In the 1980s, if you installed a word processor or spreadsheet program on your PC, they wouldn’t come with word counts, footnotes or charts. You couldn’t put a comment in a cell. You couldn’t even print in landscape. Those were all separate products from separate companies that you’d have to go out and buy for $50 or $100 each.
A few generations earlier, in the 1950s, a new car often wouldn’t have turn signal lights. That was an after-market product that you could buy and fit yourself with a drill and a screwdriver on a Saturday morning.
Over time, of course, Lotus and Microsoft, and Ford and GM, integrated - ‘bundled’ - all of that. Spreadsheets do charts, and the operating system handles printing, and no-one today argues that when you buy a car you should choose who supplies the lights. That killed a lot of third party products - it was, ipso facto, unfair competition - but we don’t argue it should be illegal.
On the other hand, in the 1960s there was a significant court case around Ford bundling a car radio and squashing competing radio suppliers. Is a radio an essential, necessary part of the product that should be integrated, or is it optional? What would that mean?
This is the centre of today’s argument about what Apple and Google include in their smartphone operating systems, and what Google includes in ‘search’, and it was the argument twenty years ago around whether and how Microsoft should include a web browser in Windows. If a company has market dominance, and it adds a feature to its product that is someone else’s entire business, this is inherently unfair, but life is unfair - does it follow that it’s bad, and that we should we do something about it, and if so, what?
People sometimes argue that this is all obvious and easy. Some things are ‘clearly’ essential features (for example, brakes) and some are ‘clearly’ optional, or at least substitutable (a web browser). But this is a fuzzy definition, and changes over time. Turn signals were optional once, and in the mid 1990s PCs didn’t come with a TCP/IP stack, and that was a separate product you could buy. Microsoft (and Apple) added one, and that became an anti-trust question - people proposed very seriously that this required intervention and that Microsoft should offer a choice for your network stack, and your file system, and indeed everything on top of the kernel.
There are, very obviously, some things that absolutely are illegitimate tying and bundling. But how do you capture that without capturing everything else? There is a huge grey area between telling Microsoft it can’t force OEMs to bundle Office with Windows and telling Microsoft that it has to make consumers choose their network stack when they turn on a new PC. Equally, we clearly don’t want to shrug, and give up, and let Google or Apple clone and squash any third party product they want to incorporate. But we also don’t want the DoJ or EU to run a monthly review of Gmail’s product roadmap. We need some intermediate level of rules - some set of general principles.
Hence, Elizabeth Warren famously proposed that ‘if you run the platform you can’t compete on it’, which sounds good, until you realise this means that you’ve just banned Google Maps on Android. If this is your principle, a platform can’t do anything that anyone else might want to do. So your new iPhone has no calendar app, and no camera app, and no email or web browsers… and indeed no app store. Those are all things that some other company wants to do too. This is saying that your car can’t come with headlights, or that a spreadsheet program can’t do charts. We have to do better than that.
A somewhat more practical approach is to think about choice screens, defaults, and self-preferencing, which together propose that “Google can still build things, but it has to be fair.” These are (what fun!) pretty complicated too, though.
Choice screens are an attempt to handle things that are necessary components but also somewhat modular and substitutable - you need a browser but you can have more than one, where you can’t usefully have two networking stacks. They also presume that the user will be able to understand and evaluate the choice (hence, again, this doesn’t work for networking stacks). So, you turn on your PC and it asks you which web browser and which media player you want to use, and which one you want to make the default. A setting to let you change the default afterwards is also a common requirement, and is pretty easy to implement in most cases, but most people never change the default later unless they’re prompted or unless one app is dramatically better, so forcing a choice is key to competition.
Choice screens were a coherent response to Microsoft including a web browser with Windows (both essential and modular). It’s harder to see how this would work on iOS or Android, where the issue is not one app but every app that comes on your new phone - do you want 20 choice screens?
The same question comes up if you apply this to Google search: you could require a choice screen to ask people if they want to see Yelp reviews or Google reviews when they search for a restaurant (and oblige Google to create APIs to enable third parties to plug into its results). But again, we’re not talking about one vertical search category, but dozens. How many choice screens do you want?
The prosecutor here would say that this is like a bank robber saying he’s robbed too many banks to be tried for all of them - if you want to address Google cloning other companies’ products, it can’t block you by making so many clones that you can’t keep up. But this is also ‘like’ a word processor adding word counts and page numbers and images and tables and a spell checker - do you want to ban all of those, or add five choice screens, or go back to the drawing board and think a little more about what might work?
Meanwhile, how many things does Google (or Apple) integrate in ways that make them invisible, such that they are no longer apparent to the user at all as a separate component that could be changed, and so where any choice screen would be baffling? And how many cases are there where the competitor doesn’t want to be a choice within the platform experience at all, but wants the platform to send the user directly to them?
This is the Yelp problem. Google could give you a choice screen to pick whether you see reviews from Google or Yelp embedded in the results page when you search for a restaurant, but Yelp doesn’t actually want you to see a summary of Yelp reviews on Google - it wants Google to link you to Yelp’s own site. The question is not which reviews, maps and phone numbers Google shows, but rather whether Google is allowed to show you anything about a restaurant when you search or it, or just give you ten blue links. Seen from Google, Yelp is the company that makes baggage carts complaining that somehow all new suitcases have wheels. Yelp doesn’t want to be one of the choices of wheels - it wants suitcases not to have wheels at all.
This loops us around to ‘self-preferencing’. Really, the concept of the ‘default’ is just one aspect of self-preferencing, which encompasses a wide range of situations in which a platform might allow all sorts of competition but puts its thumb on the scale in favour of its own option. Third-party ad tracking on iOS is moving to opt-in, but Apple’s own tracking is opt-out. Yelp’s restaurant reviews show up below Google’s. Third-party services get access to Siri a year after Apple’s services. And Apple draws up App Store payment terms in ways that it explicitly understands make some competing businesses ‘prohibitive’. This email was sent when Apple had 5% share in the USA - but now it has well over half. So what now?
A ‘self-preferencing’ regime wouldn’t ban Google Meet or Google Calendar - instead it would ban Google from stapling Meet into random parts of its UI. But again, this can get complicated pretty quickly. I think Apple’s policy on Spotify is unjustifiable (the email above references ebooks but this is the entire Spotify issue), but I also think Epic should lose its case against Apple. Yelp would like to be preferenced in search but it would prefer if Google didn’t tell you anything at all about a restaurant when you search for it. Many of Apple’s preferencing choices trade off user privacy against the viability of other people’s business models. Machine learning recommendation systems are hard enough to do with your own data structures without also incorporating data from other companies. And a lot of these questions trade off competition, and the user benefits that result from that, with simplicity and ease-of-use, which are also user benefits (that network stack again). The same applies to security: Apple’s iOS software model is a huge step change in user security and privacy at the cost of flexibility and competition. Pick one.
A common strand through all of this, and a point I’ve made repeatedly in writing about tech policy, is that important questions do tend to be full of complexity and trade-offs, not just in tech but in policy in general. That’s how policy works. There are always answers that are simple, clear and wrong, and each of the frameworks I’ve discussed here works in some cases but not others. As the old saying goes, all models are wrong but most of them are useful. This takes me to my other general thesis on tech policy: that this is a story for methodical and well-resourced on-going regulation, not slogans and breakups. It’s a story for legislation, much more than litigation.
Benedict Evans is a Venture Partner at Mosaic Ventures and previously a partner at a16z. You can read more from Benedict here, or subscribe to his newsletter.