Tim Wu has a new piece up on Medium called Ben Thompson’s “Stratechery”; the subtitle, I think, is more descriptive of Wu’s premise:
Smart, but a little too much Kool-Aid
Said premise is in the second paragraph:
Thompson has more recently begun to pronounce and analyze in the field of tech antitrust, and here he is on less solid ground. I appreciate that deep industry expertise is important in his area, especially, say, when designing remedies that make sense. Lacking a background in law or economics is not disqualifying. Nonetheless, I’d say Thompson’s readers are at risk of being misled if they rely too much on what he has to say about tech antitrust. For, as we shall see, his analysis relies too much on an idiosyncratic “digital markets are fundamentally different” thesis that really doesn’t hold up too well. Stated simply, I’d say he’s inducing his readers to drink too much of his “aggregation theory” Kool-Aid, as opposed to encouraging them to think more broadly or read more deeply to understand a slightly messier reality than he presents.
I appreciate Wu’s article, both in terms of its specifics (which I disagree with), its goals (which are implied), and its means (which I value). Time to drink some Kool-Aid!
(Orange-flavored, of course).
Wu’s primary focus is my recent piece United States v. Google:
According to Thompson, the Google case needs be understood primarily through what he calls aggregation theory, which is something of a specialized version of what economists call a two-sided markets theory. His theory asserts that 1) the quality of the user experience, rather than control over distribution, is what determines the winners in digital markets; and 2) a lead based on quality is self-reenforcing, because either more suppliers are attracted or the winner, with more customers, gets more feedback on what makes for a better product. (For those with a background in economics, Thompson’s aggregation theory is basically a mixture of a two-sided market theory with some positive feedback loop stuff thrown in.) Thompson says that “aggregators” (platforms, in economic, if not technological, parlance) are in this manner different than traditional monopolists, for they “win by building ever better products for consumers”…
The problem is that his aggregation theory isn’t aspirational. Instead, it is presented as a description of how the internet has “fundamentally changed the plane of competition” in a world where “on the internet everything is just zero marginal bits.” It also takes as its assumptions: “Zero distribution costs. Zero marginal costs. Zero transactions.” In that, in some ways, it is like the older economic models from the 1960s, except that they were at least billed as models, not depictions of reality.
Wu appears to be serious in his statement that assertions about the importance of zero marginal costs are best understood by looking to the past; his analogy for Aggregation Theory reaches back to the 1920s:
Here’s my send-up of aggregation theory: Imagine this is the 1920s and we were speaking of the invention of brand advertising, and someone says, “whichever brand has the most people attracting it will create a buzz that further favors the winner. Hence, traditional metrics of competition are out the window.” I think we’d all agree that brand matters, and indeed the invention of powerful brands did change competition. But it might be a little too easy to think competition actually has changed forever. And we can see Thompson falling into the novelty trap by asserting things like “the internet has made transaction costs zero” — a sentence that would make any serious economist howl with laughter.
I can’t speak for serious economists — I’ve generally enjoyed my interactions with them, and have been treated with nothing but respect while presenting the idea of Aggregation Theory — but the only thing I find humorous here is the idea that the Internet has not had a massive impact on transactions costs.
Consider various iterations of two-sided markets, of which Wu believes the companies I call Aggregators are a not particularly special version of. At the most basic level you might have something like a neighborhood flea market: on one side the market is a place for people to sell things, and on the other a place to buy things. Ultimately, though, every transaction entails sourcing supply, acquiring a customer, and executing the transaction itself. All three of those activities are costly.
Over the course of the 20th century, larger and larger firms became more and more efficient at managing these transaction costs. The Great Atlantic & Pacific Tea Company, more commonly known as A&P, was the best example of this. The company expanded rapidly in the late 1800s, which was critical in acquiring ever more customers (the company was also a pioneer in advertising and using low prices on select items to get customers into stores); meanwhile, A&P built a back-end operation to match, vertically integrating into being a wholesaler, particularly of its own private label goods, which, combined with A&Ps scale, helped it deliver on those low prices. By 1930 the company had 16,000 stores doing nearly $3 billion in sales, accounting for a 10% share of nationwide grocery stores.
By 1950 A&P’s market share peaked at 15%, although by that time A&P had transitioned to fewer, larger stores (around 4,500); the company was also facing an antitrust lawsuit, that it would eventually settle with a favorable consent decree. What ultimately doomed A&P, though, was its inability to adjust to a grocery market increasingly dominated by national brands advertised on television, along with uncompetitive labor costs and a failure to expand from city centers to the exploding suburbs. Each of these entailed higher transaction costs, and A&P couldn’t bear them.
A few decades later Walmart would follow in A&P’s footsteps as far as dominance is concerned, although their strategy started with exurbs and suburbs and worked backwards; the fundamental limitations of needing to open stores to acquire customers, build out logistical networks to acquire and distribute goods at scale, and actually stock shelves and check out customers remained, though. Walmart too has reached about 15% market share (albeit of a larger general merchandise market).
Amazon, meanwhile, has leveraged the Internet to dramatically decrease its customer acquisition costs in particular: the fundamental insight driving the retailer is that on the Internet shelf space is both infinite and available to anyone with an Internet connection; the company is still smaller than Walmart — 5% of general merchandise last year, although that number surely made a huge leap because of the pandemic — but it got there much more quickly: Amazon is only 26 years old, while Walmart is 58.
Still, as Wu notes, Amazon has plenty of transaction costs of its own:
Here is the danger: If you think competition is all about flavor and buzz (in the 1890s) or Thompson’s aggregation theory (right now), you might end up overlooking all of the other strategies and factors that could also lead to a lasting advantage. Consider Amazon. Thompson says that “the internet has made distribution (of digital goods) free.” But, as implied, that hasn’t made the distribution of physical goods free. And that is why a company like Amazon can, and has, gained a major advantage by building up a large physical infrastructure (warehouses), not unlike a steel producer in the 20th century, and strongly relying on a loyalty program (Prime). So, it turns out Amazon’s competitive advantage isn’t all about the fact that “on the internet everything is just zero marginal bits.”
Aggregators have all three of the following characteristics; the absence of any one of them can result in a very successful business (in the case of Apple, arguably the most successful business in history), but it means said company is not an Aggregator.
Direct Relationship with Users
This point is straight-forward, yet the linchpin on which everything else rests: Aggregators have a direct relationship with users. This may be a payment-based relationship, an account-based one, or simply one based on regular usage (think Google and non-logged in users).
Zero Marginal Costs For Serving Users
Companies traditionally have had to incur (up to) three types of marginal costs when it comes to serving users/customers directly.
- The cost of goods sold (COGS), that is, the cost of producing an item or providing a service
- Distribution costs, that is the cost of getting an item to the customer (usually via retail) or facilitating the provision of a service (usually via real estate)
- Transaction costs, that is the cost of executing a transaction for a good or service, providing customer service, etc.
Aggregators incur none of these costs:
- The goods “sold” by an Aggregator are digital and thus have zero marginal costs (they may, of course, have significant fixed costs)2
- These digital goods are delivered via the Internet, which results in zero distribution costs3
- Transactions are handled automatically through automatic account management, credit card payments, etc.4
This characteristic means that businesses like Apple hardware and Amazon’s traditional retail operations are not Aggregators; both bear significant costs in serving the marginal customer (and, in the case of Amazon in particular, have achieved such scale that the service’s relative cost of distribution is actually a moat).
Demand-driven Multi-sided Networks with Decreasing Acquisition Costs
Because Aggregators deal with digital goods, there is an abundance of supply; that means users reap value through discovery and curation, and most aggregators get started by delivering superior discovery.
Then, once an Aggregator has gained some number of end users, suppliers will come onto the Aggregator’s platform on the Aggregator’s terms, effectively commoditizing and modularizing themselves. Those additional suppliers then make the Aggregator more attractive to more users, which in turn draws more suppliers, in a virtuous cycle.
This means that for Aggregators, customer acquisition costs decrease over time; marginal customers are attracted to the platform by virtue of the increasing number of suppliers. This further means that Aggregators enjoy winner-take-all effects: since the value of an Aggregator to end users is continually increasing it is exceedingly difficult for competitors to take away users or win new ones.
This is in contrast to non-Aggregator and non-platform companies that face increasing customer acquisition costs as their user base grows. That is because initial customers are often a perfect product-market fit; however, as that fit decreases, the surplus value from the product decreases as well and quickly turns negative. Generally speaking, any business that creates its customer value in-house is not an Aggregator because eventually its customer acquisition costs will limit its growth potential.
Google is the canonical example of this definition, and the difference from Amazon, much less non-Internet two-sided markets, is significant. Start with the transaction costs: while scaling an Internet service is a profoundly difficult thing to do, requiring tremendous ingenuity, invention, and investment, the marginal transaction costs for serving one additional customer are zero. That is why Google, from the moment it launched, could be used by anyone in the world. Like Amazon, the company didn’t need to build out physical stores, but unlike Amazon, the company didn’t need to build out delivery infrastructure either. And unlike every retailer in existence it didn’t need to pay for supply. And — this is the part that makes Google truly unique — it didn’t even need to generate supply. The web already existed!
This is why Google can achieve 88 percent market share in the U.S. search market (according to the Department of Justice lawsuit), and achieve a similar level of share all over the entire world. The company’s scalability is effectively infinite, because serving additional customers is a function of fixed costs, not transaction costs; it really is not comparable to Amazon at all, in this regard, as the companies’ respective market shares demonstrates.
The same reality applies to Google’s marginal costs (including distribution); while Google spends a tremendous amount of fixed costs on its data centers and networking, any one search is “free”, including Google accepting the search term, computing the result, and delivering it to the user. Moreover, this same principle applies to Google’s advertising business: the vast majority of advertising on Google is acquired via self-service portals that price ads automatically via real-time auctions. Yes, the infrastructure necessary to enable this business requires substantial investment, but the only transaction costs on any one specific advertising purchase are credit card fees.
This is a business that requires more analysis than calling it a “two-sided market…with some positive feedback loop stuff thrown in”; for one, I would argue that all two-sided markets have positive feedback loops. Any market that touches the physical world, though, accumulates an ever increasing number of tiny costs along the way, whether that be labor costs, shipping costs, rent costs, etc.; moreover, the logistical challenges entailed in managing those costs incur their own cost in managerial complexity, and every investment to overcome those challenges become sunk costs, making it difficult to adjust when the market changes (this is particularly acute because it takes time to make these investments).
Google, on the other hand, faces none of these natural drags on scale. More search means better search, thanks to the ongoing feedback of billions of users ranking every Google search result (by clicking on the best result); more advertisers means better advertising results, for the same reason. This matters greatly for antitrust because at some point you need a theory of harm: how exactly is Google making things worse for users or advertisers?
This is also why I disagree with Wu’s characterization of switching costs:
Whether this is core to his theory or not, Thompson also takes a highly anti-empirical approach to switching costs. He endorses the old 1990s idea that “competition is just one click away,” which may have been true in 1999, but that can’t be taken seriously now — if what he means is that the costs of leaving Google or Amazon or Facebook are close to zero. The real question is whether there are, for the average person, costs to switching from Facebook or Google to use something else — leaving behind Gmail, friends, and so on. The assertion that those costs are near zero is magical thinking. Indeed, one of Google’s most important strategies over this decade — its tell — has been to increase those switching costs, those barriers to entry.
Consider Google specifically: the company’s core product, Search, has for its supply the open web. Indeed, that is what let the company be at scale, instantly, in a way no other company ever has. It follows, though, that every other search engine — including Bing, DuckDuckGo, etc. — has access to the exact same supply.
Admittedly, Google does have its own local results in particular; Yelp has an entire website complaining about this fact, arguing that the search engine should be forced to include third-party content in its local search results because those results are better for consumers. That, though, is a mark in the competition’s favor: is Wu’s position that Google’s allegedly inferior local search results is lock-in?
Meanwhile, I am puzzled by the reference to Gmail; it is unclear to me why it is a competition concern when a user chooses to use a free email service that is quite obviously locked to the service provider. The relevant market here is not Gmail, it is email, and not only is there a huge amount of competition for hosted email, it is fairly simple to set up your own email server. Critically, there is absolutely nothing Google can do to stop you from doing so, even if they wanted to.
This drives at the biggest reason why I believe a distinct definition for “Aggregators” is important; while Wu casually conflates Aggregators and platforms (“in economic, if not technological, parlance”), I believe the distinction is substantial and crucial. I explore that distinction at length in A Framework for Regulating Competition on the Internet, but the critical point goes back to the email example:
- Platforms are essential to their value chains. You can’t have a Windows app, for example, without the Windows API.
- Aggregators, in contrast, are not essential, but they are convenient. You can go to a website directly — just type in its URL — but for most consumers, for most pieces of information, it is far easier to search.
What is fascinating about many of Google’s most ardent critics is that they themselves aspire to be Aggregators. Yelp, for example, doesn’t operate any local businesses. It doesn’t prepare the food, or cut the hair, or teach the class: its business model is to aggregate so many users that local businesses feel compelled to be on Yelp, the better to reach more customers than they could on their own. The same applies to Trip Advisor, or Expedia, or any other vertical search company. All of those sites are only a click away.
And, critically, so are the entities that actually provide the services or information that the user is seeking. Airlines and hotels invest heavily in loyalty programs, for example, because they want their best customers to come to them directly, not via an Aggregator, whether that Aggregator be Google or Booking.com. That not only sounds like competition, it also sounds like an exceptionally customer-friendly outcome.
The Missing Intersection
Where I think Wu and many tech critics go wrong is missing how the question of zero marginal costs and zero switching costs intersect. First, because Wu does not believe that Google is unique as far as scalability is concerned, he appears to assume that the company must be doing something nefarious to command such market share. And, by the same token, there must be some sort of unfair lock-in, because again, companies ought not be so dominant. This is a sure recipe for lazy arguments that end up criminalizing the basics of business.
How does all this relate to antitrust? Antitrust should be dealing with the reality of anticompetitive behavior in markets, not ideals of how companies work. And it is the difficult job of the law to determine which of these durable advantages just described are part of fair competition (for example, a better user experience) and which are not (for example, buying out dangerous rivals, or exclusionary deals that keep out competitors)…
We may summarize the problem for Thompson this way: Why, exactly, did Google pay Apple billions to gain control over distribution rights? And why, to bring the law into it, hasn’t Google settled the case? If aggregation theory is right — if competition has changed in the digital market and the best user experience wins — then Google doesn’t need to spend that money.
I honestly don’t think this is too complicated: defaults do matter, and given the fact that Google makes somewhere north of $250 in revenue per U.S. user it is well worth sharing some of that revenue to ensure it gets as many users as possible. Consider iOS specifically:
- According to the Department of Justice “This agreement covers roughly 36 percent of all general search queries in the United States”; assuming that share of search queries is inline with share of revenue (which is almost certainly not the case, given the relative spending power of Apple’s userbase), the agreement covers $26.9 billion worth of revenue.
- Further assume that, were Google not the default, the company would lose 25% of Apple device searches it might have gained otherwise; this equates to a revenue loss of $6.7 billion for the U.S. alone, and again, this number is almost certainly conservative given the relative spending power of Apple users.
Moreover, those searches would go to Google’s competitors, not only giving valuable data that would help make their search engines better, but also increasing the efficiency and relative attractiveness of their ad products. I do still believe that Google would continue to win on the merits, but it would be more costly, not less.
Which, of course, is why I support the Department of Justice’s lawsuit, and why I have been outspoken about acquisitions by Aggregators. Aggregators already have intrinsic advantages given the nature of costs on the Internet; I don’t believe they should be able to augment those advantages with contracts or by acquiring customers (I do not, however, favor a ban on other types of acquisitions).
The difference I have with Wu, as far as I can tell, is that I see these agreements and acquisitions as frosting on an Aggregation cake, as opposed to the fundamental drivers of their dominance. Google isn’t dominant because they broke the law, they are (arguably) breaking the law well after their dominance was established, and that distinction matters when it comes to crafting remedies and regulations that actually work.
Differentiation and Gatekeepers
The most disappointing part of Wu’s essay, though, at least on a personal level, is the conclusion:
This may be too much for some readers, but a last problem with aggregation theory is that its “winner take all” assertion assumes away the importance of differentiated user preferences. In other words, it tends to assume that there is one “user experience” that is preferred by everyone, and by depending on feedback, the product can be improved to match that…
Perhaps Thompson has addressed this somewhere, but I thought it important to point out. The model only works well, I think, either when consumers have identical preferences or when they want the greatest number of suppliers for some reason, or maybe when consumer value convenience even over what they’d called their own stated preferences (the so-called tyranny of convenience).
The very premise of this site is that the Internet takes preference differentiation to the extreme, resulting in never-ending niches that can be profitably filled. Moreover, this isn’t simply about small websites: just last month I wrote about how Disney is pursuing an integration strategy that is an antidote to Aggregators like Google and Facebook, and how that can be model for other media companies. The Article included this distinction:
Aggregators are content agnostic. Integrators are predicated on differentiation.
Facebook reduces all content to similarly sized rectangles in your feed: a deeply reported investigative report is given the same prominence and visual presentation as photos of your classmate’s baby; all that Facebook cares about is keeping you engaged. Content created by Disney, on the other hand, must be unique to Disney, and memorable, as it is the linchpin for their entire business.
So yes, I have “addressed this somewhere” — I addressed it three weeks ago. Contrary to Wu’s caricature of me, I don’t believe that Aggregation Theory applies to everything, but the things to which it does apply — like when “consumers have identical preferences or when they want the greatest number of suppliers” for something like search results — it matters a great deal. To that end, it seems rich to criticize me for “false confidence” and an unwillingness to “think more broadly or read more deeply” when one can’t be bothered to scroll down my home page.
Then again, perhaps an honest debate isn’t the goal. Wu foreshadowed his essay on Twitter:
I think it is time to start reviewing some of the work of Ben Thompson of Stratechery. He's the kind of guy who is smart enough to sound like he knows of what he speaks, but is just too ignorant of the relevant law and economics to make the pronouncements that he does.
— Tim Wu (@superwuster) October 26, 2020
Wu added, in a tweet he seems to have deleted, that my writing is “quack medicine”:
This casts Wu’s comment that “Lacking a background in law or economics is not disqualifying”, in a different, more backhanded, light. For decades antitrust was indeed limited to those with a background in law or economics; that is the only way politicians, attorneys general, judges, general counsels, or the media would pay attention to what you had to say. The media in particular had a monopoly on the dissemination of information, and credentials were the way to get into their channel. It’s hard to escape the sense that a breakdown in gatekeepers bothers Wu.
And, by the same token, perhaps this is why I do believe that the Internet is something profoundly different; it has certainly made my career far different than it might have been were I a decade or two older. And, by extension, perhaps this is why I can see the benefit of Aggregators: Google and Facebook and Twitter may have been terrible for traditional media companies whose business models depended on controlling distribution, but they are fantastic for giving consumers exactly what they want, including a perhaps heretical view of antitrust.
This does raise the question as to why you should believe me, or anything else you read on the Internet. This interaction arose in response to Wu’s initial tweet:
Why would getting stuff right in the long run entail some degree of getting stuff wrong in the short run?
— Nicolas Bray (@yarbsalocin) October 26, 2020
I don't believe that anyone has a monopoly on the truth, and one way to figure out where you are wrong is to put it out there and have folks like Tim convince you that you got it wrong
— Ben Thompson (@benthompson) October 26, 2020
This is why, whatever Wu’s motivations, I appreciated his piece. I do disagree, in part because I don’t think Wu understands my argument, but that’s ok: it’s an opportunity to make my argument in a different way, as I tried to do today, and perhaps change his mind, or yours. Or perhaps I failed, and you agree with Wu that I have created a theory where one ought not exist. Again, that’s fine: now you know what you believe better than you did before this piece, and perhaps you will view my other writing with increased skepticism. That’s a good thing!
More broadly, the pre-Internet world, governed as it was by gatekeepers, was certainly a more unified one, at least as far as conventional wisdom was concerned — this applied to law and economics just as much as anything else. At the same time, that does not mean the pre-Internet world had a better overarching grasp on the truth, given how much more difficult it was for dissenting voices to gain distribution. If I happen to be correct about Aggregation Theory, and that the way to understand Google depends on more than “two-sided market theory with some positive feedback loop stuff thrown in”, then I would like to think the fact that Stratechery can exist, without anyone’s permission, is a good thing.
After all, I am concerned about just how powerful these companies are. I am not intrinsically opposed to regulation — I believe in democratic oversight — but the risk of getting regulation wrong is that companies like Google become more entrenched, not less. This is why I objected to GDPR before it came in force, and why I spend time writing about antitrust (which, I might add, is not a traffic driver!). Getting the law and the economics right are important, particularly if the Internet challenges the fundamental assumptions underlying them.
- Please forgive me for the long excerpt, but despite Wu’s exhortation that one needs to read deeply to understand these issues, it appears he has not done me the same honor [↩]
- And yes, in the very long run, all fixed costs are marginal costs; that said, while the amount of capital costs for Aggregators is massive, their userbase is so large that even over the long run the fixed costs per user are infinitesimal, particularly relative to revenue generated [↩]
- In terms of the marginal customer; in aggregate there are of course significant bandwidth costs, but see the previous footnote [↩]
- Credit card fees are a significant transaction cost that do limit some types of businesses, but will generally be ignored in this analysis [↩]