Resetting online commerce


This post is by Benedict Evans from Essays - Benedict Evans

“Aunt Agatha’s demeanour now was rather like that of one who, picking daisies on the railway, has just caught the down express in the small of the back.” – P.G. Wodehouse

I’ve spent a lot of time in the last few years looking at ecommerce and discovery – how do people decide what to buy online, when a shop can’t show it to them? It seems to me that pretty much every part of that question is being reset this year. There are half a dozen huge industries where all of the cards are being thrown up in the air, and no-one really knows where they’re going to land. The Covid lockdowns of 2020 and 2021 are catalysing and accelerating all sorts of changes – we’re getting five years of adoption in a few quarters, and five years of inevitability in the back of the neck. 

Physical retail itself has been a ‘boiling frog’ for 20 years. Every year ecommerce gets a little bigger and the problem gets a little worse, but the growth in any given year was never big enough for people to panic, and you could always tell yourself that sure, people would buy that other industry’s product online, but not yours. I think we all now understand that anyone will buy anything online, given the right experience, and if your retail model is based on being an end-point to a logistics chain then you have an existential problem. 

Misc slides.001.png

This is accelerated by lockdowns, partly because growth in ecommerce penetration that we all expected has been pulled forward, and partly because everyone is now forced to try buying things online that they might not have considered before (most obviously groceries, where UK online penetration has doubled from 5% of sales to 10% this year).

But the reduction in footfall itself also has cascading consequences. It’s pretty obvious that many US malls are anchored by large retailers that could very easily now go out of business, and then the other retailers in that mall that might have though they were OK now aren’t – and then the mall itself goes as well. That purchasing won’t automatically go to those retailers’ websites, and it might also go to entirely different categories. If you change the channel then you change what gets bought. 

A further and perhaps more interesting question is that a shift to working remotely might be a permanent change for many retail areas in big cities. Even if people now work from home only one day a week, how many retailers will experience that as a 20% decline in footfall, and how many cannot survive that? In some areas that might also be a vicious circle: more people working from home means less retail, and less retail in Canary Wharf or Hudson Yards might mean more people working from home. I’ve seen people call this a ‘donut’ effect – office districts of a city are hollowed out. 

Then, what gets sold in those shops? In the last couple of years there’s been an explosion and arguable a bubble in so-called direct to consumer or ‘D2C’ brands. The bubble burst at the beginning of this year (ironically just before everyone had to buy everything online), partly prompted by the realisation that if you’re not renting a store on Fifth Avenue, that money doesn’t go to the bottom line – you’ll almost certainly have to spend it on delivery, advertising, Amazon placement or returns instead (in other words, there are no free lunches). But the reasons why that explosion had happened remain: you can now make and sell a consumer product without the same kind of fixed cost and upfront capital investment in a national retail footprint, inventory and marketing that would have been necessary 20 years ago. But what does that mean? What is a sustainable customer acquisition model? For how many brands, and what aggregation and discovery models? Is there any role for ‘software’ or is this really entirely a CPG and marketing story? And at what point do you need to get bought by P&G, or LVMH, or partner with Sephora, in much the way that you would have in 2000? (One part of the question – do these kinds of companies produce venture returns?)

I’m a terrified dinosaur” – Jorge Paulo Lemann, co-founder of 3G Capital, recent purchaser of Kraft Heinz

Those traditional brand owners are also scrambling. Many of the big consumer brands we all know have historically been B2B businesses. P&G doesn’t sell soap – it sells pallets of soap. Now all of these companies are trying to work out what a customer relationship would be, and how many companies can have that (hence, for example, Lululemon buying Mirror for $500m earlier this year). What does it mean to be a brand, or a brand owner, or for manufacturing, distribution and capital for those brands, when all of that is being unbundled and rebundled? 

Misc+slides.035.jpeg

Meanwhile, if you’re now spending your acquisition budget on advertising instead of rent, what does that advertising look like? Again, no-one quite knows. Print advertising has collapsed. TV has been pretty resilient as the internet has grown (though the chart above suggests that its share of GDP has fallen significantly), but subscriptions and audiences are now decisively switching away from the old model, so what will that look like in 5 years?

Misc slides.001.png

And then there’s internet advertising, and that looks more uncertain than almost anything else I’ve written about here. Over the past 25 years, a huge inverted pyramid has been built up on top of cookies, much of which can often look more like rent-seeking, arbitrage and general spivery than rational economic optimisation. Earlier this year PwC carried out a study of UK online advertising suggesting that not only does half of ad spending not actually make it to publishers, but that 15% couldn’t be traced at all. 

Source: PwC for ISBA

Source: PwC for ISBA

Today, privacy changes by Google and Apple on one side (in Chrome, Safari and iOS), and GDPR, CCPA and a whole bunch of other blunderbuss regulation on the other, are shoving over that whole tottering mess of tracking, targeting, interest and identity management. I’ve sometimes had the impression that almost no-one in Silicon Valley that doesn’t actually work on an ad team really pays any attention to ‘ad tech’, but now that whole business is being reset.

Quite separately, Google and Facebook’s own ad market position (they have at least half of all online advertising) is attracting very serious scrutiny from competition regulators, especially in the UK and EU, with all sorts of suggestions of highly technical and specific intervention into the mechanics of their market dominance – many of which incidentally are in direct conflict with what the privacy regulator next door is demanding. The competition regulator says ‘make it easy to move data around’ and the privacy regulator says ‘don’t’. 

Online advertising is now worth perhaps $250bn, but advertising in total is $500bn and all global marketing is closer to $1tr. Telling people about things they might like or be interested has value, and it isn’t actually evil a priori, but if you can’t ‘track’ people across the web anymore, how do you do that? And how do you reconcile that with wanting more competition to Google or Amazon? I hope that the answer is not that the only companies that can do interest-based ads are Google and Facebook on one hand and brands with their own huge audiences and data such as the Guardian or New York Times on the other. Will one or other of the various industry data initiatives work? Will Apple try a generalised identity or interest platform? I don’t know, but I do know that a trillion dollar industry is up in the air. 

I don’t know the answer to most of these question – more importantly, I don’t really know the questions. What will happen in TV? I don’t know – ask a TV analyst! What will happen to all these D2C CPG companies? I don’t know – ask a CPG analyst! But of course, they don’t know either.

The end of the American internet — Benedict Evans


This post is curated by Keith Teare. It was written by Benedict Evans. The original is [linked here]

There are many arguments that flow out of this. One, for example, is how far and how many Chinese consumer internet companies will spread globally as opposed to being constrained by their domestic environment (this would be the ‘Galapagos Effect’ often suggested of Japanese tech. Tiktok worked, but WeChat failed). Another is how many ‘unicorns’ come from Europe – how fast does its population, economic, scientific and educational base produce a proportionate number of big tech companies (or if not, why not?). Yet another is the ‘Is Silicon Valley Over?’ debate, which goes back decades – when my old colleague Marc Andreessen arrived there in the early 1990s, he thought the whole thing was over and he’d missed it.

You can argue about the details of all of these all day, but it does seem inarguable that we should just presume a global diffusion of software creation and internet company creation. It doesn’t really matter if Silicon Valley ends up as 25% or 75% of the next 100 important companies – America doesn’t have a monopoly on the agenda any more.

Hence, there are all sorts of issues with the ways that the US government has addressed Tiktok in 2020, but the most fundamental, I think, is that it has acted as though this is a one-off, rather than understanding that this is the new normal – there will be hundreds more of these. You can’t one-at-a-time this – you need a systematic, repeatable approach. You can’t ask to know the citizenship of the shareholders in every popular app – you need rules that apply to everyone. Today, the rules come from Apple, or California, both of which are increasingly becoming America’s privacy regulators by default. But they will also come from the EU, which is increasingly writing laws that, intentionally or not, change how American companies do business in America, and the more different rules we have in different places, the more fragmented and complicated things get. Regulation is an export industry, and a competitive industry.

The end of the American internet


This post is by Benedict Evans from Essays - Benedict Evans

When Netscape launched in 1994 and kicked off the consumer internet, there were maybe 100m PCs on earth, and over half of them were in the USA. The web was invented in Switzerland, and computers were invented in the UK, but the internet was American. American companies set the agenda and created most of the important products and services, and American attitudes, cultures and laws around regulation and speech dominated.

This is not quite so true anymore. 80-90% of internet users are now outside the USA, there are more smartphone users in China than in the USA and western Europe combined, and the creation of venture-based startups has gone global.

2020 09 New Normal.003.png

2020 09 New Normal.002.png

Meanwhile, of course, the internet has become vastly more important. in the last decade it has gone from being interesting and exciting but not really an important part of most people’s lives to being a central part of society. This is my favorite way to illustrate this – by 2017, almost half of new (straight) relationships in the USA started online.

2020 09 New Normal.001.png

This has two pretty basic sets of consequences.

First, as I discussed in some detail here, technology is becoming a regulated industry, if only because important and specialised industries are always regulated, and that regulation will not only be determined by the USA. Other countries have their own laws, cultures and constitutions, and so we are entering a period of increasing regulatory expansion, overlap and competition from different jurisdictions, from the EU and UK to Singapore or Australia and, of course, China.

Second, you can no longer assume that the important companies and products themselves are American.

Both of these are captured in Tiktok. This is the first time that Americans have really had to deal with their teenagers using a form of mass media that isn’t created in their country by people who mostly share their values. It’s from somewhere else. That’s compounded by the fact that the ‘somewhere else’ is China, with all of the political and geopolitical issues that come with that, but I’d suggest that the core, structural issue is that it’s foreign. This is, of course, a problem that the rest of the world has been wrestling with since 1994, but it comes as something of a shock in Washington DC. There’s an old joke that war is how God teaches Americans geography – now it’s regulation.

There are many arguments that flow out of this. One, for example, is how far and how many Chinese consumer internet companies will spread globally as opposed to being constrained by their domestic environment (this would be the ‘Galapagos Effect’ often suggested of Japanese tech. Tiktok worked, but WeChat failed). Another is how many ‘unicorns’ come from Europe – how fast does its population, economic, scientific and educational base produce a proportionate number of big tech companies (or if not, why not?). Yet another is the ‘Is Silicon Valley Over?’ debate, which goes back decades – when my old colleague Marc Andreessen arrived there in the early 1990s, he thought the whole thing was over and he’d missed it.

You can argue about the details of all of these all day, but it does seem inarguable that we should just presume a global diffusion of software creation and internet company creation. It doesn’t really matter if Silicon Valley ends up as 25% or 75% of the next 100 important companies – America doesn’t have a monopoly on the agenda any more.

Hence, there are all sorts of issues with the ways that the US government has addressed Tiktok in 2020, but the most fundamental, I think, is that it has acted as though this is a one-off, rather than understanding that this is the new normal – there will be hundreds more of these. You can’t one-at-a-time this – you need a systematic, repeatable approach. You can’t ask to know the citizenship of the shareholders in every popular app – you need rules that apply to everyone. Today, the rules come from Apple, or California, both of which are increasingly becoming America’s privacy regulators by default. But they will also come from the EU, which is increasingly writing laws that, intentionally or not, change how American companies do business in America, and the more different rules we have in different places, the more fragmented and complicated things get. Regulation is an export industry, and a competitive industry.

Amazon’s profits, AWS and advertising


This post is by Benedict Evans from Essays - Benedict Evans

People argue about Amazon a lot, and one of the most common and long-running arguments is about profits. The sales keep going up, and it takes a larger and larger share of US retail every year (7-8% in 2019), but it never seems to make any money. What’s going on?

Misc slides.001.png

Well, to begin with, this idea itself is a little out of date: if we zoom in on that net income line, we can see that Amazon’s profitability appears to have shot up in the last couple of years. But what else is going on?

Misc slides.002.png

An obvious response here is that all of the profit is coming from AWS: it’s easy to assume that AWS’s profits subside losses in the rest of the company. By extension, if anti-trust intervention split AWS apart from the rest of the company, those cross-subsidies would go away and Amazon would have to put up prices, or grow more slowly, or at any rate be a less formidable and aggressive competitor.

That doesn’t really stand up to examination.

AWS has been around in some form since the early 2000s, but Amazon didn’t disclose financials, and the products looked so cheap that many people presumed that it must be making a loss. Then, in 2015, reporting regulations meant that Amazon had to start giving numbers (with historic figures back to 2013), and we discovered that in fact it was hugely profitable. Hence, in 2015 and 2016 AWS was the great majority of Amazon’s operating income. However, that’s not true anymore – the US business is also now generating substantial operating profit, and, on this basis, it’s only the international business that’s losing money.

Misc slides.003.png

The phrase ‘on this basis’ is important. Amazon reports revenue and operating income for three segments: AWS, USA and Rest of World. The chart below shows the revenue (and also lets you see that AWS is a much higher-margin business).

Misc slides.004.png

However, this is not the only kind of disclosure that Amazon gives. If you scroll a little further through the 10-K, you will find that since 2014 the company has also disclosed revenue (though not profitability) on a quite different and much more informative basis.

Misc slides.005.png

‘First party ecommerce’, where Amazon sells you things on its own behalf on the Amazon website, is now only about half of Amazon’s revenue. Another third comes from providing platforms for other people to do business: AWS is one part of this and Marketplace, or ‘third party services’ is the other.

Misc slides.006.png

Amazon lets other companies list products on its website and ship them through its warehouses as the ‘Marketplace’ business. It charges them a fee for this, and it reports the fee as revenue (‘third party services), and it makes a profit on that. Amazon doesn’t treat the value of the actual purchases as its own revenue, which is in line with US accounting rules, since technically Amazon is only acting as an agent. So, if you buy a $1,000 TV on Amazon from a third party supplier, Amazon will charge the supplier (say) $150 in fees for shipping and handling and commission, and only report $150 as revenue. However, it has started stating, in rounded numbers, what percentage third party sales make up of total sales on Amazon – so-called ‘gross marketplace value’ or GMV. Last year, it was about 60%.

Misc slides.007.png

As an intellectual exercise, it’s interesting to think about what this would look like if the accounting rules were different and everything sold and processed through Amazon’s website was reported as Amazon revenue. On an operational level, this is pretty much what happens today: Amazon’s own ecommerce product teams are charged an internal fee by the logistics platform and by the digital platform in much the same way that external marketplace vendors are charged a fee. Hence, if these were reported on a like-for-like basis, Amazon’s revenue in 2019 would have been close to $450bn.

Misc slides.008.png

Marketplace gets quite a lot of attention these days, but it’s also worth a quick look at one of the small and insignificant-looking series on that chart – ‘ads and other’. The vast majority of this is Amazon’s business selling placement on the home page and in product search results, which it has built up from almost nothing in the last five years. This is what that business looks like in isolation – it did close to $15bn of revenue in 2019. A billion here and a billion there can add up to real money.

Misc slides.009.png

Amazon doesn’t disclose profitability for this segment, but we can make some informed (wild) guesses. So: it mostly leverages existing technical infrastructure and engineering resource. It must have meaningful numbers of sales and operations people, but the system itself is mostly automated. It will have knock-on consequences to other past of the business – for example, it may steer sales to product with higher or lower profitability. And it seems reasonable to assume that it has pretty high margins.

So, for comparison, Alphabet had a 2019 operating margin before R&D and TAC (it’s at least arguable that neither apply here) of 57%, and AWS reported 2019 operating margin of 26%. On that basis it’s reasonable to suggest that the ad business is contributing as much operating income as everything else apart from AWS, and it’s not absurd to suggest it might be close to matching AWS.

Misc slides.001.png

Close to six years ago I wrote a pretty popular essay about Amazon’s business at the time – ‘Why Amazon has no profits’. That made two points.

First, Amazon is not one business – it’s many different businesses, at different stages of maturity and profitability. Some of those businesses are established highly profitable and others are new and in a start-up loss making phase, but you can’t really see from the outside, because all of the money gets reinvested. You can see exactly the same thing in these charts. Amazon is not a loss-making business that will eventually have to raise prices to make money; rather, it is many businesses leveraging a common platform and a common balance sheet.

Second, Amazon is run for cash, not net income. Jess Bezos always says that he runs it for ‘trailing 12 months’ free cashflow’, not net income, and it’s had positive cashflow since 2002. It’s a drawback of these charts that they’re based on operating income, not cashflow, but that’s what we’ve got. There are also bunch of other interesting things one could dig into – stock compensation, say, or the cash conversion cycle. But the important thing is that if you want to understand a company, it’s worth reading the accounts.

Amazon’s profits, AWS and advertising


This post is by Benedict Evans from Essays - Benedict Evans

People argue about Amazon a lot, and one of the most common and long-running arguments is about profits. The sales keep going up, and it takes a larger and larger share of US retail every year (7-8% in 2019), but it never seems to make any money. What’s going on?

Misc slides.001.png

Well, to begin with, this idea itself is a little out of date: if we zoom in on that net income line, we can see that Amazon’s profitability appears to have shot up in the last couple of years. But what else is going on?

Misc slides.002.png

An obvious response here is that all of the profit is coming from AWS: it’s easy to assume that AWS’s profits subsidise losses in the rest of the company. By extension, if anti-trust intervention split AWS apart from the rest of the company, those cross-subsidies would go away and Amazon would have to put up prices, or grow more slowly, or at any rate be a less formidable and aggressive competitor.

That doesn’t really stand up to examination.

AWS has been around in some form since the early 2000s, but Amazon didn’t disclose financials, and the products looked so cheap that many people presumed that it must be making a loss. Then, in 2015, reporting regulations meant that Amazon had to start giving numbers (with historic figures back to 2013), and we discovered that in fact it was hugely profitable. Hence, in 2015 and 2016 AWS was the great majority of Amazon’s reported operating income. However, that’s not true anymore – the US business is also now generating substantial operating profit, and, on this basis, it’s only the international business that’s losing money.

Misc slides.003.png

There is an easy way to get this wrong if you’re not careful. If you only look at AWS and total operating profit, or aggregate the numbers that Amazon reports into AWS and ‘Other’, the resulting numbers will be misleading: the losses for RoW balance out the profit in the USA and make it look as though AWS is the only profitable business, especially from 2015 to 2017. The chart below shows the result: close to $3bn of operating profit in the US business in 2017 has vanished, and so has a $3bn operating loss in the RoW business. Don’t do this.

Misc slides.001.png

This is a great illustration of a broader challenge: Amazon is lots of businesses, but you only see the profitability of the aggregate.

Hence, Amazon reports revenue and operating income for three segments: AWS, USA and Rest of World. The chart below shows the revenue (and also lets you see that AWS is a much higher-margin business).

Misc slides.004.png

However, this is not the only kind of disclosure that Amazon gives. If you scroll a little further through the 10-K, you will find that since 2014 the company has also disclosed revenue (though not profitability) on a quite different and much more informative basis.

Misc slides.005.png

‘First party ecommerce’, where Amazon sells you things on its own behalf on the Amazon website, is now only about half of Amazon’s revenue. Another third comes from providing platforms for other people to do business: AWS is one part of this and Marketplace, or ‘third party services’ is the other.

Misc slides.006.png

Amazon lets other companies list products on its website and ship them through its warehouses as the ‘Marketplace’ business. It charges them a fee for this, and it reports the fee as revenue (‘third party services’), and it makes a profit on that. Amazon doesn’t treat the value of the actual purchases as its own revenue, which is in line with US accounting rules, since technically Amazon is only acting as an agent. So, if you buy a $1,000 TV on Amazon from a third party supplier, Amazon will charge the supplier (say) $150 in fees for shipping and handling and commission, and only report $150 as revenue. However, it has started stating, in rounded numbers, what percentage third party sales make up of total sales on Amazon – so-called ‘gross marketplace value’ or GMV. Last year, it was about 60%.

Misc slides.007.png

As an intellectual exercise, it’s interesting to think about what this would look like if the accounting rules were different and everything sold and processed through Amazon’s website was reported as Amazon revenue. On an operational level, this is pretty much what happens today: Amazon’s own ecommerce product teams are charged an internal fee by the logistics platform and by the digital platform in much the same way that external marketplace vendors are charged a fee. Hence, if these were reported on a like-for-like basis, Amazon’s revenue in 2019 would have been close to $450bn.

Misc slides.008.png

Marketplace gets quite a lot of attention these days, but it’s also worth a quick look at one of the small and insignificant-looking series on that chart – ‘ads and other’. The vast majority of this is Amazon’s business selling placement on the home page and in product search results, which it has built up from almost nothing in the last five years. This is what that business looks like in isolation – it did close to $15bn of revenue in 2019. A billion here and a billion there can add up to real money.

Misc slides.009.png

Amazon doesn’t disclose profitability for this segment, but we can make some informed (wild) guesses. So: it mostly leverages existing technical infrastructure and engineering resource. It must have meaningful numbers of sales and operations people, but the system itself is mostly automated. It will have knock-on consequences to other parts of the business – for example, it may steer sales to product with higher or lower profitability. And it seems reasonable to assume that it has pretty high margins.

So, for comparison, Alphabet had a 2019 operating margin before R&D and TAC (it’s at least arguable that neither apply here) of 57%, and AWS reported 2019 operating margin of 26%. On that basis it’s reasonable to suggest that the ad business is contributing as much operating income as everything else apart from AWS, and it’s not absurd to suggest it might be close to matching AWS.

Misc slides.001.png

Close to six years ago I wrote a pretty popular essay about Amazon’s business at the time – ‘Why Amazon has no profits’. That made two points.

First, Amazon is not one business – it’s many different businesses, at different stages of maturity and profitability. Some of those businesses are established and highly profitable and others are new and in a start-up loss making phase, but you can’t really see from the outside, because all of the money gets both aggregated and reinvested. You can see exactly the same thing in these charts. Amazon is not a loss-making business that will eventually have to raise prices to make money; rather, it’s many businesses leveraging a common platform and a common balance sheet.

Second, Amazon is run for cash, not net income. Jeff Bezos always says that he runs it for ‘trailing 12 months’ absolute free cashflow’, not net income, and it’s had positive cashflow since 2002, which is before some startup founders were born. It’s a drawback of these charts that they’re based on operating income, not cashflow, but that’s what we’ve got. There’s also a bunch of other interesting things one could dig into – stock compensation, say, or the cash conversion cycle. But the important thing is that if you want to understand a company, it’s worth reading the accounts.

The ecommerce surge


This post is by Benedict Evans from Essays - Benedict Evans

Both the UK and (today) the USA have given official statistics on how ecommerce and retail have changed during lockdown. The headline numbers are pretty dramatic. The UK went from 20% ecommerce penetration to over 30% in two months, and the USA from 17% to 22%.

Misc slides.001.png

This spike partly reflects a shift in the denominator: digital increased while sales at most physical retail declined, except for groceries. This effect was much stronger in the UK, where lockdown was much deeper. Even so, absolute US ecommerce sales rose 32% in Q2. Interestingly, though, absolute UK ecommerce has continued to accelerate even as the lockdown has eased (the UK’s monthly ecommerce numbers make this visible, where the quarterly US reporting makes it harder to tell.

Misc slides.004.png

Misc slides.005.png

Meanwhile, UK online grocery sales doubled almost overnight, on a low base. Conversely, the UK lockdown has such a dramatic effect on everything else except groceries that the ecommerce share of non-grocery peaked at 60%, though it’s now falling back down.

Misc slides.003.png

Misc slides.006.png

Finally, as we all know, these effects are very unevenly distributed. In a ‘normal’ economic shock the pain is concentrated on companies with weak management, strategy, balance sheets or strategy positioning, but here it’s much more specific: bars & restaurants and travel, obviously, but also clothes.

Misc slides.007.png

As I wrote here, we’re in a period of both forced experiment and forced experimentation. In January everyone was online and willing to try anytime online: now we have no choice. So, some of this is the future happening more quickly, with years of growth being pulled forward, but some of it also is an experiment that won’t stick. We’ll find out which in the next six months or so.

The ecommerce surge


This post is by Benedict Evans from Essays - Benedict Evans

Both the UK and (today) the USA have given official statistics on how ecommerce and retail have changed during lockdown. The headline numbers are pretty dramatic. The UK went from 20% ecommerce penetration to over 30% in two months, and the USA from 17% to 22%.

Misc slides.001.png

This spike partly reflects a shift in the denominator: digital increased while sales at most physical retail declined, except for groceries. This effect was much stronger in the UK, where lockdown was much deeper. Even so, absolute US ecommerce sales rose 32% in Q2.

Misc slides.004.png

Interestingly, though, absolute UK ecommerce has remained strong even as the lockdown has eased (the UK’s monthly ecommerce numbers make this visible, where the quarterly US reporting makes it harder to tell.

Misc slides.003.png

Meanwhile, UK online grocery sales doubled almost overnight, on a low base. Conversely, the UK lockdown has such a dramatic effect on everything else except groceries that the ecommerce share of non-grocery peaked at 60%, though it’s now falling back down.

Misc slides.002.png

Misc slides.004.png

Finally, as we all know, these effects are very unevenly distributed. In a ‘normal’ economic shock the pain is concentrated on companies with weak management, strategy, balance sheets or strategy positioning, but here it’s much more specific: bars & restaurants and travel, obviously, but also clothes.

Misc slides.007.png

As I wrote here, we’re in a period of both forced experiment and forced experimentation. In January everyone was online and willing to try anytime online: now we have no choice. So, some of this is the future happening more quickly, with years of growth being pulled forward, but some of it also is an experiment that won’t stick. We’ll find out which in the next six months or so.

(Note: I updated the UK charts with July data on 21 August.)

The ecommerce surge


This post is by Benedict Evans from Essays - Benedict Evans

Both the UK and (today) the USA have given official statistics on how ecommerce and retail have changed during lockdown. The headline numbers are pretty dramatic. The UK went from 20% ecommerce penetration to over 30% in two months, and the USA from 17% to 22%.

Misc slides.001.png

This spike partly reflects a shift in the denominator: digital increased while sales at most physical retail declined, except for groceries. This effect was much stronger in the UK, where lockdown was much deeper. Even so, absolute US ecommerce sales rose 32% in Q2.

Misc slides.004.png

Interestingly, though, absolute UK ecommerce has remained strong even as the lockdown has eased (the UK’s monthly ecommerce numbers make this visible, where the quarterly US reporting makes it harder to tell.

Misc slides.003.png

Meanwhile, UK online grocery sales doubled almost overnight, on a low base. Conversely, the UK lockdown has such a dramatic effect on everything else except groceries that the ecommerce share of non-grocery peaked at 60%, though it’s now falling back down.

Misc slides.002.png

Misc slides.004.png

Finally, as we all know, these effects are very unevenly distributed. In a ‘normal’ economic shock the pain is concentrated on companies with weak management, strategy, balance sheets or strategy positioning, but here it’s much more specific: bars & restaurants and travel, obviously, but also clothes.

Misc slides.007.png

As I wrote here, we’re in a period of both forced experiment and forced experimentation. In January everyone was online and willing to try anytime online: now we have no choice. So, some of this is the future happening more quickly, with years of growth being pulled forward, but some of it also is an experiment that won’t stick. We’ll find out which in the next six months or so.

(Note: I updated the UK charts with July data on 21 August.)

The ecommerce surge


This post is by Benedict Evans from Essays - Benedict Evans

Both the UK and (today) the USA have given official statistics on how ecommerce and retail have changed during lockdown. The headline numbers are pretty dramatic. The UK went from 20% ecommerce penetration to over 30% in two months, and the USA from 17% to 22%.

Misc slides.001.png

This spike partly reflects a shift in the denominator: digital increased while sales at most physical retail declined, except for groceries. This effect was much stronger in the UK, where lockdown was much deeper. Even so, absolute US ecommerce sales rose 32% in Q2.

Misc slides.004.png

Interestingly, though, absolute UK ecommerce has remained strong even as the lockdown has eased (the UK’s monthly ecommerce numbers make this visible, where the quarterly US reporting makes it harder to tell.

Misc slides.003.png

Meanwhile, UK online grocery sales doubled almost overnight, on a low base. Conversely, the UK lockdown has such a dramatic effect on everything else except groceries that the ecommerce share of non-grocery peaked at 60%, though it’s now falling back down.

Misc slides.002.png

Misc slides.004.png

Finally, as we all know, these effects are very unevenly distributed. In a ‘normal’ economic shock the pain is concentrated on companies with weak management, strategy, balance sheets or strategy positioning, but here it’s much more specific: bars & restaurants and travel, obviously, but also clothes.

Misc slides.007.png

As I wrote here, we’re in a period of both forced experiment and forced experimentation. In January everyone was online and willing to try anytime online: now we have no choice. So, some of this is the future happening more quickly, with years of growth being pulled forward, but some of it also is an experiment that won’t stick. We’ll find out which in the next six months or so.

(Note: I updated the UK charts with July data on 21 August.)

App stores, trust and anti-trust


This post is by Benedict Evans from Essays - Benedict Evans

We all, I think, understand that the iPhone was a generational change in computing, but that change came in two parts. The multitouch interface is obvious, but the change in the software model was just as important. Apple changed how software development worked, and by doing so expanded the number of people who could comfortably, safely use a computer from a few hundred million to a few billion.

Specifically, Apple tried to solve three kinds of problem.

  • Putting apps in a sandbox, where they can only do things that Apple allows and cannot ask (or persuade, or trick) the user for permission to do ‘dangerous’ things, means that apps become completely safe. A horoscope app can’t break your computer, or silt it up, or run your battery down, or watch your web browser and steal your bank details.

  • An app store is a much better way to distribute software. Users don’t have to mess around with installers and file management to put a program onto their computer – they just press ‘Get’. If you (or your customers) were technical this didn’t seem like a problem, but for everyone else with 15 copies of the installer in their download folder, baffled at what to do next, this was a huge step forward.

  • Asking for a credit card to buy an app online created both a friction barrier and a safety barrier – ‘can I trust this company with my card?’ Apple added frictionless, safe payment.

All of this levelled the playing field. You knew you could trust Adobe or EA with your credit card, and you knew you could trust them not to abuse your PC too much. PanicRogue Amoeba or Basecamp have accumulated reputations that mean they get trust too, for tech insiders who’ve known about them for years. But what about a random Vietnamese developer who’s made a fun little game about a bird that flaps? The iOS software model removed trust as a problem, and as an advantage for big companies. You still have to hear about the app – the App Store solves distribution but not discovery – but you don’t have to worry about paying for it and you don’t have to worry what it might do to your computer.

This model has enabled an explosion of software. A billion people use iPhones today, the App Store has 500m weekly users, and those users both buy and install far more software than ever before. The new software model has, objectively, been great for software development, and also and much more importantly has been hugely and unambiguously good for actual consumers. Trust and rules are good.

The trouble is, if you have a curated, managed sandbox, where a company decides what’s safe, you have to do a good job of managing and curating, and Apple has not, always, done a good job at all.

And, it matters if Apple doesn’t do a good job, because when Apple launched the app store it had sold fewer than 10m iPhones ever, but today a billion people use iPhones, and more importantly so does over half of the US market and 80% of American teenagers. For a lot of big companies, iPhone users are the market. When your product has a few points of market share you can make whatever choices you like, but when you dominate the market, other rules start applying. Apple isn’t the pirates anymore – it’s the navy, the port and the customs house. In the last few weeks, Microsoft, Google, Facebook and Epic have been stopped at customs.

2020 Shoulders of Giants 1.5.001.png

So, what kinds of decision does Apple make about what you can do on an iPhone or iPad, and where are the problems? Splitting this apart:

  1. Most decisions do actually have a solid, rational engineering basis. You can’t run in the background and record what every other app does.You can’t run the battery down, or read all my photos without permission, or hijack my network connection and CPU to mine crypto. 

  2. But some seem to be just personal preference, or taste – most obviously, the decision in the last few weeks to block streaming games services from Microsoft and Google. This may partly be about revenue, but the real issue seems to be that Apple thinks that games on iOS ’should’ use native APIs, and, perhaps, that they ‘should’ work without you needing to buy a separate games controller. But whatever it is, there’s no safety, security or privacy issue – Apple just does’t like those apps.

  3. Some decisions in both of two previous categories cause difficulties for third party apps that compete with Apple products. That isn’t necessarily the aim, but it also might not go un-noticed at Apple.

  4. Then, there are endless horror stories around curation of the store. Apps are rejected in arbitrary, capricious, irrational and inconsistent ways, often for breaking completely unwritten rules. Only Apple actually knows how much this happens, but far too many people have far too many bad experiences. This has done real damage to Apple’s brand amongst developers.

  5. And then there are the payment rules.

Apple’s payment rules, made mandatory in 2011, created a whole load of new problems:

  • United Airlines and Uber aren’t covered at all – only content consumed on the device is affected.

  • There are apps where there’s a clear logic for Apple’s payment system to be compulsory – there are coherent, consistent reasons why that random horoscope app should use the build-in payment and not be allowed to offer extra value if you give it a card, and a level playing field means the same rules for everyone (except, we just discovered that Amazon is ‘only’ paying 15% for some Prime Video signups on iOS).

  • However, there’s a huge grey area around services that are consumed both on and off the platform – for example, Netflix, Disney and newspapers and magazines. How and where, exactly, should Apple get a cut?

  • Worse, there are companies that just can’t pay. Ebooks or music apps have to give a fixed percentage of their top line to rights-holders and don’t have a 30% margin to give Apple.

We had all these arguments in 2011 and very little has changed since: I wrote this report at the time and it’s still a pretty good summary.

View fullsize

Screenshot 2020-06-21 at 12.28.29 pm.png

View fullsize

Screenshot 2020-06-21 at 12.29.23 pm.png

View fullsize

Screenshot 2020-06-21 at 12.29.43 pm.png

View fullsize

Screenshot 2020-06-21 at 12.29.53 pm.png

View fullsize

Screenshot 2020-06-21 at 12.30.01 pm.png

View fullsize

Screenshot 2020-06-21 at 12.30.13 pm.png

Meanwhile, none of this is a surprise to Apple. As part of the recent US congress competition hearings, we saw an email from early 2011 in which Steve Jobs explicitly accepted and embraced the fact that the payment rules would be a fundamental problem for some companies.  The result, for almost a decade, has been a horrible muddle, with people using those products forced into a bad user experience.

Screenshot 2020-08-13 at 2.35.17 pm.png

(Of course, when this email was written, Apple was still a fair way away from market dominance: there were only 150-160m iOS devices in use, and iPhones were maybe 10% of all the mobile phones being used in the USA, where today they’re over 50%.)

Ironically, Epic is not in this ‘can’t pay’ category at all, and it built a huge and very profitable business following Apple’s rules. Unlike Spotify, it doesn’t have marginal cost for in-app purchases and there’s no structural reason why it can’t pay Apple (or indeed Google). It just doesn’t want to, or wants to pay less than 30%. Indeed, one could point out that the real issue is that Epic just doesn’t ’like’ Apple’s model, just as Apple doesn’t ‘like’ Stadia.

At this point, many people suggest that we can cut a Gordian knot here – that we can slash through the complexity by letting people have a choice. Allow any payment service, perhaps in parallel to Apple’s; allow third-party app stores; allow side-loading of apps; and of course let users turn off the sandboxed restrictions on the phone. Then you can have the security if you want it, or the freedom.

Unfortunately, you can’t have your cake and eat it. A secure system with a switch to turn off the security might work for Linux and a highly technical user, but when you’ve given smartphones to a few billion people, a secure system with a switch to turn off the security is just a target for malware. That horoscope app can tell you’ll get more accurate results if it has access to some computer gibberish, so please press OK, and guess what? Everyone will press OK. A computer should not ask a question that the user won’t understand, and when you have billions of users that list looks different. This has been Google’s experience with Android: it chose a less restrictive sandbox than iOS and had many more malware problems, and Google has spent the last decade slowly rowing towards Apple’s approach.

A limited version of this argument, incidentally, is that all the problems are with the store, and you could get rid of it without security problems, relying on software sandboxing on the phones to handle all security and safety issues. But there are also policies people object to on the phone itself (no replacing the default Maps app, say), and policies that we want to keep that are enforced in the store rather than on the phone (no ads in apps for kids, say). You have to tackle the whole policy question, not just part of it, and you cannot rely on the sandbox on the phone to solve all safety and security attacks.

All of this is to say that the demands Epic makes in its lawsuit are not, in fact, merely arguing that the smartphone apps market should be more competitive, with more payment options. The sandboxed app store model is not some curious, incidental feature of modern smartphones – rather, this is an essential and hugely important part of why they have such a strong software ecosystem. Epic is explicitly arguing that we should abandon the smartphone software model and security model almost entirely, and switch to what would actually be the old Windows model. Its arguments would also of course mean that we should abandon any level playing field, and move to a model where big companies and big brands have an even bigger advantage, because a trusted platform is replaced by a trusted reputation. This would be good for big established brands – like Epic – but not for may other people.

Epic’s proposal is full of holes, and Epic’s problem is really pretty peripheral, but I’m much more interested in Spotify and Stadia, where the situation now looks unsustainable, and that’s where we’re more likely to see changes. So, I think we should try to draw one more set of distinctions.

First, the App Store moderation problems are infuriating but they’re not rent-seeking or necessarily market abuse – they’re an execution failure, and indeed we’ve been here before. However, the EU, which is becoming the tech regulator by default, is already working on plans for the store review model to be regulated, with real, external rights of appeal and review, and external transparency. Apple could try to get ahead of this, or it might be too late. It might have no choice but to allow Stadia, and ‘we just don’t like that’ won’t do anymore. 

Second, I think Apple is going to have to make fundamental changes to the payment model. Epic only has margin at stake, but Spotify can’t pay at all, it’s a direct competitor, and there’s no user benefit at all to Apple’s policy, just confusion and annoyance. The EU is now pursuing two separate competition policy cases against Apple: one over the App Store, with Spotify a complainant, and the other over Apple Wallet and Apple Pay. This second one is instructive: the EU is taking the view that Apple has a monopoly of payment on the iPhone. Market definition is everything. I-am-not-a-lawyer, but I don’t see how Apple can win on Spotify (or Kindle), and I don’t think it should.

That might mean changes in who and what is covered by payment rules, but it probably also means changes to the 30%. There’s a lot of argument about principle, but there’s also a price: if the rate was, say, 10%, I’m not sure that we would be having the same conversation, and Epic would certainly get less sympathy.

That 30% adds up to real money, incidentally. When the store launched, Steve Jobs said it was aiming to break even – the 30% was to cover the running costs, and it is worth remembering how how many huge companies are getting the App Store, the manual review and the file downloads to hundreds of millions of users for nothing more than their $100 a year developer subscription. But the App Store is not running at break even anymore: in 2019 it made somewhere between $10bn and $15bn of commission – 20-30% of the ‘service revenue’ Apple likes to talk about.

Finally – we’ve been arguing about this since the store launched in 2008, but really, some of this debate is as old as personal computers. Right back to the 1970s, there’s been a religious split between people who want computers that they’re free to change however they like, and people who want computers that are easy and safe to use for as many people as possible. This is a trade-off, but there’s a certain kind of person in tech that thinks app stores and the iOS sandbox have nothing to do with the success of smartphones and the iPhone – they’re just a stupid Apple thing you could get rid of with no ill effects. 30 years ago they thought the same about GUIs, and indeed a lot of Epic’s PR comes straight out of furious Usenet posts from the 1990s about how GUIs are evil and infantilising. But the whole direction of computing since the Apple 1 has been about more abstraction, less access to the lower levels of the system and inherent in that more accessibility for more people.

Apple has always been at one, extreme, end of that debate, taking a strong opinion on how it thought a good computer ‘should’ work and letting you choose it or not. From 1976 to, say, 2015 or so, it was just one fairly niche vendor, and some people chose Apple’s opinion and some didn’t. But with the iPhone, Apple finally won the argument with users’ wallets, and that means it’s not niche anymore – Apple has become the navy, and different rules apply.

Subscribers to my premium newsletter were sent a version of this on Sunday.

App stores, trust and anti-trust


This post is by Benedict Evans from Essays - Benedict Evans

We all, I think, understand that the iPhone was a generational change in computing, but that change came in two parts. The multitouch interface is obvious, but the change in the software model was just as important. Apple changed how software development worked, and by doing so expanded the number of people who could comfortably, safely use a computer from a few hundred million to a few billion.

Specifically, Apple tried to solve three kinds of problem.

  • Putting apps in a sandbox, where they can only do things that Apple allows and cannot ask (or persuade, or trick) the user for permission to do ‘dangerous’ things, means that apps become completely safe. A horoscope app can’t break your computer, or silt it up, or run your battery down, or watch your web browser and steal your bank details.

  • An app store is a much better way to distribute software. Users don’t have to mess around with installers and file management to put a program onto their computer – they just press ‘Get’. If you (or your customers) were technical this didn’t seem like a problem, but for everyone else with 15 copies of the installer in their download folder, baffled at what to do next, this was a huge step forward.

  • Asking for a credit card to buy an app online created both a friction barrier and a safety barrier – ‘can I trust this company with my card?’ Apple added frictionless, safe payment.

All of this levelled the playing field. You knew you could trust Adobe or EA with your credit card, and you knew you could trust them not to abuse your PC too much. PanicRogue Amoeba or Basecamp have accumulated reputations that mean they get trust too, for tech insiders who’ve known about them for years. But what about a random Vietnamese developer who’s made a fun little game about a bird that flaps? The iOS software model removed trust as a problem, and as an advantage for big companies. You still have to hear about the app – the App Store solves distribution but not discovery – but you don’t have to worry about paying for it and you don’t have to worry what it might do to your computer.

This model has enabled an explosion of software. A billion people use iPhones today, the App Store has 500m weekly users, and those users both buy and install far more software than ever before. The new software model has, objectively, been great for software development, and also and much more importantly has been hugely and unambiguously good for actual consumers. Trust and rules are good.

The trouble is, if you have a curated, managed sandbox, where a company decides what’s safe, you have to do a good job of managing and curating, and Apple has not, always, done a good job at all.

And, it matters if Apple doesn’t do a good job, because when Apple launched the app store it had sold fewer than 10m iPhones ever, but today a billion people use iPhones, and more importantly so does over half of the US market and 80% of American teenagers. For a lot of big companies, iPhone users are the market. When your product has a few points of market share you can make whatever choices you like, but when you dominate the market, other rules start applying. Apple isn’t the pirates anymore – it’s the navy, the port and the customs house. In the last few weeks, Microsoft, Google, Facebook and Epic have been stopped at customs.

2020 Shoulders of Giants 1.5.001.png

So, what kinds of decision does Apple make about what you can do on an iPhone or iPad, and where are the problems? Splitting this apart:

  1. Most decisions do actually have a solid, rational engineering basis. You can’t run in the background and record what every other app does.You can’t run the battery down, or read all my photos without permission, or hijack my network connection and CPU to mine crypto. 

  2. But some seem to be just personal preference, or taste – most obviously, the decision in the last few weeks to block streaming games services from Microsoft and Google. This may partly be about revenue, but the real issue seems to be that Apple thinks that games on iOS ’should’ use native APIs, and, perhaps, that they ‘should’ work without you needing to buy a separate games controller. But whatever it is, there’s no safety, security or privacy issue – Apple just does’t like those apps.

  3. Some decisions in both of two previous categories cause difficulties for third party apps that compete with Apple products. That isn’t necessarily the aim, but it also might not go un-noticed at Apple.

  4. Then, there are endless horror stories around curation of the store. Apps are rejected in arbitrary, capricious, irrational and inconsistent ways, often for breaking completely unwritten rules. Only Apple actually knows how much this happens, but far too many people have far too many bad experiences. This has done real damage to Apple’s brand amongst developers.

  5. And then there are the payment rules.

Apple’s payment rules, made mandatory in 2011, created a whole load of new problems:

  • United Airlines and Uber aren’t covered at all – only content consumed on the device is affected.

  • There are apps where there’s a clear logic for Apple’s payment system to be compulsory – there are coherent, consistent reasons why that random horoscope app should use the build-in payment and not be allowed to offer extra value if you give it a card, and a level playing field means the same rules for everyone (except, we just discovered that Amazon is ‘only’ paying 15% for some Prime Video signups on iOS).

  • However, there’s a huge grey area around services that are consumed both on and off the platform – for example, Netflix, Disney and newspapers and magazines. How and where, exactly, should Apple get a cut?

  • Worse, there are companies that just can’t pay. Ebooks or music apps have to give a fixed percentage of their top line to rights-holders and don’t have a 30% margin to give Apple.

We had all these arguments in 2011 and very little has changed since: I wrote this report at the time and it’s still a pretty good summary.

View fullsize

Screenshot 2020-06-21 at 12.28.29 pm.png

View fullsize

Screenshot 2020-06-21 at 12.29.23 pm.png

View fullsize

Screenshot 2020-06-21 at 12.29.43 pm.png

View fullsize

Screenshot 2020-06-21 at 12.29.53 pm.png

View fullsize

Screenshot 2020-06-21 at 12.30.01 pm.png

View fullsize

Screenshot 2020-06-21 at 12.30.13 pm.png

Meanwhile, none of this is a surprise to Apple. As part of the recent US congress competition hearings, we saw an email from early 2011 in which Steve Jobs explicitly accepted and embraced the fact that the payment rules would be a fundamental problem for some companies.  The result, for almost a decade, has been a horrible muddle, with people using those products forced into a bad user experience.

Screenshot 2020-08-13 at 2.35.17 pm.png

(Of course, when this email was written, Apple was still a fair way away from market dominance: there were only 150-160m iOS devices in use, and iPhones were maybe 10% of all the mobile phones being used in the USA, where today they’re over 50%.)

Ironically, Epic is not in this ‘can’t pay’ category at all, and it built a huge and very profitable business following Apple’s rules. Unlike Spotify, it doesn’t have marginal cost for in-app purchases and there’s no structural reason why it can’t pay Apple (or indeed Google). It just doesn’t want to, or wants to pay less than 30%. Indeed, one could point out that the real issue is that Epic just doesn’t ’like’ Apple’s model, just as Apple doesn’t ‘like’ Stadia.

At this point, many people suggest that we can cut a Gordian knot here – that we can slash through the complexity by letting people have a choice. Allow any payment service, perhaps in parallel to Apple’s; allow third-party app stores; allow side-loading of apps; and of course let users turn off the sandboxed restrictions on the phone. Then you can have the security if you want it, or the freedom.

Unfortunately, you can’t have your cake and eat it. A secure system with a switch to turn off the security might work for Linux and a highly technical user, but when you’ve given smartphones to a few billion people, a secure system with a switch to turn off the security is just a target for malware. That horoscope app can tell you’ll get more accurate results if it has access to some computer gibberish, so please press OK, and guess what? Everyone will press OK. A computer should not ask a question that the user won’t understand, and when you have billions of users that list looks different. This has been Google’s experience with Android: it chose a less restrictive sandbox than iOS and had many more malware problems, and Google has spent the last decade slowly rowing towards Apple’s approach.

A limited version of this argument, incidentally, is that all the problems are with the store, and you could get rid of it without security problems, relying on software sandboxing on the phones to handle all security and safety issues. But there are also policies people object to on the phone itself (no replacing the default Maps app, say), and policies that we want to keep that are enforced in the store rather than on the phone (no ads in apps for kids, say). You have to tackle the whole policy question, not just part of it, and you cannot rely on the sandbox on the phone to solve all safety and security attacks.

All of this is to say that the demands Epic makes in its lawsuit are not, in fact, merely arguing that the smartphone apps market should be more competitive, with more payment options. The sandboxed app store model is not some curious, incidental feature of modern smartphones – rather, this is an essential and hugely important part of why they have such a strong software ecosystem. Epic is explicitly arguing that we should abandon the smartphone software model and security model almost entirely, and switch to what would actually be the old Windows model. Its arguments would also of course mean that we should abandon any level playing field, and move to a model where big companies and big brands have an even bigger advantage, because a trusted platform is replaced by a trusted reputation. This would be good for big established brands – like Epic – but not for may other people.

Epic’s proposal is full of holes, and Epic’s problem is really pretty peripheral, but I’m much more interested in Spotify and Stadia, where the situation now looks unsustainable, and that’s where we’re more likely to see changes. So, I think we should try to draw one more set of distinctions.

First, the App Store moderation problems are infuriating but they’re not rent-seeking or necessarily market abuse – they’re an execution failure, and indeed we’ve been here before. However, the EU, which is becoming the tech regulator by default, is already working on plans for the store review model to be regulated, with real, external rights of appeal and review, and external transparency. Apple could try to get ahead of this, or it might be too late. It might have no choice but to allow Stadia, and ‘we just don’t like that’ won’t do anymore. 

Second, I think Apple is going to have to make fundamental changes to the payment model. Epic only has margin at stake, but Spotify can’t pay at all, it’s a direct competitor, and there’s no user benefit at all to Apple’s policy, just confusion and annoyance. The EU is now pursuing two separate competition policy cases against Apple: one over the App Store, with Spotify a complainant, and the other over Apple Wallet and Apple Pay. This second one is instructive: the EU is taking the view that Apple has a monopoly of payment on the iPhone. Market definition is everything. I-am-not-a-lawyer, but I don’t see how Apple can win on Spotify (or Kindle), and I don’t think it should.

That might mean changes in who and what is covered by payment rules, but it probably also means changes to the 30%. There’s a lot of argument about principle, but there’s also a price: if the rate was, say, 10%, I’m not sure that we would be having the same conversation, and Epic would certainly get less sympathy.

That 30% adds up to real money, incidentally. When the store launched, Steve Jobs said it was aiming to break even – the 30% was to cover the running costs, and it is worth remembering how many huge companies are getting the App Store, the manual review and the file downloads to hundreds of millions of users for nothing more than their $100 a year developer subscription. But the App Store is not running at break even anymore: in 2019 it made somewhere between $10bn and $15bn of commission – 20-30% of the ‘service revenue’ Apple likes to talk about.

Finally – we’ve been arguing about this since the store launched in 2008, but really, some of this debate is as old as personal computers. Right back to the 1970s, there’s been a religious split between people who want computers that they’re free to change however they like, and people who want computers that are easy and safe to use for as many people as possible. This is a trade-off, but there’s a certain kind of person in tech that thinks app stores and the iOS sandbox have nothing to do with the success of smartphones and the iPhone – they’re just a stupid Apple thing you could get rid of with no ill effects. 30 years ago they thought the same about GUIs, and indeed a lot of Epic’s PR comes straight out of furious Usenet posts from the 1990s about how GUIs are evil and infantilising. But the whole direction of computing since the Apple 1 has been about more abstraction, less access to the lower levels of the system and inherent in that more accessibility for more people.

Apple has always been at one, extreme, end of that debate, taking a strong opinion on how it thought a good computer ‘should’ work and letting you choose it or not. From 1976 to, say, 2015 or so, it was just one fairly niche vendor, and some people chose Apple’s opinion and some didn’t. But with the iPhone, Apple finally won the argument with users’ wallets, and that means it’s not niche anymore – Apple has become the navy, and different rules apply.

Subscribers to my premium newsletter were sent a version of this on Sunday.

App stores, trust and anti-trust


This post is curated by Keith Teare. It was written by Benedict Evans. The original is [linked here]

We all, I think, understand that the iPhone was a generational change in computing, but that change came in two parts. The multitouch interface is obvious, but the change in the software model was just as important. Apple changed how software development worked, and by doing so expanded the number of people who could comfortably, safely use a computer from a few hundred million to a few billion.

Specifically, Apple tried to solve three kinds of problem.

  • Putting apps in a sandbox, where they can only do things that Apple allows and cannot ask (or persuade, or trick) the user for permission to do ‘dangerous’ things, means that apps become completely safe. A horoscope app can’t break your computer, or silt it up, or run your battery down, or watch your web browser and steal your bank details.

  • An app store is a much better way to distribute software. Users don’t have to mess around with installers and file management to put a program onto their computer – they just press ‘Get’. If you (or your customers) were technical this didn’t seem like a problem, but for everyone else with 15 copies of the installer in their download folder, baffled at what to do next, this was a huge step forward.

  • Asking for a credit card to buy an app online created both a friction barrier and a safety barrier – ‘can I trust this company with my card?’ Apple added frictionless, safe payment.

All of this levelled the playing field. You knew you could trust Adobe or EA with your credit card, and you knew you could trust them not to abuse your PC too much. PanicRogue Amoeba or Basecamp have accumulated reputations that mean they get trust too, for tech insiders who’ve known about them for years. But what about a random Vietnamese developer who’s made a fun little game about a bird that flaps? The iOS software model removed trust as a problem, and as an advantage for big companies. You still have to hear about the app – the App Store solves distribution but not discovery – but you don’t have to worry about paying for it and you don’t have to worry what it might do to your computer.

This model has enabled an explosion of software. A billion people use iPhones today, the App Store has 500m weekly users, and those users both buy and install far more software than ever before. The new software model has, objectively, been great for software development, and also and much more importantly has been hugely and unambiguously good for actual consumers. Trust and rules are good.

The trouble is, if you have a curated, managed sandbox, where a company decides what’s safe, you have to do a good job of managing and curating, and Apple has not, always, done a good job at all.

And, it matters if Apple doesn’t do a good job, because when Apple launched the app store it had sold fewer than 10m iPhones ever, but today a billion people use iPhones, and more importantly so does over half of the US market and 80% of American teenagers. For a lot of big companies, iPhone users are the market. When your product has a few points of market share you can make whatever choices you like, but when you dominate the market, other rules start applying. Apple isn’t the pirates anymore – it’s the navy, the port and the customs house. In the last few weeks, Microsoft, Google, Facebook and Epic have been stopped at customs.

2020 Shoulders of Giants 1.5.001.png

So, what kinds of decision does Apple make about what you can do on an iPhone or iPad, and where are the problems? Splitting this apart:

  1. Most decisions do actually have a solid, rational engineering basis. You can’t run in the background and record what every other app does.You can’t run the battery down, or read all my photos without permission, or hijack my network connection and CPU to mine crypto. 

  2. But some seem to be just personal preference, or taste – most obviously, the decision in the last few weeks to block streaming games services from Microsoft and Google. This may partly be about revenue, but the real issue seems to be that Apple thinks that games on iOS ’should’ use native APIs, and, perhaps, that they ‘should’ work without you needing to buy a separate games controller. But whatever it is, there’s no safety, security or privacy issue – Apple just does’t like those apps.

  3. Some decisions in both of two previous categories cause difficulties for third party apps that compete with Apple products. That isn’t necessarily the aim, but it also might not go un-noticed at Apple.

  4. Then, there are endless horror stories around curation of the store. Apps are rejected in arbitrary, capricious, irrational and inconsistent ways, often for breaking completely unwritten rules. Only Apple actually knows how much this happens, but far too many people have far too many bad experiences. This has done real damage to Apple’s brand amongst developers.

  5. And then there are the payment rules.

Apple’s payment rules, made mandatory in 2011, created a whole load of new problems:

  • United Airlines and Uber aren’t covered at all – only content consumed on the device is affected.

  • There are apps where there’s a clear logic for Apple’s payment system to be compulsory – there are coherent, consistent reasons why that random horoscope app should use the build-in payment and not be allowed to offer extra value if you give it a card, and a level playing field means the same rules for everyone (except, we just discovered that Amazon is ‘only’ paying 15% for some Prime Video signups on iOS).

  • However, there’s a huge grey area around services that are consumed both on and off the platform – for example, Netflix, Disney and newspapers and magazines. How and where, exactly, should Apple get a cut?

  • Worse, there are companies that just can’t pay. Ebooks or music apps have to give a fixed percentage of their top line to rights-holders and don’t have a 30% margin to give Apple.

We had all these arguments in 2011 and very little has changed since: I wrote this report at the time and it’s still a pretty good summary.


View fullsize
Screenshot 2020-06-21 at 12.28.29 pm.png


View fullsize
Screenshot 2020-06-21 at 12.29.23 pm.png


View fullsize
Screenshot 2020-06-21 at 12.29.43 pm.png


View fullsize
Screenshot 2020-06-21 at 12.29.53 pm.png


View fullsize
Screenshot 2020-06-21 at 12.30.01 pm.png


View fullsize
Screenshot 2020-06-21 at 12.30.13 pm.png

Meanwhile, none of this is a surprise to Apple. As part of the recent US congress competition hearings, we saw an email from early 2011 in which Steve Jobs explicitly accepted and embraced the fact that the payment rules would be a fundamental problem for some companies.  The result, for almost a decade, has been a horrible muddle, with people using those products forced into a bad user experience.

Screenshot 2020-08-13 at 2.35.17 pm.png

(Of course, when this email was written, Apple was still a fair way away from market dominance: there were only 150-160m iOS devices in use, and iPhones were maybe 10% of all the mobile phones being used in the USA, where today they’re over 50%.)

Ironically, Epic is not in this ‘can’t pay’ category at all, and it built a huge and very profitable business following Apple’s rules. Unlike Spotify, it doesn’t have marginal cost for in-app purchases and there’s no structural reason why it can’t pay Apple (or indeed Google). It just doesn’t want to, or wants to pay less than 30%. Indeed, one could point out that the real issue is that Epic just doesn’t ’like’ Apple’s model, just as Apple doesn’t ‘like’ Stadia.

At this point, many people suggest that we can cut a Gordian knot here – that we can slash through the complexity by letting people have a choice. Allow any payment service, perhaps in parallel to Apple’s; allow third-party app stores; allow side-loading of apps; and of course let users turn off the sandboxed restrictions on the phone. Then you can have the security if you want it, or the freedom.

Unfortunately, you can’t have your cake and eat it. A secure system with a switch to turn off the security might work for Linux and a highly technical user, but when you’ve given smartphones to a few billion people, a secure system with a switch to turn off the security is just a target for malware. That horoscope app can tell you’ll get more accurate results if it has access to some computer gibberish, so please press OK, and guess what? Everyone will press OK. A computer should not ask a question that the user won’t understand, and when you have billions of users that list looks different. This has been Google’s experience with Android: it chose a less restrictive sandbox than iOS and had many more malware problems, and Google has spent the last decade slowly rowing towards Apple’s approach.

A limited version of this argument, incidentally, is that all the problems are with the store, and you could get rid of it without security problems, relying on software sandboxing on the phones to handle all security and safety issues. But there are also policies people object to on the phone itself (no replacing the default Maps app, say), and policies that we want to keep that are enforced in the store rather than on the phone (no ads in apps for kids, say). You have to tackle the whole policy question, not just part of it, and you cannot rely on the sandbox on the phone to solve all safety and security attacks.

All of this is to say that the demands Epic makes in its lawsuit are not, in fact, merely arguing that the smartphone apps market should be more competitive, with more payment options. The sandboxed app store model is not some curious, incidental feature of modern smartphones – rather, this is an essential and hugely important part of why they have such a strong software ecosystem. Epic is explicitly arguing that we should abandon the smartphone software model and security model almost entirely, and switch to what would actually be the old Windows model. Its arguments would also of course mean that we should abandon any level playing field, and move to a model where big companies and big brands have an even bigger advantage, because a trusted platform is replaced by a trusted reputation. This would be good for big established brands – like Epic – but not for may other people.

Epic’s proposal is full of holes, and Epic’s problem is really pretty peripheral, but I’m much more interested in Spotify and Stadia, where the situation now looks unsustainable, and that’s where we’re more likely to see changes. So, I think we should try to draw one more set of distinctions.

First, the App Store moderation problems are infuriating but they’re not rent-seeking or necessarily market abuse – they’re an execution failure, and indeed we’ve been here before. However, the EU, which is becoming the tech regulator by default, is already working on plans for the store review model to be regulated, with real, external rights of appeal and review, and external transparency. Apple could try to get ahead of this, or it might be too late. It might have no choice but to allow Stadia, and ‘we just don’t like that’ won’t do anymore. 

Second, I think Apple is going to have to make fundamental changes to the payment model. Epic only has margin at stake, but Spotify can’t pay at all, it’s a direct competitor, and there’s no user benefit at all to Apple’s policy, just confusion and annoyance. The EU is now pursuing two separate competition policy cases against Apple: one over the App Store, with Spotify a complainant, and the other over Apple Wallet and Apple Pay. This second one is instructive: the EU is taking the view that Apple has a monopoly of payment on the iPhone. Market definition is everything. I-am-not-a-lawyer, but I don’t see how Apple can win on Spotify (or Kindle), and I don’t think it should.

That might mean changes in who and what is covered by payment rules, but it probably also means changes to the 30%. There’s a lot of argument about principle, but there’s also a price: if the rate was, say, 10%, I’m not sure that we would be having the same conversation, and Epic would certainly get less sympathy.

That 30% adds up to real money, incidentally. When the store launched, Steve Jobs said it was aiming to break even – the 30% was to cover the running costs, and it is worth remembering how many huge companies are getting the App Store, the manual review and the file downloads to hundreds of millions of users for nothing more than their $100 a year developer subscription. But the App Store is not running at break even anymore: in 2019 it made somewhere between $10bn and $15bn of commission – 20-30% of the ‘service revenue’ Apple likes to talk about.

Finally – we’ve been arguing about this since the store launched in 2008, but really, some of this debate is as old as personal computers. Right back to the 1970s, there’s been a religious split between people who want computers that they’re free to change however they like, and people who want computers that are easy and safe to use for as many people as possible. This is a trade-off, but there’s a certain kind of person in tech that thinks app stores and the iOS sandbox have nothing to do with the success of smartphones and the iPhone – they’re just a stupid Apple thing you could get rid of with no ill effects. 30 years ago they thought the same about GUIs, and indeed a lot of Epic’s PR comes straight out of furious Usenet posts from the 1990s about how GUIs are evil and infantilising. But the whole direction of computing since the Apple 1 has been about more abstraction, less access to the lower levels of the system and inherent in that more accessibility for more people.

Apple has always been at one, extreme, end of that debate, taking a strong opinion on how it thought a good computer ‘should’ work and letting you choose it or not. From 1976 to, say, 2015 or so, it was just one fairly niche vendor, and some people chose Apple’s opinion and some didn’t. But with the iPhone, Apple finally won the argument with users’ wallets, and that means it’s not niche anymore – Apple has become the navy, and different rules apply.

Subscribers to my premium newsletter were sent a version of this on Sunday.

Would breaking up ‘big tech’ work? What would?


This post is by Benedict Evans from Essays - Benedict Evans

We’re clearly going to be arguing about the size, power and market share of large technology companies a great deal in the next couple of years. Many of the underlying concerns we have around technology are complicated, and involve deep-seated trade-offs where we actually have to make choices, and not everything is a competition problem anyway ( I wrote about this here). But if we presume that something is a competition problem, what do we do about it? The discussion here tends to jump straight to ‘break them up’, which also means presuming that break-ups would actually work. I’m not sure about that. 

The folk memory here, of course, is Standard Oil. John Rockefeller built a network of production, processing and distribution companies that he bundled, tied and cross-leveraged in all sorts of ruthless and devious ways to squeeze out competition. Then in 1911, when Standard Oil was forcibly split up into over 30 different companies, that market power was broken and the oil industry became competitive again, or so the story goes. 

This is a great story, but I’d suggest it’s more useful to look at, and contrast, the breakup of AT&T and the proposed breakup of Microsoft, which give a rather more mixed picture. 

In 1982 AT&T settled an anti-trust case by splitting off its local access telephone network into seven regional companies (the Regional Bell Operating Companies or ‘Baby Bells’), keeping long-distance and the telecoms equipment business (which was also later split off and is now part of Nokia). This is another of the stories that anti-trust lawyers tell their young around the campfire, but its real effect was limited in important ways. Splitting local from long-distance opened up a new market for competitive long-distance carriers, and the market for telecoms equipment was liberated by breaking one customer into eight. It also led to the emergence of companies building fibre networks in city centres to connect corporate customers. However, if you were a normal consumer, and lived in the suburbs of New York or Miami, and wanted a telephone, there was still a monopoly. There was commotion for long-distance, but not for your phone line; a national (near) monopoly had just been replaced by local monopolies. 

There’s a pretty good law-of-physics reason for this: fixed-line local access telephone networks are a natural monopoly, in much the same way as water, gas or electricity networks. split the wires in half, and Building a network of copper wires to every home in a neighbourhood is not quite as expensive as laying water pipes, but it’s expensive enough, with a long payback period, and it’s very hard to cover the cost of building two parallel networks. It’s not impossible – cable TV companies did it by selling a separate and much more expensive product, but then we also don’t have multiple, parallel CATV networks either. And, of course, you can’t split the wire going into your home into two and give each half to a different company.

If local access telephone networks are mostly a natural monopoly that cannot be made competitive by break-ups, what about network effects? Two decades after AT&T was broken up, the DoJ proposed that Microsoft should be broken up – that it be split into Office and Windows. As we know, this didn’t happen, but what it it had? What would that have changed?

Going back to 1911, splitting up Standard Oil did three things. First, it replaced a company that was often the only buyer or the only seller with many competing companies. Second, it addressed the cross-leveraging, bundling and tying whereby the oil fields, refineries, pipelines, rail cars and retailers all worked together to squeeze out competition, by breaking those into separate competing companies. And third, more abstractly, it replaced a huge company with huge financial and market power with many smaller companies with less individual mass. 

Now suppose that Windows and Office had become separate companies. So what? Well, the third point would be addressed; the overall mass of the company would be reduced. So, arguably, would the second; to the extent that you believe Microsoft was cross-leveraging Windows and Office, that would be ended. There would be no more Office/Windows bundles. 

However, it’s not clear that this would have resulted in more actual competition for Office or for Windows. There would not have been a wave of new companies making new PC operating systems, nor new PC productivity suites, any more than there was a wave of companies building new phone networks in American suburbs in the 1980s. Microsoft might have been doing all sorts of mean and sneaky things, but people used Windows and Office because of network effects, and those network effects were and are internal to each product. People used Windows because it had the software and people wrote software for Windows because it had the users, and that had very little to do with Office. If Office had been in a different company, that wouldn’t have prompted Adobe to port Photoshop to BeOS, nor id Software to write Quake for Mac before Windows. 

The strength of Windows was not that it was bundled or tied or leveraged, but that it had a network effect. The same for Office – everyone used Office because everyone used it, not because it was part of the same company as Windows. Breaking it up wouldn’t have changed this. 

Indeed, these network effects would have limited the companies emerging from a broken-up Microsoft (the ‘Baby Bills’) in just the same way that they limited everyone else. The Office company could not have made a new PC operating system to compete with Windows – no-one would have written software for it. The Windows company could not have made a new productivity suite to compete with Office – no-one would have used it, any more than anyone used Open Office. These are hypotheticals, but Microsoft really was caught by exactly this mechanic in mobile a decade later – no-one made apps for Window Phone because it had no users, and it had no users because no-one wrote apps for it, and all the power Microsoft had in Office and Windows meant nothing. 

In other words, one should think of network effects as comparable to a natural monopoly. In a network effect product as for a natural monopoly, once you have market dominance, that dominance persists not because of any anti-competitive behaviour by the company that owns it (even if there appears to be plenty) but because of the mechanics and economics of the product. 

Network effects do not dictate that there can be only one network – it depends on the market, just as you can have both cable TV and telephones on one street but only one water pipe. Hence, in early 1990s the PC market, with only 50-100m users globally, was too small to sustain more than one network – Microsoft won, Apple clung on in a niche and almost disappeared and the other contenders did disappear. The global smartphone market, with now over 4bn global users, is big enough for two networks – iOS and Android. In the early days of social networks many people thought there would be a winner in each region – Bebo was strong in the UK, Orkut in Brazil and so on – and this had happened with instant messaging in the first internet boom, but in the end Facebook turned out to have mostly global network effects. A few years later we had the same discussion about on-demand car services – many people thought thought that the network effects would be city-by-city, but in fact we had national and regional winners. However, in some countries the market did turn out to be big enough to sustain more than one network – in the USA both Lyft and Uber. 

Now, a generation after Microsoft’s antitrust case and two generations after AT&T’s breakup, we come to talking about Google, Apple, Facebook or Amazon. There is little serious talk of breaking up Apple, perhaps because it’s so obviously a single unit. There is some argument for splitting AWS apart from Amazon – I find this unconvincing (and I’ll return to this in a future essay) but regardless, that would still leave the Amazon retail business itself as a single hugely powerful company that’s generating a torrent of cash. But there’s a lot of talk of breaking apart Google and Facebook, and here I think comparisons with Standard Oil, AT&T and Microsoft are most interesting. On one hand, there are clearly divisible component parts (Youtube, Instagram etc) in a way that’s much less true for Apple and Amazon. But on the other hand, I’d suggest that, as for Office and Windows, the competitive strength of these component parts doesn’t come from the combined ownership, but from networks effects. Hence, breaking them apart might achieve very little.

As a first observation, Google and Facebook have two-sided business models: they address advertisers and they address consumers. There’s no question that they have market dominance in online advertising (especially if you define the relevant market for Google as search advertising and for Facebook as social advertising). Equally, there isn’t much question that they bundle and cross-leverage all of their different properties when doing business with advertisers. Break them up, and advertisers would have more leverage and the successor companies to Google and Facebook would have less leverage and less market power.

(Ironically, more leverage for advertisers over search or social networking companies would, all things being equal, mean less privacy for consumers, which isn’t typically what anti-trust advocates argue for and points to the fact that privacy isn’t necessarily a straightforward competition problem – but then, real policy is about trade-offs). 

However, though advertisers could now play Facebook off against Instagram and Google against Youtube, consumers would have the same choices that they had before. Just as breaking up AT&T liberalised the telecoms equipment market but not the natural monopoly local access market, changing who owns Instagram doesn’t alter the network effects that make Instagram strong, nor YouTube, nor WhatsApp, because, as for Office or Windows, the network effects are internal to the product. You don’t use WhatsApp because Facebook owns it. Google Search isn’t far ahead of Bing because it also owns Youtube. And yes, just as Microsoft was accused of doing all sorts of things to cross-leverage its businesses, so are these companies, but that’s ultimately peripheral – the market dominance comes from the products themselves.

At this point it’s sometimes suggested that if Google and YouTube became separate companies, Google would build a new video sharing product and Youtube would make an important new search engine. This is hard to take seriously – all the reasons that ‘Office Inc’ and ‘Windows Inc’ could not have competed with each other apply here in the same ways. ‘Youtube Inc’ would have none of the data and much more importantly none of the ongoing network effects that make Google a leader in search – it would start not just far behind Google but far behind Bing. Equally, there’s no reason for Google’s new video site to do any better than its last one – it would be on the wrong side of network effects. Indeed, the basic problems with this idea becomes clearer if one asks, rhetorically, why it is that Facebook does not already compete with Google in search, or why Google has failed so many times at creating social products itself? Why would Youtube Inc’s search engine do any better than Bing – what special advantage would it have? This is all so much magical thinking.

🤔

If network effects are equivalent to natural monopolies, and the market position of some of the companies that you worry about are based by network effects, what do you do? Well, when faced by a natural monopoly with problems, we don’t just shrug and give up – we regulate it. 

Going back to AT&T, the local access network is a natural monopoly, but you can unbundle competitive access to that ‘last mile’ of copper at the local exchange, with wholesale access to the data streams or direct physical access to the actual copper wire itself as it comes into the building. The trigger for this was DSL. In the USA unbundling this access this was called ‘UNE-P’ and lasted a short while before being shut down, returning the copper monopoly to the Baby Bells – outside the USA regulators persevered (calling it ‘unbundling the local loop’), creating an entire new competitive layer in local access. The chart below shows the result: in pretty much every large country in Europe the former monopoly (‘the ‘incumbent’) now has less than 50% share of DSL. It still owns the copper, but it rents it out for other people to provide services on top, under a legally controlled wholesale model. The result is that in these countries most consumers have a choice of a dozen or more broadband providers. You can, in fact, combine a natural monopoly with competition.

Screenshot 2020-08-10 at 11.05.20 am.png

A little later, something similar happened to roaming prices in Europe. Phone calls and data had become cheap, but roaming prices had not, and stories were widespread of hapless tourists getting vast bills for trivial amounts of use when they turned their phone on abroad. The EU responded with a set of rules that removed the consumer harm, and today roaming in Europe is effectively free.

Screenshot 2020-08-08 at 4.13.01 pm.png

The EU went about this by constructing an argument that this was a competition problem: the roaming price you were charged was a function of the wholesale rate agreed between the host operator and your operator, and you had no say in this. You could argue that this is this pretty tendentious, but it doesn’t matter – the regulator found a legal mechanism to address a real consumer harm. (Ironically, a decade earlier Vodafone, which had networks in most European markets, had tried to sell a discounted roaming deal across those networks and was blocked by the EU on the grounds that since other operators could not match this it was anti-competitive.)

Something rather similar has happened over the last five years in European credit card interchange rates. When you swipe your card the retailer is charged a fee by Visa, MasterCard or Amex; the retailer can’t negotiate this and can’t chose not to support those card providers, so this really is a competitive question. Starting from 2015, the EU has capped these pieces, pushing down interchange rates. (These rates are where loyalty points come from, so this has also reduced the value of such schemes to Europeans).

What all of these have in common is that regulators inserted competition, cut prices, or both, by digging deep inside a monopoly or oligopoly and addressing mechanics, infrastructure and internal pricing schemes that consumers never saw. They didn’t ‘break them up’ – they mandated wholesale access or price changes to things that you would never see on the P&L. Local loop unbundling came with very specific rules and pricing about every aspect connecting to the local access network. As we look at the regulation of parts of the technology industry today, we can see some pretty similar things coming.

Hence, the UK’s competition authority, the CMA, analyses Google’s dominance position in search, and doesn’t propose breaking it up, because it understands that search is (mostly) a natural monopoly. Instead, it proposes a long list of highly specific internal, mechanical interventions. For example:

  • “The power to require Google to provide click and query data to third-party search engines to allow them to improve their search algorithms”

  • “The power to restrict Google’s ability to secure default positions, to restrict the monetisation of default positions on devices (i.e. Apple selling the default search engine slot to Google) and to introduce choice screens”

  • “ Facebook should offer a defined find contacts service to users of a third-party platform, but rival platforms should not be required to reciprocate”

View fullsize

1.png

View fullsize

2.png

View fullsize

3.png

View fullsize

4.png

There’s lots to argue about in specific proposals like this (including how much of it will be enacted), but that’s not really the point – rather, one should ask which problems you can resolve by splitting the company apart, or by fining people, and which by getting right inside the operations and writing rules. As I pointed out here, we didn’t make cars safer by breaking up GM or Ford, but by writing rules about how you can make a car.

However, while the US does regulates cars (and many other things), most of my examples come from Europe, and this points to two distinct problems. 

First, what happens when monolithic global software systems are regulated by different authorities in different places? What if they mandate things that are mutually contradictory? Worse, what if those contradictions reflect fundamental differences in philosophy? Adtech is relatively apolitical, but attitudes to free speech vary in important ways even between different liberal democracies. 

Second, different jurisdictions can have rather different operating models for regulation itself. The US tends to have a rules-based, lawyer-led system that moves forward one court case at a time, whereas Europe tends to have a principle-based, outcome-based, practitioner-led system. You can see that very clearly in the images of the CMA report above. The US discussion tends to circle back to the Sherman Antitrust act of 1890 and what is or is not a violation, whereas the CMA argues for a new regulator that can write new rules about the operations of Google’s data centre whenever it thinks necessary.

Finally – markets change, and in technology they change vey fast. Detailed, line-by-line regulation of the internal operations of a company might make sense when the market is set in place for 50 years, but IBM’s mainframes dominated the tech industry for only 15 years, and Windows/Intel for only another 15, and as I wrote here, neither one lost their dominance because of anti-trust, but because the whole basis of of their dominance became irrelevant. There are people applying to YC now who weren’t born the last time anyone started a company to write Windows software. 

This of course takes us back to the shift in regulatory models. Competition regulators are very conscious that they move too slowly. If you take five or ten years to resolve a complaint, then the company being harmed might have disappeared, the people who did the harm have moved to other jobs and forgotten all about it, and more fundamentally the whole market structure might have changed again – you can fine people, but it’s far too late. Hence, a big part of the shift in regulator attitudes is a shift to ex ante regulation – to thinking about what might happen instead of what did happen. (In the same vein, US regulators are also starting to think about whether moving away from their historic narrow focus on low prices for consumer might be a good idea, when looking at companies whose entire model is to be cheap or free.)

Of course, predictions are hard. The main reason that Americans do now have a (moderately) competitive market for telephones is that a completely different set of physics came along, in the form of cellular, where you actually can justify building three or four competing networks. Ironically, the legendary McKinsey study that said mobile would be a tiny market, and that there would only be 900,000 mobile subscribers in the USA by 2000, was commissioned by AT&T as part of this process – when AT&T was broken up, no-one expected mobile to provide mass-market competition for telephone service. Equally, the anti-trust process that Microsoft went through 20 year ago was utterly ineffective – but then a few years later smartphones turned Windows PCs into accessories. That’s not an argument against regulation, but it may be an argument for humility.

Would breaking up ‘big tech’ work? What would?


This post is by Benedict Evans from Essays - Benedict Evans

We’re clearly going to be arguing about the size, power and market share of large technology companies a great deal in the next couple of years. Many of the underlying concerns we have around technology are complicated, and involve deep-seated trade-offs where we actually have to make choices, and not everything is a competition problem anyway ( I wrote about this here). But if we presume that something is a competition problem, what do we do about it? The discussion here tends to jump straight to ‘break them up’, which also means presuming that break-ups would actually work. I’m not sure about that. 

The folk memory here, of course, is Standard Oil. John Rockefeller built a network of production, processing and distribution companies that he bundled, tied and cross-leveraged in all sorts of ruthless and devious ways to squeeze out competition. Then in 1911, when Standard Oil was forcibly split up into over 30 different companies, that market power was broken and the oil industry became competitive again, or so the story goes. 

This is a great story, but I’d suggest it’s more useful to look at, and contrast, the breakup of AT&T and the proposed breakup of Microsoft, which give a rather more mixed picture. 

In 1982 AT&T settled an anti-trust case by splitting off its local access telephone network into seven regional companies (the Regional Bell Operating Companies or ‘Baby Bells’), keeping long-distance and the telecoms equipment business (which was also later split off and is now part of Nokia). This is another of the stories that anti-trust lawyers tell their young around the campfire, but its real effect was limited in important ways. Splitting local from long-distance opened up a new market for competitive long-distance carriers, and the market for telecoms equipment was liberated by breaking one customer into eight. It also led to the emergence of companies building fibre networks in city centres to connect corporate customers. However, if you were a normal consumer, and lived in the suburbs of New York or Miami, and wanted a telephone, there was still a monopoly. There was commotion for long-distance, but not for your phone line; a national (near) monopoly had just been replaced by local monopolies. 

There’s a pretty good law-of-physics reason for this: fixed-line local access telephone networks are a natural monopoly, in much the same way as water, gas or electricity networks. Building a network of copper wires to every home in a neighbourhood is not quite as expensive as laying water pipes, but it’s expensive enough, with a long payback period, and it’s very hard to cover the cost of building two parallel networks. It’s not impossible – cable TV companies did it by selling a separate and much more expensive product, but then we also don’t have multiple, parallel CATV networks either. And, of course, you can’t split the wire going into your home into two and give each half to a different company.

If local access telephone networks are mostly a natural monopoly that cannot be made competitive by break-ups, what about network effects? Two decades after AT&T was broken up, the DoJ proposed that Microsoft should be broken up – that it be split into Office and Windows. As we know, this didn’t happen, but what it it had? What would that have changed?

Going back to 1911, splitting up Standard Oil did three things. First, it replaced a company that was often the only buyer or the only seller with many competing companies. Second, it addressed the cross-leveraging, bundling and tying whereby the oil fields, refineries, pipelines, rail cars and retailers all worked together to squeeze out competition, by breaking those into separate competing companies. And third, more abstractly, it replaced a huge company with huge financial and market power with many smaller companies with less individual mass. 

Now suppose that Windows and Office had become separate companies. So what? Well, the third point would be addressed; the overall mass of the company would be reduced. So, arguably, would the second; to the extent that you believe Microsoft was cross-leveraging Windows and Office, that would be ended. There would be no more Office/Windows bundles. 

However, it’s not clear that this would have resulted in more actual competition for Office or for Windows. There would not have been a wave of new companies making new PC operating systems, nor new PC productivity suites, any more than there was a wave of companies building new phone networks in American suburbs in the 1980s. Microsoft might have been doing all sorts of mean and sneaky things, but people used Windows and Office because of network effects, and those network effects were and are internal to each product. People used Windows because it had the software and people wrote software for Windows because it had the users, and that had very little to do with Office. If Office had been in a different company, that wouldn’t have prompted Adobe to port Photoshop to BeOS, nor id Software to write Quake for Mac before Windows. 

The strength of Windows was not that it was bundled or tied or leveraged, but that it had a network effect. The same for Office – everyone used Office because everyone used it, not because it was part of the same company as Windows. Breaking it up wouldn’t have changed this. 

Indeed, these network effects would have limited the companies emerging from a broken-up Microsoft (the ‘Baby Bills’) in just the same way that they limited everyone else. The Office company could not have made a new PC operating system to compete with Windows – no-one would have written software for it. The Windows company could not have made a new productivity suite to compete with Office – no-one would have used it, any more than anyone used Open Office. These are hypotheticals, but Microsoft really was caught by exactly this mechanic in mobile a decade later – no-one made apps for Window Phone because it had no users, and it had no users because no-one wrote apps for it, and all the power Microsoft had in Office and Windows meant nothing. 

In other words, one should think of network effects as comparable to a natural monopoly. In a network effect product as for a natural monopoly, once you have market dominance, that dominance persists not because of any anti-competitive behaviour by the company that owns it (even if there appears to be plenty) but because of the mechanics and economics of the product. 

Network effects do not dictate that there can be only one network – it depends on the market, just as you can have both cable TV and telephones on one street but only one water pipe. Hence, in early 1990s the PC market, with only 50-100m users globally, was too small to sustain more than one network – Microsoft won, Apple clung on in a niche and almost disappeared and the other contenders did disappear. The global smartphone market, with now over 4bn global users, is big enough for two networks – iOS and Android. In the early days of social networks many people thought there would be a winner in each region – Bebo was strong in the UK, Orkut in Brazil and so on – and this had happened with instant messaging in the first internet boom, but in the end Facebook turned out to have mostly global network effects. A few years later we had the same discussion about on-demand car services – many people thought thought that the network effects would be city-by-city, but in fact we had national and regional winners. However, in some countries the market did turn out to be big enough to sustain more than one network – in the USA both Lyft and Uber. 

Now, a generation after Microsoft’s antitrust case and two generations after AT&T’s breakup, we come to talking about Google, Apple, Facebook or Amazon. There is little serious talk of breaking up Apple, perhaps because it’s so obviously a single unit. There is some argument for splitting AWS apart from Amazon – I find this unconvincing (and I’ll return to this in a future essay) but regardless, that would still leave the Amazon retail business itself as a single hugely powerful company that’s generating a torrent of cash. But there’s a lot of talk of breaking apart Google and Facebook, and here I think comparisons with Standard Oil, AT&T and Microsoft are most interesting. On one hand, there are clearly divisible component parts (Youtube, Instagram etc) in a way that’s much less true for Apple and Amazon. But on the other hand, I’d suggest that, as for Office and Windows, the competitive strength of these component parts doesn’t come from the combined ownership, but from networks effects. Hence, breaking them apart might achieve very little.

As a first observation, Google and Facebook have two-sided business models: they address advertisers and they address consumers. There’s no question that they have market dominance in online advertising (especially if you define the relevant market for Google as search advertising and for Facebook as social advertising). Equally, there isn’t much question that they bundle and cross-leverage all of their different properties when doing business with advertisers. Break them up, and advertisers would have more leverage and the successor companies to Google and Facebook would have less leverage and less market power.

(Ironically, more leverage for advertisers over search or social networking companies would, all things being equal, mean less privacy for consumers, which isn’t typically what anti-trust advocates argue for and points to the fact that privacy isn’t necessarily a straightforward competition problem – but then, real policy is about trade-offs). 

However, though advertisers could now play Facebook off against Instagram and Google against Youtube, consumers would have the same choices that they had before. Just as breaking up AT&T liberalised the telecoms equipment market but not the natural monopoly local access market, changing who owns Instagram doesn’t alter the network effects that make Instagram strong, nor YouTube, nor WhatsApp, because, as for Office or Windows, the network effects are internal to the product. You don’t use WhatsApp because Facebook owns it. Google Search isn’t far ahead of Bing because it also owns Youtube. And yes, just as Microsoft was accused of doing all sorts of things to cross-leverage its businesses, so are these companies, but that’s ultimately peripheral – the market dominance comes from the products themselves.

At this point it’s sometimes suggested that if Google and YouTube became separate companies, Google would build a new video sharing product and Youtube would make an important new search engine. This is hard to take seriously – all the reasons that ‘Office Inc’ and ‘Windows Inc’ could not have competed with each other apply here in the same ways. ‘Youtube Inc’ would have none of the data and much more importantly none of the ongoing network effects that make Google a leader in search – it would start not just far behind Google but far behind Bing. Equally, there’s no reason for Google’s new video site to do any better than its last one – it would be on the wrong side of network effects. Indeed, the basic problems with this idea becomes clearer if one asks, rhetorically, why it is that Facebook does not already compete with Google in search, or why Google has failed so many times at creating social products itself? Why would Youtube Inc’s search engine do any better than Bing – what special advantage would it have? This is all so much magical thinking.

🤔

If network effects are equivalent to natural monopolies, and the market position of some of the companies that you worry about are based by network effects, what do you do? Well, when faced by a natural monopoly with problems, we don’t just shrug and give up – we regulate it. 

Going back to AT&T, the local access network is a natural monopoly, but you can unbundle competitive access to that ‘last mile’ of copper at the local exchange, with wholesale access to the data streams or direct physical access to the actual copper wire itself as it comes into the building. The trigger for this was DSL. In the USA unbundling this access this was called ‘UNE-P’ and lasted a short while before being shut down, returning the copper monopoly to the Baby Bells – outside the USA regulators persevered (calling it ‘unbundling the local loop’), creating an entire new competitive layer in local access. The chart below shows the result: in pretty much every large country in Europe the former monopoly (‘the ‘incumbent’) now has less than 50% share of DSL. It still owns the copper, but it rents it out for other people to provide services on top, under a legally controlled wholesale model. The result is that in these countries most consumers have a choice of a dozen or more broadband providers. You can, in fact, combine a natural monopoly with competition.

Screenshot 2020-08-10 at 11.05.20 am.png

A little later, something similar happened to roaming prices in Europe. Phone calls and data had become cheap, but roaming prices had not, and stories were widespread of hapless tourists getting huge bills for trivial amounts of use when they turned their phone on abroad. The EU responded with a set of rules that removed the consumer harm, and today roaming in Europe is effectively free.

Screenshot 2020-08-08 at 4.13.01 pm.png

The EU went about this by constructing an argument that this was a competition problem: the roaming price you were charged was a function of the wholesale rate agreed between the host operator and your operator, and you had no say in this. You could probably debate whether this really is a competition problem, but it doesn’t matter – the regulator found a legal mechanism to address a real consumer harm. (Ironically, a decade earlier Vodafone, which had networks in most European markets, had tried to sell a discounted roaming deal across those networks and was blocked by the EU on the grounds that since other operators could not match this it was anti-competitive.)

Something rather similar has happened over the last five years in European credit card interchange rates. When you swipe your card the retailer is charged a fee by Visa, MasterCard or Amex: the retailer can’t negotiate this and can’t chose not to support those card providers, so this really is a competitive question. So, starting from 2015, the EU has capped these prices, pushing down interchange rates. (These rates are where loyalty points come from, so this has also reduced the value of such schemes to Europeans.)

What all of these have in common is that regulators inserted competition, cut prices, or both, by digging deep inside a monopoly or oligopoly and addressing mechanics, infrastructure and internal pricing schemes that consumers never saw. They didn’t ‘break them up’ – they mandated wholesale access or price changes to things that you would never see on the P&L. Local loop unbundling came with very specific rules and pricing about every aspect of connecting to the local access network. As we look at the regulation of parts of the technology industry today, we can see some pretty similar things coming.

Hence, the UK’s competition authority, the CMA, analyses Google and Facebook’s dominant positions, and doesn’t focus on breaking them up. Instead, it proposes a long list of highly specific internal, mechanical interventions. For example:

  • “The power to require Google to provide click and query data to third-party search engines to allow them to improve their search algorithms”

  • “The power to restrict Google’s ability to secure default positions, to restrict the monetisation of default positions on devices [i.e. Apple selling the default search engine slot to Google] and to introduce choice screens”

  • “ Facebook should offer a defined find contacts service to users of a third-party platform, but rival platforms should not be required to reciprocate”

View fullsize

1.png

View fullsize

2.png

View fullsize

3.png

View fullsize

4.png

There’s lots to argue about in specific proposals like this (including how much of it will be enacted), but that’s not really the point – rather, one should ask which problems you can resolve by splitting the company apart, or by fining people, and which by getting right inside the operations and writing rules. As I pointed out here, we didn’t make cars safer by breaking up GM or Ford, but by writing rules about how you can make a car.

However, while the US does regulates cars (and many other things), most of my examples come from Europe, and this points to two distinct problems. 

First, what happens when monolithic global software systems are regulated by different authorities in different places? What if those authorities mandate things that are mutually contradictory? Worse, what if those contradictions reflect fundamental differences in philosophy? Adtech is relatively apolitical, but attitudes to free speech vary in important ways even between different liberal democracies. 

Second, different jurisdictions can have rather different operating models for regulation itself. The US tends to have a rules-based, lawyer-led system that moves forward one court case at a time, whereas Europe tends to have a principle-based, outcome-based, practitioner-led system. You can see that very clearly in the images of the CMA report above. The US discussion tends to circle back to the Sherman Antitrust act of 1890 and what is or is not a violation, whereas the CMA argues for a new regulator that can write new rules about the operations of Google’s data centre whenever it thinks necessary.

Finally – markets change, and in technology they change very fast. Detailed, line-by-line regulation of the internal operations of a company is straight-forward when the market is set in place for 50 years, but IBM’s mainframes dominated the tech industry for only 15 years, and Windows/Intel for only another 15, and as I wrote here, neither one lost their dominance because of anti-trust, but because the whole basis of of their dominance became irrelevant. There are people applying to YC now who weren’t born the last time anyone started a company to write Windows software. 

This of course takes us back to the shift in regulatory models. Competition regulators are very conscious that they have moved too slowly. In tech, if you take five or ten years to resolve a complaint, then the company being harmed might have disappeared, the people who did the harm have moved to other jobs and forgotten all about it, and more fundamentally the whole market structure might have changed again – you can fine people, but it’s far too late. Hence, a big part of the shift in regulator attitudes is a shift to ex ante regulation – to thinking about what might happen instead of what did happen. (In the same vein, US regulators are also starting to think about whether moving away from their historic narrow focus on low prices for consumer might be a good idea, when looking at companies whose entire model is to be cheap or free.)

Of course, predictions are hard. The main reason that Americans do now have a (moderately) competitive market for telephones is that a completely different set of physics came along, in the form of cellular, where you actually can justify building three or four competing networks. Ironically, the legendary McKinsey study that said mobile would be a tiny market, and that there would only be 900,000 mobile subscribers in the USA by 2000, was commissioned by AT&T as part of this – when AT&T was broken up, no-one expected mobile to provide mass-market competition for telephone service. Equally, the anti-trust process that Microsoft went through 20 year ago was utterly ineffective – but then a few years later smartphones turned Windows PCs into accessories and Microsoft from a monopolist into just another big tech company. That’s not an argument against regulation, but it may be an argument for humility.

Would breaking up ‘big tech’ work? What would?


This post is by Benedict Evans from Essays - Benedict Evans

We’re clearly going to be arguing about the size, power and market share of large technology companies a great deal in the next couple of years. Many of the underlying concerns we have around technology are complicated, and involve deep-seated trade-offs where we actually have to make choices, and not everything is a competition problem anyway ( I wrote about this here). But if we presume that something is a competition problem, what do we do about it? The discussion here tends to jump straight to ‘break them up’, which also means presuming that break-ups would actually work. I’m not sure about that. 

The folk memory here, of course, is Standard Oil. John Rockefeller built a network of production, processing and distribution companies that he bundled, tied and cross-leveraged in all sorts of ruthless and devious ways to squeeze out competition. Then in 1911, when Standard Oil was forcibly split up into over 30 different companies, that market power was broken and the oil industry became competitive again, or so the story goes. 

This is a great story, but I’d suggest it’s more useful to look at, and contrast, the breakup of AT&T and the proposed breakup of Microsoft, which give a rather more mixed picture. 

In 1982 AT&T settled an anti-trust case by splitting off its local access telephone network into seven regional companies (the Regional Bell Operating Companies or ‘Baby Bells’), keeping long-distance and the telecoms equipment business (which was also later split off and is now part of Nokia). This is another of the stories that anti-trust lawyers tell their young around the campfire, but its real effect was limited in important ways. Splitting local from long-distance opened up a new market for competitive long-distance carriers, and the market for telecoms equipment was liberated by breaking one customer into eight. It also led to the emergence of companies building fibre networks in city centres to connect corporate customers. However, if you were a normal consumer, and lived in the suburbs of New York or Miami, and wanted a telephone, there was still a monopoly. There was competition for long-distance, but not for your phone line; a national (near) monopoly was replaced by local monopolies. 

There’s a pretty good law-of-physics reason for this: fixed-line local access telephone networks are a natural monopoly, in much the same way as water, gas or electricity networks. Building a network of copper wires to every home in a neighbourhood is not quite as expensive as laying water pipes, but it’s expensive enough, with a long payback period, and it’s very hard to cover the cost of building two parallel networks. It’s not impossible – cable TV companies did it by selling a separate and much more expensive product, but then we also don’t have multiple, parallel CATV networks either. And, of course, you can’t split the wire going into your home into two and give each half to a different company.

If local access telephone networks are mostly a natural monopoly that cannot be made competitive by break-ups, what about network effects? Two decades after AT&T was broken up, the DoJ proposed that Microsoft should be broken up – that it be split into Office and Windows. As we know, this didn’t happen, but what it it had? What would that have changed?

Going back to 1911, splitting up Standard Oil did three things. First, it replaced a company that was often the only buyer or the only seller with many competing companies. Second, it addressed the cross-leveraging, bundling and tying whereby the oil fields, refineries, pipelines, rail cars and retailers all worked together to squeeze out competition, by breaking those into separate competing companies. And third, more abstractly, it replaced a huge company with huge financial and market power with many smaller companies with less individual mass. 

Now suppose that Windows and Office had become separate companies. So what? Well, the third point would be addressed; the overall mass of the company would be reduced. So, arguably, would the second; to the extent that you believe Microsoft was cross-leveraging Windows and Office, that would be ended. There would be no more Office/Windows bundles. 

However, it’s not clear that this would have resulted in more actual competition for Office or for Windows. There would not have been a wave of new companies making new PC operating systems, nor new PC productivity suites, any more than there was a wave of companies building new phone networks in American suburbs in the 1980s. Microsoft might have been doing all sorts of mean and sneaky things, but people used Windows and Office because of network effects, and those network effects were and are internal to each product. People used Windows because it had the software and people wrote software for Windows because it had the users, and that had very little to do with Office. If Office had been in a different company, that wouldn’t have prompted Adobe to port Photoshop to BeOS, nor id Software to write Quake for Mac before Windows. 

The strength of Windows was not that it was bundled or tied or leveraged, but that it had a network effect. The same for Office – everyone used Office because everyone used it, not because it was part of the same company as Windows. Breaking them up wouldn’t have changed this. 

Indeed, these network effects would have limited the companies emerging from a broken-up Microsoft (the ‘Baby Bills’) in just the same way that they limited everyone else. The Office company could not have made a new PC operating system to compete with Windows – no-one would have written software for it. The Windows company could not have made a new productivity suite to compete with Office – no-one would have used it, any more than anyone used Open Office. These are hypotheticals, but Microsoft really was caught by exactly this mechanic in mobile a decade later – no-one made apps for Window Phone because it had no users, and it had no users because no-one wrote apps for it, and all the power Microsoft had in Office and Windows meant nothing. 

In other words, one should think of network effects as comparable to a natural monopoly. In a network effect product as for a natural monopoly, once you have market dominance, that dominance persists not because of any anti-competitive behaviour by the company that owns it (even if there appears to be plenty) but because of the mechanics and economics of the product. 

Network effects do not dictate that there can be only one network – it depends on the market, just as you can have both cable TV and telephones on one street but only one water pipe. Hence, in early 1990s the PC market, with only 50-100m users globally, was too small to sustain more than one network – Microsoft won, Apple clung on in a niche and almost disappeared and the other contenders did disappear. The global smartphone market, with now over 4bn global users, is big enough for two networks – iOS and Android. In the early days of social networks many people thought there would be a winner in each region – Bebo was strong in the UK, Orkut in Brazil and so on – and this had happened with instant messaging in the first internet boom, but in the end Facebook turned out to have mostly global network effects. A few years later we had the same discussion about on-demand car services – many people thought thought that the network effects would be city-by-city, but in fact we had national and regional winners. However, in some countries the market did turn out to be big enough to sustain more than one network – in the USA both Lyft and Uber. 

Now, a generation after Microsoft’s antitrust case and two generations after AT&T’s breakup, we come to talking about Google, Apple, Facebook or Amazon. There is little serious talk of breaking up Apple, perhaps because it’s so obviously a single unit. There is some argument for splitting AWS apart from Amazon – I find this unconvincing (and I’ll return to this in a future essay) but regardless, that would still leave the Amazon retail business itself as a single hugely powerful company that’s generating a torrent of cash. But there’s a lot of talk of breaking apart Google and Facebook, and here I think comparisons with Standard Oil, AT&T and Microsoft are most interesting. On one hand, there are clearly divisible component parts (Youtube, Instagram etc) in a way that’s much less true for Apple and Amazon. But on the other hand, I’d suggest that, as for Office and Windows, the competitive strength of these component parts doesn’t come from the combined ownership, but from networks effects. Hence, breaking them apart might achieve very little.

As a first observation, Google and Facebook have two-sided business models: they address advertisers and they address consumers. There’s no question that they have market dominance in online advertising (especially if you define the relevant market for Google as search advertising and for Facebook as social advertising). Equally, there isn’t much question that they bundle and cross-leverage all of their different properties when doing business with advertisers. Break them up, and advertisers would have more leverage and the successor companies to Google and Facebook would have less leverage and less market power.

(Ironically, more leverage for advertisers over search or social networking companies would, all things being equal, mean less privacy for consumers, which isn’t typically what anti-trust advocates argue for and points to the fact that privacy isn’t necessarily a straightforward competition problem – but then, real policy is about trade-offs). 

However, though advertisers could now play Facebook off against Instagram and Google against Youtube, consumers would have the same choices that they had before. Just as breaking up AT&T liberalised the telecoms equipment market but not the natural monopoly local access market, changing who owns Instagram doesn’t alter the network effects that make Instagram strong, nor YouTube, nor WhatsApp, because, as for Office or Windows, the network effects are internal to the product. You don’t use WhatsApp because Facebook owns it. Google Search isn’t far ahead of Bing because it also owns Youtube. And yes, just as Microsoft was accused of doing all sorts of things to cross-leverage its businesses, so are these companies, but that’s ultimately peripheral – the market dominance comes from the products themselves.

At this point it’s sometimes suggested that if Google and YouTube became separate companies, Google would build a new video sharing product and Youtube would make an important new search engine. This is hard to take seriously – all the reasons that ‘Office Inc’ and ‘Windows Inc’ could not have competed with each other apply here in the same ways. ‘Youtube Inc’ would have none of the ongoing network effects that make Google a leader in search – it would start not just far behind Google but far behind Bing. Equally, there’s no reason for Google’s new video site to do any better than its last one – it would be on the wrong side of network effects. Indeed, the weakness of this idea becomes clearer if one asks, rhetorically, why it is that Facebook does not already compete with Google in search, or why Google has failed so many times at creating social products itself? Why would Youtube Inc’s search engine do any better than Bing – what special advantage would it have? This is all just magical thinking.

🤔

If network effects are equivalent to natural monopolies, and the market position of some of the companies that you worry about are based on network effects, what do you do? Well, when faced by a natural monopoly with problems, we don’t just shrug and give up – we regulate it. 

Going back to AT&T, the local access network is a natural monopoly, but you can unbundle competitive access to that ‘last mile’ of copper at the local exchange, with wholesale access to the data streams or direct physical access to the actual copper wire itself as it comes into the building. The trigger for this was DSL. In the USA, unbundling this access this was called ‘UNE-P’ and lasted a short while before being shut down, returning the copper monopoly to the Baby Bells. Outside the USA, regulators persevered (calling it ‘unbundling the local loop’) and created an entire new competitive layer in local access. The chart below shows the result: in pretty much every large country in Europe the former monopoly (‘the ‘incumbent’) now has less than 50% share of DSL. It still owns the copper, but it rents it out for other people to provide services on top, under a legally controlled wholesale model. The result is that in these countries most consumers have a choice of a dozen or more broadband providers. You can, in fact, combine a natural monopoly with competition.

Screenshot 2020-08-10 at 11.05.20 am.png

A little later, something similar happened to roaming prices in Europe. Phone calls and data had become cheap, but roaming prices had not, and stories were widespread of hapless tourists getting huge bills for trivial amounts of use when they turned their phone on abroad. The EU responded with a set of rules that removed the consumer harm, and today roaming in Europe is effectively free.

Screenshot 2020-08-08 at 4.13.01 pm.png

The EU went about this by constructing an argument that this was a competition problem: the roaming price you were charged was a function of the wholesale rate agreed between the host operator and your operator, and you had no say in this. You could probably debate whether this really is a competition problem, but it doesn’t matter – the regulator found a legal mechanism to address a real consumer harm. (Ironically, a decade earlier Vodafone, which had networks in most European markets, had tried to sell a discounted roaming deal across those networks and was blocked by the EU on the grounds that since other operators could not match this it was anti-competitive. Yes, really.)

Something rather similar has happened over the last five years in European credit card interchange rates. When you swipe your card the retailer is charged a fee by Visa, MasterCard or Amex: the retailer can’t negotiate this and can’t chose not to support those card providers, so this really is a competitive question. Starting from 2015, the EU has capped these prices, pushing down interchange rates. (These rates are where loyalty points come from, so this has also reduced the value of such schemes to Europeans.)

What all of these have in common is that regulators inserted competition, cut prices, or both, by digging deep inside a monopoly or oligopoly and addressing mechanics, infrastructure and internal pricing schemes that consumers never see. They didn’t ‘break them up’ – they mandated wholesale access or price changes to things that you would never see on the P&L. Local loop unbundling came with very specific rules and pricing about every aspect of connecting to the local access network. As we look at the regulation of parts of the technology industry today, we can see some pretty similar things coming.

Hence, the UK’s competition authority, the CMA, analyses Google and Facebook’s dominant positions, and doesn’t focus on breaking them up. Instead, it proposes a long list of highly specific internal, mechanical interventions. For example:

  • “The power to require Google to provide click and query data to third-party search engines to allow them to improve their search algorithms”

  • “The power to restrict Google’s ability to secure default positions, to restrict the monetisation of default positions on devices [i.e. Apple selling the default search engine slot to Google] and to introduce choice screens”

  • “ Facebook should offer a defined find contacts service to users of a third-party platform, but rival platforms should not be required to reciprocate”

There’s lots to argue about in specific proposals like this (including how much of it will be enacted), but that’s not really the point – rather, one should ask which problems you can resolve by splitting the company apart, or by fining people, and which by getting right inside the operations and writing rules. As I pointed out here, we didn’t make cars safer by breaking up GM or Ford, but by writing rules about how you can make a car.

However, while the US does regulates cars (and many other things), most of my examples come from Europe, and this points to two distinct problems. 

First, what happens when monolithic global software systems are regulated by different authorities in different places? What if those authorities mandate things that are mutually contradictory? Worse, what if those contradictions reflect fundamental differences in philosophy? Adtech is relatively apolitical, but attitudes to free speech vary in important ways even between different liberal democracies. 

Second, different jurisdictions can have rather different operating models for regulation itself. The US tends to have a rules-based, lawyer-led system that moves forward one court case at a time, whereas Europe tends to have a principle-based, outcome-based, practitioner-led system. You can see that very clearly in the CMA report above. The US discussion tends to circle back to the Sherman Antitrust act of 1890 and what is or is not a violation, whereas the CMA argues for a new regulator that can write new rules about the operations of Google’s data centre whenever it thinks necessary.

Finally – markets change, and in technology they change very fast. Detailed, line-by-line regulation of the internal operations of a company is straight-forward when the market is set in place for 50 years, but IBM’s mainframes dominated the tech industry for only 15 years, and Windows/Intel for only another 15, and as I wrote here, neither one lost their dominance because of anti-trust, but because the whole basis of of their dominance became irrelevant. There are people applying to YC now who weren’t born the last time anyone started a company to write Windows software. 

This of course takes us back to the shift in regulatory models. Competition regulators are very conscious that they have moved too slowly. In tech, if you take five or ten years to resolve a complaint, then the company being harmed might have disappeared, the people who did the harm have moved to other jobs and forgotten all about it, and more fundamentally the whole market structure might have changed again – you can fine people, but it’s far too late. Hence, a big part of the shift in regulator attitudes is a shift to ex ante regulation – to thinking about what might happen instead of what did happen. (In the same vein, US regulators are also starting to think about whether moving away from their historic narrow focus on low prices for consumer might be a good idea, when looking at companies whose entire model is to be cheap or free.)

Of course, predictions are hard. The main reason that Americans do now have a (moderately) competitive market for telephones is that a completely different set of physics came along, in the form of cellular, where you actually can justify building three or four competing networks. Ironically, the legendary McKinsey study that said mobile would be a tiny market, and that there would only be 900,000 mobile subscribers in the USA by 2000, was commissioned by AT&T as part of this – when AT&T was broken up, no-one expected mobile to provide mass-market competition for telephone service. Equally, the anti-trust process that Microsoft went through 20 year ago was utterly ineffective – but then a few years later smartphones turned Windows PCs into accessories and Microsoft from a monopolist into just another big tech company. That’s not an argument against regulation, but it may be an argument for humility.

Regulating technology


This post is by Benedict Evans from Essays - Benedict Evans

Technology was a small industry until very recently. It was exciting and interesting, and it was on lots of magazine covers, but it wasn‘t actually an important part of most people’s lives. When Bill Gates was on every magazine cover, Microsoft was a small company that sold accounting tools to big companies. When Netscape kicked off the consumer internet in 1994, there were only 100m or so PCs on earth, and most of them were in offices. Today 4bn people have a smartphone – three quarters of all the adults on earth. In most developed countries, 90% of the adult population is online. 

2020 07 New Normal Google.053.png

The change isn’t just that almost all of us have a computer now, but that we’ve changed how we use them.  This is my favourite chart to show this – in 2017, 40% of new couples in the USA met online. It’s probably over 50% now. Anyone does anything online now. 

2020 Shoulders of Giants 1.1.053.png

Tech has gone from being just one of many industries to being systemically important to society. My old colleague Marc Andreessen liked to say that ‘software is eating the world’ – well, it did.

The trouble is, when tech becomes the world, all of tech’s problems matter much more, because they become so much bigger and touch so many more people; and in parallel all of the problems that society had already are expressed in this new thing, and are amplified and changed by it, and channeled in new ways. When you connect all of society, you connect all of society’s problems as well. You connect all the bad people, and more importantly you connect all of our own worst instincts. And then, of course, all of these combine and feed off each other, and generate new externalities. The internet had hate speech in 1990, but it didn’t affect elections, and it didn’t involve foreign intelligence agencies.

When something is systemically important to society and has systemically important problems, this brings attention from governments and regulators. At a very high level, one could say that all industries are subject to general legislation, and some also have industry-specific legislation. All companies have to follow employment law, and accounting law, and workplace safety law, and indeed criminal law. But some also have their own laws as well, because they have some very specific and important questions that need them. This chart is an attempt to capture some of this industry-specific law. Banks, airlines and oil refineries are regulated industries, and technology is going to become a regulated industry as well. 

2020 Shoulders of Giants 1.1.088.png

Part of the point of this chart is that regulation isn’t simple, and that it can’t be. Each of these industries has lots of different issues, in different places, with different people in positions to do something and different kinds of solution. We regulate ‘banks’, but that’s not one thing – we regulate credit cards, mortgages, futures & options and the money supply and these are different kinds of question with different kinds of solutions. De La Rue is good at making banknotes that are hard to forge, but we don’t ask it what affordability tests should be applied to mortgages. 

To take this point further, cars brought many different kinds of problem, and we understood that responsibility for doing something about them sat in different places, and that solutions are probably limited and probably have tradeoffs. We can tell car companies to make their cars safer, and punish them if they cut corners, but we can’t tell them to end all accidents or make gasoline that doesn’t burn. We can tell Ford to reduce emissions, but we can’t tell it to fix parking or congestion, or build more bike lanes – someone else has to do that, and we might or might not decide to pay for it out of taxes on cars. We worry that criminals use cars, but that’s a social problem and a law enforcement problem, not a mechanical engineering problem. And we mandate speed limits, but we don’t build them into cars. This is how policy works: there are many largely unrelated problems captured by words like ‘cars’ or ‘banking’ or ‘tech’, some things are impossible, most things are tradeoffs, there are generally unintended consequences, and everything is complicated. 

‘Tech’, of course, has all of this complexity, but we’re having to work this out a lot more quickly. It took 75 years for seatbelts to become compulsory, but tech has gone from interesting to crucial only in the last five to ten years. That speed means we have to form opinions about things we didn’t grow up with and don’t always understand quite so well as, say, supermarkets. 

In addition, in the US and UK I’d suggest the triggers for the change in awareness of just how much tech suddenly mattered were the election of Donald Trump and the Brexit referendum, both in 2016, and the roles that social media may or may not have played in those. This has polarised and intensified some of these conversations around tech by linking them to much broader polarising themes, and sometimes to a tendency to displacement – it can be tempting to blame an external force rather than ask yourself why your fellow-citizens didn’t vote the right way. 

All of this means that the move towards regulation has sometimes been accompanied by a moral panic, and a rush for easy answers. That’s a lot of the appeal of a phrase like ‘break them up!’ – it has a comforting simplicity, but doesn’t really give us a route to solutions. Indeed, ‘break them up’ reminds me a lot of ‘Brexit’ – it sounds simple until you ask questions. Break them up into what, and what problems would that solve? The idea that you can solve the social issues connected to the internet with anti-trust intervention is rather like thinking that you can solve the social issues that come from cars by breaking up GM and Ford. Competition tends to produce better cars, but we didn’t rely on it to reduce emissions or improve safety, and breaking up GM wouldn’t solve traffic congestion. Equally, changing Instagram’s ownership would give advertisers more leverage, but it’s not a path to stopping teenaged girls from seeing self-harm content. Not everything is captured by the pricing system, which is why economics text books talk about ‘market failure’. And if you ask the average person on the street why they worry about ‘big tech’, they’re unlikely to talk about their concern that Facebook and Google might be overcharging Unilever for video prerolls. 

Part of the appeal of applying anti-trust to any problem connected to ‘tech’ is that it sounds simple – it’s a way to avoid engaging with the complexity of real policy – but it’s also worth noting that the rise of tech to systemic importance has coincided with the second half of an industry cycle. Smartphones and social have matured and the leading companies in those industries mostly have dominant market shares, just as IBM did in mainframes in the late 1970s or Microsoft in PC software in the late 1990s (and because tech itself is so large, and global, being a leading company in tech makes you very big). This lends itself to a post hoc ergo propter hoc fallacy: these companies have gained big market shares at the same time as the problems emerged, so that must be the cause of the problem. 

The more one thinks about real policy, though, the more one sees that many of the most important debates we have around technology have deeply embedded tradeoffs. At the beginning of this year I attended a conference of European competition regulators: the head of one agency said that they tell a tech company that it must do X, and then the company goes down the road to the privacy regulator, who says ‘you must not under any circumstances do X’. Competition policy says ’remove friction in moving data between competing services’ and privacy policy says, well, ‘add friction’. In other words, policy objectives come with conflicts. We are probably about to get into another big argument about how Apple controls what you do on an iPhone, and there’s a Venn Digram to be drawn here: there are Apple policies that protect the user’s privacy and security, policies that protect Apple’s competitive position (or just make it money), policies that do both, and policies that really just reflect Apple’s preferences for the kind of apps it would like to see. How exactly do these intersect? You might not want to let privacy regulators or competition regulators have the only word on this. 

Of course, this is how policy works – you have to pick tradeoffs. You can have cheaper food or more sustainable food supply chains; you can make home-owning a wealth-building asset class or you can have cheaper housing. As voters, of course, we want both – I want my parents’ home to appreciate and the home I plan to buy to get cheaper. A UK minister recently told me that his constituents complain about two aspects of government data collection: the government knows too much about them, and also they have to enter the same information into too many different government websites. 

This is how policy works, but in the past these were all national debates. The UK, France and USA have very different models of libel law, but that wasn’t a big problem because no-one was really publishing a newspaper in all three countries. But network effects are global: any software platform of any scale grows globally, so how does it follow local law? For its first 25 years, the consumer internet has operated by default on US ideas of free speech, regulation, privacy law (or the absence of it) and competition, partly because that’s where most of the internet came from and partly because the internet wasn’t really important enough for clashes of these cultures to move from irritation to legislation. That’s clearly changed, and we’re moving to a world of multiple, overlapping regulatory spheres. 

2020 Shoulders of Giants 1.1.107.png

These vary in several ways. Some of this is undoubtedly nationalism and protectionism. Some rests on concerns about national security, either a fear of spying or of media tools being used to promote particular narratives by unfriendly states. But the core of it, I’d suggest, is the rather basic Westfalian principle that a country’s government has the right to say what can happen in that country. This isn’t just about ‘China’ versus ‘the west’ – different liberal democracies have different views on how free speech works, for example, and no-one outside the USA cares or even knows what the US constitution says about it. Equally, each sphere will find its own approach to the liability questions that America groups under ‘section 230’, and again, no-one will care what solution the US comes with. The same variance applies to privacy, competition itself and whole bunch of other issues, right down to very micro issues like whether an Uber driver is legally an employee or Airbnb’s impact on house prices.  

2020 Shoulders of Giants 1.1.099.png

Indeed, even the basic mechanisms of regulation themselves can look very different in different places. To simplify hugely, the US has a rules-based, lawyer-led system in which the basic unit of regulation is generally a court case, with a guilty or innocent verdict, and a punishment. Conversely, both the UK and EU have outcome-based, practitioner-led systems that focus on principles and ‘reasonableness tests’ rather than detailed rules and may never go to a court – this can look very alien to US lawyers, and vice versa.

We are clearly going to have multiple regulatory spheres. GDPR also made it clear that those rules would increasingly apply no matter where your servers are: if your users are in the EU, you have to obey EU rules, and for practical reasons that probably means you have to obey them for all of your users. CCPA effectively does the same in the USA, where California has increasingly become the national privacy regulator by default. An intriguing further step came from this case, in which an EU court held that Facebook must take down libellous content not just in Austria, where the case began, but globally. Meanwhile, the new Hong Kong security law appears to apply to behaviour by non-HK residents outside HK, which is truly extra-territorial. The obvious next question is what happens when an extraterritorial rule collides with a trade-off. What happens when the UK says you must do X and Germany says you must not? 

So, we have divergence in regulatory policy. We also have divergence in where the companies themselves are coming from. In this, Tiktok is the symptom of a broader change, or perhaps a catalyst for realising it. For most of the last 25 years the internet was American by default partly because that was where the companies came from, and that may be changing. In 2008 the US was 80% of global VC investing; now it’s 50%.

2020 07 New Normal Google.059.png

Silicon Valley is still the global cluster, but it’s no longer the only place you can make great products – software is diffusing. So what happens if now lots of people love a Chinese product? (For a fun contrast, note that the Russians tried to subvert US social networks but the Chinese built their own.) Tiktok claims 800m MAUs outside China. The US talks about banning it, but you can’t ban every cool new app, and yet we need to be conscious that any Chinese company (or company with people in China) can be told to do anything by the Chinese state, and they don’t have a choice.  What are the scalable, repeatable rules?

All of this really just scratches the surface of the complexity we might see, as companies used to thinking in terms of inherently borderless platforms collide with five or ten or fifty different regulators and governments around the world. Complexity itself is an important consequence: this chart attempts to capture the cost of regulation for different US retail banks by size: the smaller the bank, the higher the proportionate cost of compliance. This isn’t an original observation: regulation is a regressive tax that tends to slow down innovation and impede startups.

2020 Shoulders of Giants 1.1.117.png

This, of course, is another tradeoff. Some time between 1850 and 1900 or so the industrial world worked out that regulating industry is necessary, and since then we’ve been arguing about how and how much, industry by industry, from industrial food to banking to airlines. Now that gets applied to tech. That’s not a great book title, but it’s probably the next decade or two. I started my career, in 1999, at an investment bank in London that had 900 people and one (1) compliance person. Today it might have 100x that. So how many compliance people will Google have in five years?

Note: many of these slides come from a presentation I gave in Davos at the beginning of 2020: ‘Standing on the Shoulders of Giants’.

Regulating technology


This post is by Benedict Evans from Essays - Benedict Evans

Technology was a small industry until very recently. It was exciting and interesting, and it was on lots of magazine covers, but it wasn‘t actually an important part of most people’s lives. When Bill Gates was on every magazine cover, Microsoft was a small company that sold accounting tools to big companies. When Netscape kicked off the consumer internet in 1994, there were only 100m or so PCs on earth, and most of them were in offices. Today 4bn people have a smartphone – three quarters of all the adults on earth. In most developed countries, 90% of the adult population is online. 

2020 07 New Normal Google.053.png

The change isn’t just that almost all of us have a computer now, but that we’ve changed how we use them.  This is my favourite chart to show this – in 2017, 40% of new couples in the USA met online. It’s probably over 50% now. Anyone does anything online now. 

2020 Shoulders of Giants 1.1.001.png

Tech has gone from being just one of many industries to being systemically important to society. My old colleague Marc Andreessen liked to say that ‘software is eating the world’ – well, it did.

The trouble is, when software becomes part of society, all of society’s problems get expressed in software. We connected everyone, so we connected the bad people, and more importantly we connected all of our own worst instincts. All the things we worried about before now happen online, and are amplified, changed and channeled in new ways. Meanwhile, the problems that tech always had matter much more, because they become so much bigger and touch so many more people. And then, of course, all of these combine and feed off each other, and generate new externalities. The internet had hate speech in 1990, but it didn’t affect elections, and it didn’t involve foreign intelligence agencies.

When something is systemically important to society and has systemically important problems, this brings attention from governments and regulators. All industries are subject to general legislation, but some also have industry-specific legislation. All companies have to follow employment law, and accounting law, and workplace safety law, and indeed criminal law. But some also have their own laws as well, because they have some very specific and important questions that need them. This chart is an attempt to capture some of this industry-specific law. Banks, airlines and oil refineries are regulated industries, and technology is going to become a regulated industry as well. 

2020 Shoulders of Giants 1.1.088.png

Part of the point of this chart is that regulation isn’t simple, and that it can’t be. Each of these industries has lots of different issues, in different places, with different people in positions to do something and different kinds of solution. We regulate ‘banks’, but that’s not one thing – we regulate credit cards, mortgages, futures & options and the money supply and these are different kinds of question with different kinds of solutions. De La Rue is good at making banknotes that are hard to forge, but we don’t ask it what affordability tests should be applied to mortgages. 

To take this point further, cars brought many different kinds of problem, and we understood that responsibility for doing something about them sat in different places, and that solutions are probably limited and probably have tradeoffs. We can tell car companies to make their cars safer, and punish them if they cut corners, but we can’t tell them to end all accidents or make gasoline that doesn’t burn. We can tell Ford to reduce emissions, but we can’t tell it to fix parking or congestion, or build more bike lanes – someone else has to do that, and we might or might not decide to pay for it out of taxes on cars. We worry that criminals use cars, but that’s a social problem and a law enforcement problem, not a mechanical engineering problem. And we mandate speed limits, but we don’t build them into cars. This is how policy works: there are many largely unrelated problems captured by words like ‘cars’ or ‘banking’ or ‘tech’, some things are impossible, most things are tradeoffs, there are generally unintended consequences, and everything is complicated. 

‘Tech’, of course, has all of this complexity, but we’re having to work this out a lot more quickly. It took 75 years for seatbelts to become compulsory, but tech has gone from interesting to crucial only in the last five to ten years. That speed means we have to form opinions about things we didn’t grow up with and don’t always understand quite so well as, say, supermarkets. 

In addition, in the US and UK I’d suggest the triggers for the change in awareness of just how much tech suddenly mattered were the election of Donald Trump and the Brexit referendum, both in 2016, and the roles that social media may or may not have played in those. This has polarised and intensified some of these conversations around tech by linking them to much broader polarising themes, and sometimes to a tendency to displacement – it can be tempting to blame an external force rather than ask yourself why your fellow-citizens didn’t vote the right way. 

All of this means that the move towards regulation has sometimes been accompanied by a moral panic, and a rush for easy answers. That’s a lot of the appeal of a phrase like ‘break them up!’ – it has a comforting simplicity, but doesn’t really give us a route to solutions. Indeed, ‘break them up’ reminds me a lot of ‘Brexit’ – it sounds simple until you ask questions. Break them up into what, and what problems would that solve? The idea that you can solve the social issues connected to the internet with anti-trust intervention is rather like thinking that you can solve the social issues that come from cars by breaking up GM and Ford. Competition tends to produce better cars, but we didn’t rely on it to reduce emissions or improve safety, and breaking up GM wouldn’t solve traffic congestion. Equally, changing Instagram’s ownership would give advertisers more leverage, but it’s not a path to stopping teenaged girls from seeing self-harm content. Not everything is captured by the pricing system, which is why economics text books talk about ‘market failure’. And if you ask the average person on the street why they worry about ‘big tech’, they’re unlikely to reply that Facebook and Google might be overcharging Unilever for video prerolls. 

Part of the appeal of applying anti-trust to any problem connected to ‘tech’ is that it sounds simple – it’s a way to avoid engaging with the complexity of real policy – but it’s also worth noting that the rise of tech to systemic importance has coincided with the second half of an industry cycle. Smartphones and social have matured and the leading companies in those industries mostly have dominant market shares, just as IBM did in mainframes in the late 1970s or Microsoft in PC software in the late 1990s (and because tech itself is so large, and global, being a leading company in tech makes you very big). This lends itself to a post hoc ergo propter hoc fallacy: these companies have gained big market shares at the same time as the problems emerged, so that must be the cause of the problem. 

The more one thinks about real policy, though, the more one sees that many of the most important debates we have around technology have deeply embedded tradeoffs. At the beginning of this year I attended a conference of European competition regulators: the head of one agency said that they tell a tech company that it must do X, and then the company goes down the road to the privacy regulator, who says ‘you must not under any circumstances do X’. Competition policy says ’remove friction in moving data between competing services’ and privacy policy says, well, ‘add friction’. In other words, policy objectives come with conflicts. We are probably about to get into another big argument about how Apple controls what you do on an iPhone, and there’s a Venn Digram to be drawn here: there are Apple policies that protect the user’s privacy and security, policies that protect Apple’s competitive position (or just make it money), policies that do both, and policies that really just reflect Apple’s preferences for the kind of apps it would like to see. How exactly do these intersect? You might not want to let privacy regulators or competition regulators have the only word on this. 

Of course, this is how policy works – you have to pick tradeoffs. You can have cheaper food or more sustainable food supply chains; you can make home-owning a wealth-building asset class or you can have cheaper housing. As voters, of course, we want both – I want my parents’ home to appreciate and the home I plan to buy to get cheaper. A UK minister recently told me that his constituents complain about two aspects of government data collection: the government knows too much about them, and also they have to enter the same information into too many different government websites. 

This is how policy works, but in the past these were all national debates. The UK, France and USA have very different models of libel law, but that wasn’t a big problem because no-one was really publishing a newspaper in all three countries. But network effects are global: any software platform of any scale grows globally, so how does it follow local law? For its first 25 years, the consumer internet has operated by default on US ideas of free speech, regulation, privacy law (or the absence of it) and competition, partly because that’s where most of the internet came from and partly because the internet wasn’t really important enough for clashes of these cultures to move from irritation to legislation. That’s clearly changed, and we’re moving to a world of multiple, overlapping regulatory spheres. 

2020 Shoulders of Giants 1.1.107.png

Some of this is undoubtedly nationalism and protectionism. Some rests on concerns about national security, either a fear of spying or of media tools being used to promote particular narratives by unfriendly states. But the core of it, I’d suggest, is the rather basic Westphalian principle that a country’s government has the right to say what can happen in that country. This isn’t just about ‘China’ versus ‘the west’ – different liberal democracies have different views on how free speech works, for example, and no-one outside the USA cares or even knows what the US constitution says about it. Equally, each sphere will find its own approach to the liability questions that America groups under ‘section 230’, and again, no-one will care what solution the US comes with. The same variance applies to privacy, competition itself and whole bunch of other issues, right down to very micro issues like whether an Uber driver is legally an employee or Airbnb’s impact on house prices.  

2020 Shoulders of Giants 1.1.099.png

Indeed, even the basic mechanisms of regulation themselves can look very different in different places. To simplify hugely, the US has a rules-based, lawyer-led system in which the basic unit of regulation is generally a court case, with a guilty or innocent verdict, and a punishment. Conversely, both the UK and EU have outcome-based, practitioner-led systems that focus on principles and ‘reasonableness tests’ rather than detailed rules and may never go to a court – this can look very alien to US lawyers, and vice versa.

These regulatory spheres are probably going to start bumping into each other. GDPR made it clear that rules would increasingly apply no matter where your servers are: if your users are in the EU, you have to obey EU rules, and for practical reasons that probably means you have to obey them for all of your users. CCPA effectively does the same in the USA, where California has increasingly become the national privacy regulator by default. An intriguing further step came from this case, in which an EU court held that Facebook must take down libellous content not just in Austria, where the case began, but globally. Meanwhile, the new Hong Kong security law appears to apply to behaviour by non-HK residents outside HK, which is truly extra-territorial. The obvious next question is what happens when an extraterritorial rule collides with a trade-off. What happens when the UK says you must do something and Germany says you must not? 

So, we have divergence in regulatory policy. We also have divergence in where the companies themselves are coming from. In this, Tiktok is the symptom of a broader change, or perhaps a catalyst for realising it. For most of the last 25 years the internet was American by default partly because that was where the companies came from, and that may be changing. In 2008 the US was 80% of global VC investing; now it’s 50%.

2020 07 New Normal Google.059.png

Silicon Valley is still the global cluster, but it’s no longer the only place you can make great products – software is diffusing. So what happens if now lots of people love a Chinese product? (For a fun contrast, note that the Russians tried to subvert US social networks but the Chinese built their own.) Tiktok claims 800m MAUs outside China. The US talks about banning it, but you can’t ban every cool new app, and yet we need to be conscious that any Chinese company (or company with people in China) can be told to do anything by the Chinese state, and they don’t have a choice. So, what are the scalable, repeatable rules?

This really just scratches the surface of the complexity we might see, as companies used to thinking in terms of inherently borderless platforms collide with five or ten or fifty different regulators and governments around the world. Complexity itself is an important consequence: this chart attempts to capture the cost of regulation for different US retail banks by size: the smaller the bank, the higher the proportionate cost of compliance. This isn’t an original observation: regulation is a regressive tax that tends to slow down innovation and impede startups.

2020 Shoulders of Giants 1.1.117.png

Regulation, of course, is another tradeoff. Some time between 1850 and 1900 or so the industrial world worked out that regulating industry is necessary, and since then we’ve been arguing about how and how much, industry by industry, from industrial food to banking to airlines. Now that gets applied to tech. That’s not a great book title, but it’s probably the next decade or two. I started my career, in 1999, at an investment bank in London that had 900 people and one (1) compliance person. Today it might have 100x that. So, how many compliance people will Google have in five years?

Note: many of these slides come from a presentation I gave in Davos at the beginning of 2020: ‘Standing on the Shoulders of Giants’.

What comes after Zoom?


This post is by Benedict Evans from Essays - Benedict Evans

VIdeo conferencing in the 1990s

We had video calls in science fiction, and we had video conferencing in the 1990s, just as the web was taking off, as a very expensive and impractical tool for big companies. It was proposed as a use case for 3G, which didn’t happen at all, and with the growth of consumer broadband we got all sorts of tools that could do it, but it never really became a mass-market consumer behaviour. Now, suddenly, we’re all locked down, and we’re all on video calls all the time, doing team stand-ups, play dates and family birthday parties, and suddenly Zoom is a big deal. At some point many of those meetings will turn back into coffees, we hope, but video will remain.

Will it still be Zoom, though?

As a breakthrough product, I think it’s useful to compare Zoom with two previous products – Dropbox and Skype. 

Part of the founding legend of Dropbox is that Drew Houston told people what he wanted to do, and everyone said ‘there are hundreds of these already’ and he replied ‘yes, but which one do you use?’ That’s what Zoom did – video calls are nothing new, but Zoom solved a lot of the small pieces of friction that made it fiddly to get into a call.  

The other comparison, though, is Skype. Just as for video, VOIP had been around for a long time, but Skype solved a lot of pieces of friction, in both engineering and user experience, and by doing so made VOIP a consumer product. 

Two things happened to Skype after that, though. The first is that the product drifted for a long time, and the quality of the user experience declined. But the second is that everything has voice now. Imagine trying to do a market map today of which apps on a smartphone, Mac or PC might have voice – it would be absurd. Everything can have voice. And though there’s still a lot of engineering under the hood, it became a commodity. Whether you buy it from Twilio or someone else, saying ‘our app has free computer voice’ is meaningless – what matters is how you wrap it. Why do you have voice? Hence, Clubhouse is built on ideas about psychology and behaviour, not VOIP, and it’s not trying to win ‘voice’.  Equally, Pindrop looks at every call going into a call centre and tries to work out which might be fraudulent. If you’d looked at Skype in 2004 and argued that it would own ‘voice’ on ‘computers’, that would not have been the right mental model. 

I think this is where we’ll go with video – there will continue to be hard engineering, but video itself will be a commodity and the question will be how you wrap it. There will be video in everything, just as there is voice in everything, and there will be a great deal of proliferation into industry verticals on one hand and into unbundling pieces of the tech stack on the other. On one hand video in healthcare, education or insurance is about the workflow, the data model and the route to market, and lots more interesting companies will be created, and on the other hand Slack is deploying video on top of Amazon’s building blocks, and lots of interesting companies will be created here as well. There’s lots of bundling and unbundling coming, as always. Everything will be ‘video’ and then it will disappear inside.

An important part of this is that there seem to be few real network effects in a video call per se. You don’t necessarily need an account to join a call, and you generally don’t need an application either, especially on the desktop – you just click on a link in your calendar and the call opens in the browser. Indeed, the calendar is often the aggregation layer – you don’t need to know what service the next call uses, just when it is. Skype needed both an account and an app, so had a network effect (and lost even so). WhatsApp uses the telephone numbering system as an address and so piggybacked on your phone’s contact list- effectively it used the PSTN as the social graph rather than having to build its own. But a group video call is a URL and a calendar invitation – it has no graph of its own.

Incidentally, one of the ways that this all feels very 1.0 is the rather artificial distinction between calls that are based on a ‘room’, where the addressing system is a URL and anyone can join without an account, and calls that are based on ‘people’, where everyone joining needs their own address, whether it’s a phone number, an account or something else. Hence Google has both Meet (URLs) and Duo (people) – Apple’s FaceTime is only people (no URLs).

Taking this one step further, a big part of the friction that Zoom removed was that you don’t need an account, an app or a social graph to use it: Zoom made network effects irrelevant. But, that means Zoom doesn’t have those network effects either. It grew by removing defensibility.

I compared Zoom with Dropbox and Skype, but another useful comparison is with photo sharing. There have always been hundreds of things that did this, but we saw a succession of companies that worked out something new around user experience and psychology that took them beyond ‘photos’ to some deeper insight – first Flickr, then Facebook and Instagram, and then Snap. 

When Snap launched, there were infinite way to share images, but Snap asked a bunch of weird questions that no-one had really asked before. Why do you have to press the camera button – why doesn’t the app open in the camera? Why are you saving your messages – isn’t that like saving all your phone calls? Fundamentally, Snap asked ‘why, exactly, are you sending a picture? What is the underlying social purpose?’ You’re not really sending someone a sheet of pixels – you’re communicating. 

That’s the question Zoom and all its competitors haven’t really asked. Zoom has done a good job of asking why it was hard to get into a call, but hasn’t really asked why you’re in the call in the first place. Why, exactly, are you sending someone a video stream and watching another one? Why am I looking at a grid of little thumbnails of faces? Is that the purpose of this moment? What is the ‘mute’ button for – background noise, or so I can talk to someone else, or is it so I can turn it off to raise my hand? What social purpose is ‘mute’ actually serving? What is screen-sharing for? What other questions could one ask? And so if Zoom is the Dropbox or Skype of video, we are waiting for the Snap, Clubhouse and Yo. 

What comes after Zoom?


This post is by Benedict Evans from Essays - Benedict Evans

VIdeo conferencing in the 1990s

We had video calls in science fiction, and we had video conferencing in the 1990s, just as the web was taking off, as a very expensive and impractical tool for big companies. It was proposed as a use case for 3G, which didn’t happen at all, and with the growth of consumer broadband we got all sorts of tools that could do it, but it never really became a mass-market consumer behaviour. Now, suddenly, we’re all locked down, and we’re all on video calls all the time, doing team stand-ups, play dates and family birthday parties, and suddenly Zoom is a big deal. At some point many of those meetings will turn back into coffees, we hope, but video will remain.

Will it still be Zoom, though?

As a breakthrough product, I think it’s useful to compare Zoom with two previous products – Dropbox and Skype. 

Part of the founding legend of Dropbox is that Drew Houston told people what he wanted to do, and everyone said ‘there are hundreds of these already’ and he replied ‘yes, but which one do you use?’ That’s what Zoom did – video calls are nothing new, but Zoom solved a lot of the small pieces of friction that made it fiddly to get into a call.  

The other comparison, though, is Skype. Just as for video, VOIP had been around for a long time, but Skype solved a lot of pieces of friction, in both engineering and user experience, and by doing so made VOIP a consumer product. 

Two things happened to Skype after that, though. The first is that the product drifted for a long time, and the quality of the user experience declined. But the second is that everything has voice now. Imagine trying to do a market map today of which apps on a smartphone, Mac or PC might have voice – it would be absurd. Everything can have voice. And though there’s still a lot of engineering under the hood, it became a commodity. Whether you buy it from Twilio or someone else, saying ‘our app has free computer voice’ is meaningless – what matters is how you wrap it. Why do you have voice? Hence, Clubhouse is built on ideas about psychology and behaviour, not VOIP, and it’s not trying to win ‘voice’.  Pindrop also does voice, but it’s enterprise software that looks at every call going into a call centre and tries to work out which might be fraudulent. If you’d looked at Skype in 2004 and argued that it would own ‘voice’ on ‘computers’, that would not have been the right mental model. 

I think this is where we’ll go with video – there will continue to be hard engineering, but video itself will be a commodity and the question will be how you wrap it. There will be video in everything, just as there is voice in everything, and there will be a great deal of proliferation into industry verticals on one hand and into unbundling pieces of the tech stack on the other. On one hand video in healthcare, education or insurance is about the workflow, the data model and the route to market, and lots more interesting companies will be created, and on the other hand Slack is deploying video on top of Amazon’s building blocks, and lots of interesting companies will be created here as well. There’s lots of bundling and unbundling coming, as always. Everything will be ‘video’ and then it will disappear inside.

An important part of this is that there seem to be few real network effects in a video call per se. You don’t necessarily need an account to join a call, and you generally don’t need an application either, especially on the desktop – you just click on a link in your calendar and the call opens in the browser. Indeed, the calendar is often the aggregation layer – you don’t need to know what service the next call uses, just when it is. Skype needed both an account and an app, so had a network effect (and lost even so). WhatsApp uses the telephone numbering system as an address and so piggybacked on your phone’s contact list- effectively it used the PSTN as the social graph rather than having to build its own. But a group video call is a URL and a calendar invitation – it has no graph of its own.

Incidentally, one of the ways that this all feels very 1.0 is the rather artificial distinction between calls that are based on a ‘room’, where the addressing system is a URL and anyone can join without an account, and calls that are based on ‘people’, where everyone joining needs their own address, whether it’s a phone number, an account or something else. Hence Google has both Meet (URLs) and Duo (people) – Apple’s FaceTime is only people (no URLs).

Taking this one step further, a big part of the friction that Zoom removed was that you don’t need an account, an app or a social graph to use it: Zoom made network effects irrelevant. But, that means Zoom doesn’t have those network effects either. It grew by removing defensibility.

I compared Zoom with Dropbox and Skype, but another useful comparison is with photo sharing. There have always been hundreds of things that did this, but we saw a succession of companies that worked out something new around user experience and psychology that took them beyond ‘photos’ to some deeper insight – first Flickr, then Facebook and Instagram, and then Snap. 

When Snap launched, there were infinite way to share images, but Snap asked a bunch of weird questions that no-one had really asked before. Why do you have to press the camera button – why doesn’t the app open in the camera? Why are you saving your messages – isn’t that like saving all your phone calls? Fundamentally, Snap asked ‘why, exactly, are you sending a picture? What is the underlying social purpose?’ You’re not really sending someone a sheet of pixels – you’re communicating. 

That’s the question Zoom and all its competitors haven’t really asked. Zoom has done a good job of asking why it was hard to get into a call, but it hasn’t asked why you’re in the call in the first place. Why, exactly, are you sending someone a video stream and watching another one? Why am I looking at a grid of little thumbnails of faces? Is that the purpose of this moment? What is the ‘mute’ button for – background noise, or so I can talk to someone else, or is it so I can turn it off to raise my hand? What social purpose is ‘mute’ actually serving? What is screen-sharing for? What other questions could one ask? And so if Zoom is the Dropbox or Skype of video, we are waiting for the Snap, Clubhouse and Yo. 

News by the ton: 75 years of US advertising


This post is by Benedict Evans from Essays - Benedict Evans

There are two ways you can talk about newspapers. You can talk about the ‘third estate’, and newspapers’ role in culture, politics, governance, the exchange of ideas and civil society. But you can also talk about newspapers as a specialised light manufacturing industry, that aggregated attention to sell advertising. There’s a common line about Google and Facebook that ‘if you’re not paying, you’re the product’, but this is pretty much what newspapers did: if you read old accounts for, say the New York Times Company, you can see that they were giving the product away at close to cost and making the money from selling your attention.

Misc slides.040.jpeg

(‘Wages’ in this chart includes both editorial staff and production & delivery staff – the New York Times was delivering 45% of its circulation, and the Boston Globe 70%, and charging a premium for this.)

I picked 1994 because that’s the year Netscape launched and kicked off the consumer internet, and as we all know, that has been somewhat material for newspapers’ print ad business.

Misc slides.023.jpeg

I wrote recently about the way the numbers in this chart didn’t really change until the financial crisis in 2008, almost 15 years after the consumer internet began, but it’s also worth looking at how those numbers have changed over a longer period.

Misc slides.022.jpeg

There are two interesting stories here – the collapse in the last two decades, but also the earlier growth.

First, the collapse. Let’s add the internet to the chart, and let’s show all the rest of mass media advertising as well.

Misc slides.028.jpeg

The problem with this chart is that long-term inflation calculations get increasingly meaningless the further back you take them, and since historically advertising has been closely indexed to the overall economy, the growth in the first part of this chart is telling us much more about the growth of the US economy than it is about advertising. So, let’s take it as a share of GDP. At this point I’ll start the time series at 1950, because WW2 adds a big distraction, and let’s try changing the series order.

Misc slides.035.jpeg

There are lots of things going on here, but I would start with the top line: advertising share of GDP started sliding immediately after the Dotcom bubble, had a major step down in the financial crisis and has been suspiciously flat ever since. That decline was very obviously concentrated in print but actually affected TV and radio as well. We think of TV advertising as being pretty much unaffected by the internet so far, but on this data it’s down by 40% as a share of GDP. The economy grew and advertising didn’t get its historic share of that growth.

Misc slides.001.jpeg

It’s very common for people – especially newspaper people – to look at the newspaper and internet series in these charts and conclude that all the money went from newspapers to internet. There’s also a tendency to try to calculate Google and Facebook’s share of that ‘internet’ line. This can get you onto shaky ground quite quickly. As that change in share of GDP (and my phase ‘suspiciously flat’) should suggest, what’s actually happened is that the market has been both reallocated and repriced, a lot of money left the data that’s being captured here, and a lot of other money came in.

So: if you talk to people at both Google and Facebook and in the agency world, you’ll hear that perhaps two thirds to three quarters of money spent on Google and Facebook is money that was never spent on traditional advertising – it’s coming from SMEs and local businesses that might have spent in classified at most but probably wouldn’t have done even that. $60bn of consumer spending went through Shopify last year – it’s safe to assume those vendors spent money on advertising, but how many of them would have bought an ad in a local newspaper? This has also come at much lower prices: Facebook in particular has been massively deflationary to online advertising: it offers vast quantities of relevant advertising inventory at much lower prices and much lower entry costs than you’d have needed in print, let alone TV.

Meanwhile, ‘advertising’ is only a subset of all the money being spent to ‘get people to buy things’, and the data captured in traditional metrics is only another subset, and money moves between those categories. This graphic made a couple of years ago (by an old colleague – Hi Omar!) captures some of that complexity. Some parts of this are called ‘advertising’ and others ‘marketing’, but that’s not always the best way to look at them.

EaPQBhbX0AEDuWu.jpeg

If an estate agent shifts their budget from an insert in a local newspaper to their own website and to fees for Zillow or Rightmove, where does that appear in charts of ad spend? What if they close a branch and put the rent budget into Rightmove? Booking.com and Expedia between them spent $10bn on Google ads last year: did that money ‘come from’ newspaper advertising? Or from retail rents paid by travel agents? Wix.com now has $740m of ARR: that’s mostly doing the job of advertising and marketing, so did that money come from newspapers and magazines? Or from rent? Is that even a useful question? (And meanwhile Wix itself is often a top-ten advertiser itself in Europe.) I pay Squarespace and Mailchimp a fee for this site and for my weekly newsletter: 20 years ago I might have bought a listing in a professional directory, but is that really comparable? If Procter & Gable pays a supermarket for placement in the store, that’s categorised as ‘marketing’, but if it pays Amazon for placement in search results that’s categorised as ‘advertising’, and Amazon sold over $10bn of that placement last year. So, what does a chart of ‘internet advertising’ really capture, and what is it missing, and what share, exactly, do Google and Facebook have? Are you sure?

One of the most interesting charts I’ve seen on this theme comes from Google Trends. Like all Google Trends data it needs to be taken with a large pinch of salt, but this one is pretty instructive. I’d suggest this shows the internet moving up the ‘funnel’ – it moves from utility price comparison to recommendation and authority. That might be a rather more profound conflict with traditional forms of media that those raw ad sales.

Misc slides.091.jpeg

Finally, so far I’ve focused on the impact of the internet on revenue, but it’s worth digging a little into the some more second-order metrics for newspapers. As shown implicitly in some of the previous charts, but worth making explicit here, newspaper ad revenue may only have collapsed since 2008, but newspapers have been losing share of ad spend since the 1950s.

Misc slides.026.jpeg

That’s not all they were losing. Absolute circulation rose until 1990, and has been falling since (‘gradually, then suddenly’), but circulation per capita has been falling since 1950, in common with most other developed markets.

Misc slides.044.jpeg

Indexing these to 1950 makes the trend clearer.

Misc slides.045.jpeg

If we now add both revenue and newsprint consumption to this chart, we can see that newspapers combined stagnant-to-falling circulation with a rising page count and hence rising ad inventory and ad revenue. Prices went up as well, but (on this data) the increase in inventory was the key change.

Misc slides.047.jpeg

In other words, if your users are flat, increasing the ad load can lead to higher revenue (this will sound familiar to many social network users). Of course, more ad pages needed more content pages to pad them out (the ratio was typically 65:35), and that meant more people to fill them. This chart suggests what one often hears anecdotally: that over time newspapers expected each person to produce more and more column inches.

Misc slides.048.jpeg

This is, of course, the cynical way to look at the numbers. You could equally well say that newspapers’ audiences shrank and moved upmarket, leading them to print more coverage, aimed at a different reader, and that that in turn supported more and higher value advertising. The result is the same either way: more journalists, more pages, more ads, and fewer readers, first per capita and then absolutely.

Conversely, of course, rapidly declining circulation, fewer pages per newspaper and less revenue has meant fewer journalists. Indeed, there are now far fewer people working for American newspaper publishers than at any point since at WW2, and probably rather longer.

Misc slides.044.jpeg

None of these charts, of course, show newspaper companies’ internet advertising revenue, let alone the subscription businesses that some of them have managed to build. On one level that’s because the employment numbers above speak pretty clearly about how well that’s gone for most of them (and I haven’t talked at all about the polarisation of success between large and small brands). But it’s also because newspapers were an oligopoly, and they lose that oligopoly online. Newspapers are, yes, a content business, but they were also a light manufacturing business, and it was the replacement of light manufacturing and trucking with bits that removed the barrier to entry and unbundled their attention. So, here’s what that business looks like. In 2018, the US newspaper industry shipped ~2.5m tons of newspaper, down from 12.5m tons at its peak.

Misc slides.055.jpeg