Breaking up Big Tech: scissor-line suggestions for smart cuts (2025)

Abstract

1. INTRODUCTION

Under the alias MrBeast, Jimmy Donaldson is one of the most successful online content creators on YouTube, with over 300 million subscribers and more than 50 billion views. When asked about how he puts his faith in YouTube, producing million-dollar videos for the platform, Donaldson is certain of his business model:

I think YouTube is going to get bigger [slam on the table], bigger [slam on the table] and bigger [slam on the table]! It comes installed on every Android device, which is like 85 percent of all operating systems. And 90 percent of searches in the Western world are on Google, which YouTube pops up on. (…) I think YouTube in ten years is going to be bigger than we all imagine. I have no freaking doubt in my mind.1

The insight shows the forces that maintain the dominance of what is now the ‘MAMBBAAs’—Meta, Alphabet, Microsoft, ByteDance, Booking, Apple, and Amazon—and grow the powers of other superstar platforms like X, Netflix, and Spotify.2 In this domain, being big is a good predictor of becoming bigger.

Let there be no misunderstanding: these tech platforms have hugely expanded consumer choices and have facilitated market access for many small and medium-sized enterprises, while steeply decreasing transaction costs.3 They have opened up a sea of knowledge and possibilities. Anyone with an internet connection now has all the world’s writing, music, and images at their fingertips. If consumers’ habitats were until recently a little larger than the local town’s main street, today each start-up shop is able to reach clientele everywhere from anywhere, just by setting up an online store in nine easy steps with Shopify, accepting payments with PayPal or Adyen, and using Google Ads to advertise and Amazon Marketplace to sell its products globally. The platforms’ ability to pool information generates tremendous benefits, as well as new and innovative services and products.4 Without question, their contribution to the growth of consumer welfare has been enormous.

Yet over time, these once highly innovative start-ups have become ‘Big Techs’: a few powerful platforms that control digital markets and the ecosystems they have built around them. These powers have become entrenched by positive network effects: the more users are active on a single platform, the better the service that the platform can provide to all of its users, and hence the more users want to be active on that platform.5 A search engine, for example, has a better ability to respond to a query with search results and recommendations that are likely to suit the user when more people use it and so feed the algorithm with information about what is relevant through their search behaviour.6 Likewise, the value of reviews on hotel or restaurant comparison platforms is higher, the more people evaluate their experiences there.7 Therefore, in principle, everyone gains from joining the same platform for a particular class of services, to enjoy the benefits of all those active there—provided these benefits are indeed passed on to the users.8 As a result, once market dynamics have tipped towards one platform becoming dominant, it is very hard for an alternative platform to ever catch up.9

While their users gradually became locked into their ecosystems, these platforms have found ways to generate huge revenues, mostly through advertising, fees, and referral commissions. Their enormous profits have subsequently allowed the platforms to expand into various commercial activities, investing and further aggrandizing their profits.10 They have thus all become hybrids with strong tentacles in the real, non-digital economy. Lina Khan observed years ago that the dominant digital platforms ‘have integrated across business lines such that they both operate a platform and market their own goods and services on it’.11 These commercial ties have only become more interwoven and stronger since. Today, Big Techs are entrenched conglomerates with vast commercial interests. Amazon’s sales revenue was $575 billion in 2023, which corresponds roughly to the gross domestic product (GDP) of Belgium. Apple’s sales value totalled $386 billion, and Alphabet’s ad revenue was $238 billion. The current market valuation of the MAMBBAAs taken together is over $10 trillion, equivalent to a ‘Big Tech country’ having the third highest GDP in the world—roughly double the size of Germany and Japan, and more than half the GDP of China.

The MAMBBAAs’ big intermediaries are the gatekeepers of online traffic. They attract and redirect nearly all eyeballs, including the wide range of services that their own subsidiaries are offering within their ecosystems. A concerning consequence is that the integrated, hybrid business models have given the dominant platforms incentives to modify their online advisory functions to suit their own interests, which need not be fully aligned with those of their users. The existing commercial interests in certain redirects, clicks, and purchases by these few large companies can thus fundamentally affect what people are presented with, what they watch and share online, and which transactions they make in the non-digital economy. As a result, we, the users, may not get the information that is best for us—despite it being available—but that is best for them, the Big Techs.12

Information bias in the recommender systems is the intentional, typically pre-programmed distortion of information presented to a user to steer the user away from the best matches to her query and towards content that benefits the platform instead.13 It has different identifiers in the literature, including ‘search bias’, ‘self-preferencing’, ‘third-party discrimination’, ‘default option’, ‘steering’, and preventing ‘side-loading’. In all these practices, users are intentionally not given information about the best matches available for them. Users really have no way of knowing the extent to which this skewing of the information that the platforms provide towards their own interests is actually happening, but the dominant tech platforms have both the ability and the incentive to engage in it. Information bias can even take on extreme forms, such as algorithms that lead users down rabbit holes, lock them into echo chambers, and spread fake news. Even though in response to health and safety concerns, platforms have taken some measures to protect their users, those measures are often perceived to be too limited—if even genuine and not ‘screenwashing’.14

The Big Techs’ omnipresence and powers have led to numerous social concerns and concrete accusations of monopoly behaviour towards various sides of their markets, in both the United States of America (USA) and the European Union (EU).15 These include the abusive resale of private information, excessive commission and advertising fees, tying and bundling, and third-party product line imitations. In attempts to curb their market powers, competition cases have been tried against the dominant platforms, leading to fines, injunctions, and behavioural remedies.16 The USA has stepped up antitrust enforcement again, more than 20 years after the seminal case of United States v Microsoft Corp, which was finally concluded in 2001, bringing cases against Google and Apple. In Google (2024), the court decided that the platform monopolized the search market through exclusionary distribution contracts, upon which the Department of Justice suggested that divesting Chrome and Android may prevent Google from advantaging Google Search.17 Apple is alleged to have monopolized the smartphone market.18

The European Commission found abuses of dominance in Microsoft (2007), Google Shopping (2017), Google Android (2018), and Apple (2024b).19 It obtained commitments from Apple to open up its Near-Field-Communication technology on iPhones for third-party mobile app developers in Apple (2024c).20 The new draft Guidelines on Exclusionary Abuses explains how self-preferencing produces exclusionary effects and network effects that cause ‘winner-take-all’ dynamics, to which Article 102 TFEU will be applied ‘vigorously’.21 Some of the unwanted behaviours were also cast in regulation, in the Digital Services Act and Digital Markets Act (DMA), which aim to ban exclusionary behaviours and increase interoperability, data portability, and neutrality.22 They are a foundation for trying to make platform markets contestable. First preliminary findings under the DMA, against Apple’s steering rules for the App Store, have been adopted, and further non-compliance investigations are underway.23 Meanwhile, Meta’s ‘pay or consent model’ and Alphabet’s self-preferencing in search services are being scrutinized by the Commission.24 Also, new gatekeepers continue to be added under the DMA—Booking being the latest addition.25

There are concerns, however, that behavioural remedies may not suffice to control the dominant platform companies’ behaviours.26 They can be difficult to enforce effectively, may not be a perfect fit for novel cases, will require constant and ever-changing policing, and are likely to inspire creative compliance and avoidance strategies—even ‘bullying the regulator’ by withholding certain product features, such as Apple’s Intelligence and Google Maps, while blaming (potential) regulation.27 The platforms can draw up novel market-expanding strategies, including anti-competitive ones that are not straightforwardly classified as such under the DMA and take time to unmask. Competition policy can remain a forerunner in identifying new business behaviours to ban in future revisions of the DMA, but that may all not suffice against the natural monopoly character of the platforms. Therefore, where behavioural remedies are cumbersome to enforce effectively, more invasive structural remedies may be called for—and they have been proposed by Khan (2017, 2019), Wu (2018), and Van Loo (2020).28 In fact, some of the platforms’ own co-founders have stressed the need for their reorganization, such as Facebook’s Chris Hughes and Apple’s Steve Wozniak.29 The latter said on Bloomberg News:

Big Tech can get away with a lot of bad things. (…) I am in favor of looking into splitting up companies. I wish Apple on its own had split up a long time ago and spun off independent divisions to faraway places and let them think independently. (…) I think that Big Tech has become too big, and it’s too powerful a force in our life, and has taken our choice away.30

In the 2020 US presidential elections, the slogan ‘Breaking up Big Tech’ was a prominent part of the Democrats’ campaign. In a 2021 poll, 59 percent of Americans somewhat strongly supported breaking up Big Tech, and only a minority of 27 percent opposed it.31 Since then, there have been several legislative proposals in the USA to unwind and ban acquisitions by platforms. The EU’s approach to structural interventions was initially more reserved.32 In the DMA, mandatory divestments are a remedy of last resort only.33 But cuts in Alphabet are being suggested in the statement of objections in Google Adtech (2023).34 Moreover, in January 2024, the European Parliament adopted a resolution in which it ‘invites the Commission to make better use of structural remedies as a matter of last resort [and] reiterates its call on the Commission to end the primacy of behavioral remedies in EU law’.35

Blunt breakups, however, risk destroying the positive network effects that single platforms generate, may leave unviable shredded businesses that cannot be sustained on their own, and can discourage innovation. More specifically, it is not obvious what type of structural remedy would effectively address the problems thought to be caused by Big Tech, and how. The thinking about possible structural interventions in hybrid platform companies is still rudimentary, both conceptually and practically. The delicate question in designing them is: How can we preserve the highly valuable network benefits of pooled information, which essentially require the existence of single platforms, and at the same time ensure that those benefits are passed on to the users, rather than the platforms exploiting them for their own profit?

In this article, we focus on the dominant platforms’ ability to act as an intermediary of information, a recommender that can steer its users around the internet. We define ‘recommender systems’ as the parts of the firm that provide primary information to users, such as search results, rankings, and location-based information. Those are the functions of a platform that determine the personalized match values of search results in response to a user’s query and their presentation, based on the information the platform possesses about all the possible options that could be of interest to the user and his/her/their likely preferences. Users’ sole reliance on a recommender system is what gives the platform the power to bias the information it curates. We lay out a concrete type of breakup approach for ‘smart cuts’ that addresses information bias problems within this recommendation function while preserving the platforms’ network benefits as much as possible.36

The idea is to insulate the core part of a hybrid platform that provides advice to consumers from the commercial tentacles that the platform grew into the real economy. The proposed separation between the recommender systems and the commercial peripheral services creates distinct entities with the sole incentive of providing unbiased advice, so that users can be assured of receiving objective search results and recommendations on which they can base their purchasing decisions. The recommender systems can be seen as essential facilities for all the peripheral services that cater to the needs of platform users, including those separated under the company’s umbrella if such suffices, where commercial transactions can be executed. The structural intervention we sketch invites competition between suppliers on a platform, and ultimately between recommendation systems too, when complemented with mandatory data portability requirements.

The remainder of this article is organized as follows. In the next section, we first elaborate on the problem of bias in the platforms’ central recommender systems. Section 3 reviews the debate on breakups. Section 4 discusses the unwinding of past mergers, which are concrete alternative breakup proposals. Section 5 sets out the idea of smart cuts conceptually, with a wide application to platforms. Section 6 illustrates the concept with suggestions for smart cut scissor lines in the organigrams of Alphabet, Amazon, and Apple. When properly executed, the approach has several main benefits, which are discussed in Section 7. Section 8 concludes and explains how the concept of smart cuts has wider applications to other hybrid platforms as well.

2. THE BIG TECH INFORMATION BIAS PROBLEM

One of the dominant tech platforms’ greatest assets is that they have a huge information advantage over their users, which is both the cause and the consequence of their dominance, and their potential for creating immense social value. Their bigness pools the information of many, with the potential to benefit all. Their advisory role is in promoting particular recommendations, while suppressing less relevant information. Google’s infamous ranking algorithm, which remains largely a secret architecture, returns organic search results in an order that answers the user’s query with the best possible matches: relevant links first, less relevant ones next, followed by the least. Users benefit as long as the algorithm does this to the best of its ability in their service.

However, a fundamental problem that comes with the dominance of information gatekeepers is that it remains essentially unknowable to users, or knowable only at a high cost, whether the information that they are presented with in response to their search query on the platform is truly the best available to them. After all, in the vast ocean of knowledge out there, no one can know what he/she/they does not know exists. By the very nature of the market structure, there really is no truly good alternative platform or other practical way to compare the top-ranked search results, videos, or online purchase suggestions given by the dominant platform against better alternatives that it may have available but that it has ranked lower, out of sight, or not at all. The platform’s near-monopoly position, in other words, allows it to present lower quality matches to the user than those it has available, to the platform’s own benefit. This can be seen as an exploitative abuse of dominance.

The platforms’ information advantage makes it nearly impossible to see through their recommender systems and determine whether there are hidden biases. This was already true for secret ranking algorithms, and it is even more so with the accelerating sophistication of artificial intelligence (AI) and deep learning techniques applied to them. These programmes offer ample opportunities for modification. When Alphabet launched Gemini as an image generation AI tool, for example, the system turned out to have been programmed to add search terms for diversity to the prompt, which distorted the information.37 Similar subtle modifications of an AI-based search engine could easily lead to self-preferencing or steering behaviour. The more users rely on the recommender, the more possibilities for bias. In its updated AI Overview, Google offers a complete plan of action in a blurb text first, summarizing what it has machine-learned online in response to a single search query before listing the search rankings.38 Users no longer need to scroll through search results to find information on the Internet, as Google’s advice is neatly summarized on top. This can pull users further into the Google ecosystem and makes it less transparent how the system arrived at its recommendations.

The perversion subsequently comes from the dominant platforms commercializing their advisory function through the promotion and (indirect) selling of services and products to their users, earning profits through subscriptions, commissions, or margins. This introduces incentives for information bias in the recommender systems, by which the information presented to a user is intentionally distorted away from the best matches to that user’s query, towards a delivery that is beneficial to the platform instead. That is, information with a higher match value for the user, based on all tangible and intangible aspects of the query available to the recommender system, is deliberately withheld or given less prominence in the recommendations, with the intent of increasing the probability that the user takes another action than their first-best, to the benefit of the company that owns the recommender system. This results in consumer harm compared to if the recommender system would present the information with the highest match values that it has available, unbiased.

The problem arises particularly when the information provider has a stake in the choices that the user is likely to make on the basis of the information provided. That is, when the curation leads to some kind of ‘transaction’ from which the recommender system obtains a (commercial) benefit. If Google earns money from a sale made through a link it provides, it has an incentive to present that link higher up, in a spot where the user is more likely to click on it, rather than giving another link, even if that other link would have been a better deal for that user—which Google is aware of. In other words, by acquiring commercial interests, Google has additional reasons to change the order of the links that it presents as the ‘organic search results’ from what the original, unbiased ranking algorithms would have been—and no one would know. An obvious example is Amazon placing a product in the Buy Box from which it makes the highest commission. Yet also feeding apocalypse conspiracies on YouTube, against the user’s better judgment, can be beneficial for a platform that happens to own a preppers’ webstore. The loose and opaque integration of a Big Tech’s recommender system and commercial intermediation interests have thus created both the ability and incentive to perturb, however slightly, its advice to its own benefit rather than to the benefit of the user.

Notably, already in their original explanation of Google’s algorithm, Larry Page and Sergey Brin point out this potential problem when they wrote:

[A]dvertising funded search engines will be inherently biased towards the advertisers and away from the needs of the consumers. (…) [A]dvertising income often provides an incentive to provide poor quality search results.39

A hypothetical example serves to illustrate the core incentive problem for a straightforwardly optimal advisory service, to a (hungry) car driver: the shortest safe route (say, in time) from A to B using Google Maps. The positive network effects of Google Maps are evident and potentially large. The more drivers use the platform, the more information Google Maps can collect on roadworks ahead, busy stretches, and traffic jams, and so the better it can predict delays, as well as relay and divert traffic streams to direct all drivers to the fastest (or shortest) possible safe way to where they want to go (location B). This provides good quality reasons for all drivers to have a preference for Google Maps. However, it also gives the platform power over all these drivers, which it may use to detour them, unbeknownst to the drivers, to pass by places where Google Maps, or one of the other Alphabet subsidiaries, can make money.

One such additional service Google Maps offers is displaying nearby restaurants, showing their quality reviews, prices, and current busyness. This information, too, serves the driver who would like to make a restaurant stop on her way from A to B well. However, suppose Google has a stake in another restaurant C, which is nearby but not quite on the fastest route from A to B.40 In fact, there is a better restaurant D that is even closer, taking less time to visit. Of course, drivers do not know this—which is why they search for advice. But their trusted platform does. Now the platform has an incentive to recommend the driver to take a detour that will pass by restaurant C, even though restaurant D would be a better stop. The restaurant that is prominently displayed is not necessarily the one that best fits the user. The best match might be ignored, because it was not making use of Google’s advertising services, whereas the second-best restaurant is. Drivers, relying on Google Maps for route recommendations, barely have the time or energy to check the appropriateness of the recommendations. Focused on the road while driving, also once arrived—at restaurant C or B—most people remain unaware of better alternatives and are unable to even obtain the information necessary to second-guess the ride Google Maps just took them on. For even if one is suspicious of a diner detour, how could you ever be sure in hindsight that the other restaurant on the quicker route was not worse—for example, more crowded maybe? The information was given in real time, and the next best alternative, probably TripAdvisor, is of substantially lower quality, exactly because Google Maps has so many more users.

These subliminal forces play much the same in the other dominant tentacled platforms as well, seducing them to mix advisory with commercial functions. This information bias problem may not arise if there were sufficient competition between platforms, so that consumers could switch. In that case, the platform could not afford to bias. It would have to balance showing good search results, in order to retain its users, and find a way to generate high sales or click-through rates on the ads it shows, to obtain commissions. With alternative platforms present, mixing too many second-best sales suggestions or sponsors into search results and recommendations would turn users off and away. When the search value users obtain decreases, it would make them leave for an alternative platform, if one existed. More high-quality searches and services instead lead to more users who, in turn, view and click on advertisements and purchase items. But the problem is that the Big Techs are not subject to such cleansing competition. This situation edges them to choose taking revenue over giving good advice more often.

There is indeed ample evidence of Big Tech meddling with information. Microsoft (2007), Google Android (2018), and Google (2024) were about the default options presented to users, which is a form of self-preferencing that limits the visibility of competing services. In Google Shopping (2017), Google Search demoted rival comparison shopping services that were ‘Panda-ed down’ from the top of the search results to pages far lower on the list in order to advance the use of Google Shopping.41 Amazon, after having been investigated for abuse of dominance by the European Commission, accepted commitments to not apply discriminative rules in the selection of the items it presented in the Buy Box.42 Yelp has complained that Google made it harder to find in the run-up to compliance with the DMA.43 Apple has been accused of pushing certain Apple apps in its App Store over others.44 It prohibited content providers from informing their users about Apple devices of the price and availability of alternative streaming services outside the App Store.45 In response to the European Commission forcing Apple to open up its ecosystem, the company flexed its market muscle by imposing higher fees on entrants and denying Epic Games access to its development platform.46 Meta was found to push transactions on Facebook Marketplace by automatically referring Facebook and Instagram users to it rather than alternative marketplaces.47 The companies’ settlement offer was rejected.48 Microsoft is again accused of forcing its browser, Edge, by suppressing the default browser choice when users open links in Outlook and Teams and making it difficult to uninstall.49 It is also alleged to have anti-competitively tied Teams with Office 365 and Microsoft 365 and to have excluded rival cloud services on Windows Server to advance its own cloud service, Azure.50 While antitrust enforcement exposed those behaviours and tried to address them individually, it does not seem to have been able to take away the core problem yet.

3. BREAKUP PROPOSALS

In the meantime, calls for breaking up the Big Techs to curb their powers are being made in the USA. The most pronounced ones are by Senator Warren, who proposed designating large online marketplaces as ‘platform utilities’ and prohibiting companies with more than $25 billion in revenue from both owning an intermediary and participating on it.51 President Biden intended structural remedies with his Executive Order to the Federal Trade Commission (FTC) to block mergers between firms with network effects and appointed Lina Khan to the FTC.52 The final report of the 2020 Digital Markets Investigation by the House Judiciary Committee, which held spectacular hearings with the Big Tech CEOs, mentions the use of breakups as a potential ‘easier to administer’ remedy but makes no concrete proposals.53 Several ideas have been floated. One is to separate hosting services, where data are stored, from curation services that multiple parties could provide on the basis of differentiated algorithms—for example, focused on privacy, child safety, or free speech. Those recommenders would all make use of the same data stored by the hosting services, which would be regulated to ensure non-discriminatory access.54 Tim Wu suggested a ‘quarantine’ policy for ecosystems, by which the owner of essential services is not allowed to be involved in services active on the network.55 Still, so far, the only concrete breakup proposal is a lawsuit prepared by the FTC against Facebook to divest Instagram and WhatsApp, which is not expected to go to trial before the end of 2024, if at all.56 In the Google (2024) case, remedies, which might include separating Chrome and Android from Google Search, are only expected by the end of 2025.57 Meanwhile, the bipartisan Ending Platform Monopolies Act aims to bar platforms from owning an online platform as well as concurrent businesses.58 Legislation is being put forward that would allow regulators to block large technology firms from buying competitors, such as the Platform Competition and Opportunity Act.59 These cases cover different aspects of platform dominance, which may not deliver a coherent remedy approach.60

John Kwoka and Tommaso Valletti argue, on the basis of a number of historical cases such as Microsoft and AT&T, as well as self-initiated corporate restructurings, that structural remedies are perfectly possible and effective as long as divestitures contain the firm to its core business.61 Documented structural interventions provide promising examples that dispel concerns that breakups destroy value. The AT&T breakup is widely considered a success.62 According to Martin Watzinger and Monika Schnitzer, it led to a significant increase in innovation, both by the divested companies and by competitors.63 Similarly, Felix Poege shows that the breakup of IG Farben led to an industry-wide wave of innovation.64 In addition, post-breakup, the seven Baby Bells consistently outperformed the market, providing evidence that breakups can unlock value.65 Interestingly, also among venture capitalists breakups and part sales are eagerly discussed as profitable investment opportunities, on the view that Alphabet could break itself up into parts that would be more valuable separately than as a whole. While these investors seem to overlook the competition agencies’ intent to reduce market power and hence monopoly rents, the ease with which they talk about breakups is encouraging for policy.66

However, economists have warned against possible damage of disallowing synergistic acquisitions and cutting in the wrong places, possibly too deep. The welfare effects of separating an intermediary and sellers are complex, because the outcomes of separation are highly context-dependent. Merger blocks and breakups, after all, can impede and destroy the value-generating network effects that can justify the platforms’ bigness. They could lead to gaps in the product portfolios, introduce double marginalization, and reduce on-platform competition.67 The seamless integration of various services in the hybrid model adds value for consumers, so that breakups could lead to hassle costs and possibly higher prices too. Fiona Scott-Morton fears that ‘just breaking them up’ indiscriminately could actually harm consumers, workers, and innovation.68 Still, she sees benefits of separating Android from Alphabet.69 Eleanor Fox and Donald Baker note that breakups have historically been proven to be rare and difficult.70 Aviv Nevo points out that litigation will be lengthy and costly, and it is not clear that competition will increase.71 Andrei Hagiu and others indeed show that if a tech platform is not allowed to sell on its own marketplace, the loss of the one-stop-shop convenience benefit could increase search costs and may weaken price competition.72 Carl Shapiro concludes:

Economic theory and evidence indicate that a widespread campaign to break up large firms on a “no-fault” basis would slow economic growth by making our economy less competitive and less innovative.73

For concrete interventions, Jean Tirole cautions that while structural remedies may be intellectually appealing: ‘The devil is in the detail’.74 In fast-moving digital markets, Tirole points out that the challenges of successful structural interventions are in identifying the stable essential facility to set apart, and how to insulate it without destroying the benefits of network externalities. Cutting in the wrong places may deprive users of their interconnectivity or create unnatural separations that will quickly be rejoined again by the platforms re-monopolizing the formerly detached segments. Moreover, the platforms have incentives to make these complex policy exercises all the more difficult by strategically intertwining their various services.

4. UNWINDING PAST MERGERS

Some first concrete and more careful proposals for platform breakups seek to undo recent mergers. John Kwoka and Tommaso Valletti propose breakups along ‘natural fault lines’, which they argue are often, yet not exclusively, along the joints of recently consummated mergers and acquisitions.75 Herbert Hovenkamp, even though highly critical of Big Tech breakups, agrees that if divestitures are deemed necessary, it is best to unwind previous acquisitions, as those would likely do the least damage to network effects and economies of scale.76 The approach seems implied in the US Competition and Transparency in Digital Advertising Act, which intends to prohibit companies that process more than $20 billion in digital ad transactions from participating in more than one part of the ad process.77 The bill would essentially force Alphabet to divest its ad exchange, Google AdX.78 Going further back, it may stretch to unwinding mergers of the past decade, such as Zappos, Instagram, and DoubleClick.79 Similarly, in Google Adtech (2023), the European Commission suggested it may force Alphabet to sell DFP and AdX, both of which grew out of DoubleClick after Google had acquired it.80

All Big Techs are different, however. Their integration processes are long-running, and some transformed into hybrid platforms essentially in opposite directions to others. Microsoft, Alphabet, and Meta, on the one hand, began as intermediaries and then expanded into many commercial ventures. Microsoft soon started collecting its own application software suites to sell on Windows OS. It added other intermediaries, such as LinkedIn and the Xbox gaming console division, for which it also became a seller of complementary products and services, such as games, cloud software, and server space. From 2002, Google diversified into shopping services with the development of Froogle, which became Google Shopping, while riding the tech merger wave that started in the early 2000s with the acquisitions of YouTube in 2006 and DoubleClick in 2007.81 Facebook expanded with the acquisitions of Instagram and WhatsApp and has been growing Facebook Marketplace since 2016, directing trade in services and products there, with even an attempt at launching a cryptocurrency.82

Apple and Amazon, on the other hand, started out essentially as sellers of luxury hardware devices and mail-order books, respectively, and then added an intermediary later for finding apps and all kinds of products. Apple’s iPhone, first introduced in 2007, became the hardware basis for an extensive ecosystem running on iOS. Apple developed App Store in 2008 as the only distributor of mobile apps on its devices, along with Apple Pay as a payment service. It offers its own productions and rights on Apple TV and Apple Music. Amazon invested heavily in a worldwide online warehouse and distribution infrastructure for years and since around 2000 has collected a wide array of complementors that now sell their products on its Amazon Marketplace.83 The company started rolling out its subscription service Amazon Prime in 2005, initially offering enhanced shipping and delivery options, and later entertainment content via Prime Video, Prime Music, and Prime Gaming.

With these growth strategies in opposite direction, trimming back the Big Techs by undoing their recent mergers and acquisitions may be too cautious, if at all effective against information bias. To really get to the root cause of that problem, divestitures probably need to go far back, to where a platform that started out as an intermediary became a hybrid. In the case of Google, that might imply going back almost two decades to the acquisition of DoubleClick. Yet undoing that merger today would hardly rewind the market structure. Also, the core incentive problem arose rather from the growth of Google Shopping, which was not a merger but the result of in-house development. Moreover, a reversal strategy does not apply well to platforms that mostly grew organically, rather than through large acquisitions, like Apple did. Also, more recent and smaller acquisitions are not particularly significant. For example, the deterred acquisition of iRobot by Amazon was problematic because Amazon has incentives to push its own product lines on its Marketplace—and preventing such mergers is not a solution to that central concern.84

In summary, the big structural changes proposed in the USA along roughly determined revenue thresholds, business categories, and past acquisitions, seem rather coarse, whereas the European approach of tailored behavioral remedies may be too case dependent. If structural interventions are called for, it seems appropriate to think more conceptually and precisely about where to aim them with surgical precision, in order to address the information bias problem. That would be ‘smart cuts’.

5. SMART CUTS

A smart cut in a platform company is a bespoke, precisely identified, minimal yet complete separation of the platform’s core advisory function, to insulate it from incentives to bias information, with the least possible damage to positive network effects for users. The goal is to implement the minimally necessary structural measures that ensure the impartiality of a viable recommender system, as we defined it, by eliminating any direct connections between divisions that could potentially influence advice in favour of the company’s commercial interests rather than giving objective, fair, and honest ratings. The smart cut lines are drawn to isolate the body that provides such type of information from the many commercial tentacles of the company resulting from its peripheral services. By creating distinct entities with the sole incentive of providing unbiased information, users are better assured of receiving objective search results, content, and recommendations to base their purchasing decisions on. In the example of Google Maps, in isolation, it would have no business sending its users on any diner detours anymore; the platform’s sole focus is on recommending the optimal route from the user’s perspective, from A to B.

In spirit, the concept of smart cuts closely aligns with existing proposals to not allow Big Tech to both own an intermediary, which is an essential facility for suppliers trying to reach users, and participate in it with commercial transactions as well.85 Smart cuts make this idea with surgical precision. The smart cut separates the recommender system of the platform from participating on the platform with its own commercial activities, intending to sever any commercial ties that may create incentives to bias the information supplied by the recommender system to its users. This bases the intervention on a conceptual assessment of incentives, makes it more precise, and extends it beyond the mere undoing of recent mergers. Naturally, the main challenge is in properly identifying the recommendation function for a clean separation. It should be separated as much as possible from its commercial ties while maintaining its data collection parts as much as possible.

Whether the breakup is best executed with functional or ownership separation is case-dependent. In a functional separation, the resulting two parts of the hybrid platform, namely its recommender system and peripheral services, can remain part of the same umbrella company if firewalled. To prevent any possibility of control, it is essential that the selling parts are kept at arm’s length from the advisory function so they do not influence each other. After all, the primary objective is that, once separated, the recommender system is no longer affected by incentives to provide alternate information services to first- and third parties. If functional separation is sufficient to reach this goal, then there is no need to fully divest the recommender system. However, it may not be possible to guarantee independence with shared ownership, because the autonomous parts of the firm will still be beholden to the parent company, which maintains the possibility of meddling by the platform in the activities of the recommender system. In that case, ownership separation may be needed, whereby the recommender system becomes a fully independent company.86

The key is that by maintaining the complete advisory function of the platform, there continues to be the possibility for maximal network benefits. All users still feed the recommender system with their data. However, once users have been informed by the intermediary about their options and have decided what they want to buy, they go elsewhere to sellers that are independent of the recommender system to make their purchases. Their links, as well as payment options, packaging, and shipping services, are all offered in the search results, but as the best ones for the user, not the ones that suit the sellers. Thus, the indirect sources of income for the hybrid platforms from referrals by the recommender system to its own in-company peripheral services, where sales are subsequently made, fall away. The profits from those own referrals now no longer benefit the spun-off recommender system, which is essential for removing its incentives to introduce information bias.

In order for them to remain sustainable, the isolated recommender system could be allowed to continue to finance itself through the sale of advertising space, along the edges of its organic search results, from demand-side platforms. Crucially then, the advertising function cannot interfere with the advisory function. This can be assured by mandating that the organic search results are determined prior to the offering of advertising positions, so that advertisers cannot influence the order in which they are presented. Google signals to the demand-side platform that a certain number of advertising positions are available, with information only about the user, the query, and the position of the advertisement. Furthermore, paid-for messages must be clearly indicated as advertisements, so that the user can always tell the difference from the organic search results.

In the case of Google Maps, to prevent incentives for detouring riders, these requirements imply that the route is to be set before the advertised locations are put on the map and these are clearly flagged as sponsored venues. Multiple suppliers of advertisements, products, and peripheral services have non-discriminatory access to the platform and compete for these indicated advertising spots in a transparent bidding process. This increases the information value of paid-for messages too. The smart cuts assure that the recommender system has the sole objective of providing objective information, as the best possible user service will ensure a sustained maximum number of eyeballs, which, in turn, makes the function attractive for advertisers to purchase advertising space on—when properly done, in fact, even more attractive.

With its own steady source of income, the recommender system is stable in the sense that it does not need to reintegrate with the type of commercial functions it was separated from—which should be prohibited. Despite its dominance, the recommender system would be careful to refrain from using too much of the user’s screen space for advertisements, or it would lose them, which would reduce the prices of advertising space. There will still be incentives to reintroduce commercial interests into the recommender system. Sellers may try to pervert the setup by offering money for the hidden promotion of certain references, but that would be illegal. Such interference needs to be regulated and prevented through strict oversight. Also, the intermediaries cannot be allowed to expand into hybrid platforms again and must be fully transparent about their sponsors. The key is that commercial interests cannot interfere with the organic search results.

In essence, the target structure is no different from the classic newspaper model. The medium offers independent news articles to a certain readership based on its reputation for relevance and reliability. To protect it, readers are made aware of what is news and what are advertisements. The latter are clearly indicated as such, at the top of the ad, so that the user can assess its information value. Journalists and editors are held to an ethical code and social norms, while the advertising department independently sells selected advertising positions in the newspaper. In fact, this has always been the original advertising model of Google Search, with genuine organic search results, generated entirely independently from any other interests than providing the user with the best matching responses to their query, and designated advertising space at the top or margins of the pages that is auctioned off to the highest bidder. Smart cuts aim to bring back the structure necessary for assuring this clear distinction between objective organic and sponsored information.

Should instead a full non-profit structure be preferred, in which the insulated recommender systems are not allowed to engage in any advertising, alternative funding models will have to be found. These could be based on subscriptions or per-use payments, or public subsidies, which are reminiscent of Senator Warren’s view that recommender systems are essential facilities that should be run as public utilities. Alternative funding may also come from donations, like Wikipedia or The Guardian are funded, or from a membership model. A non-profit model may further enhance public trust in the reliability of the recommender systems. On the other hand, closed systems, such as a membership model, would be at odds with the open nature of most recommender systems and would likely reduce usage and, therefore, impact the valuable network effects. Moreover, these types of sources of income may be too unstable to support those important platform functions, of which the platforms should probably be allowed a substantial share. Therefore, advertising revenues seem the better option for assured viability, provided the separation is clean, which also allow for easier entry of rivals, as we will set out below.

Also, with the clear distinction in place, the remaining elements of the original hybrid platform can remain fully integrated, further divest, or grow other functions, depending on how it or its parts want to position themselves in their market(s). For example, in principle, there would be no problem for Amazon to acquire products such as iRobot, because it would no longer have the ties to Marketplace that would give the company incentives to favour them there. Likewise, an insulated Google Search would have no reason to distort its organic search results for AdX or even prefer the exclusive use of it, so that Alphabet’s advertising division can stay integrated. Similarly, the App Store would shake its incentives to steer users to particular payment providers, including Apple Pay, if set at arm’s length from Apple. With the recommender system isolated, the peripheral services, advertising divisions, and other parts of the platform would have to compete with other companies for the display of positions next to the objective search results. The seller with the best match to a user should float to the top of the rankings, while those with the highest bid would occupy the top advertising spot. In principle, there would be no need, from the point of view of information bias harming consumers, to further police or interfere with mergers and acquisitions, or for behavioural remedies to discipline the parts that are separated from the intermediary. Innovation incentives would be maintained, and scale economies in the sellers’ services, for example, in the payment and handling of purchased items, would be preserved to the benefit of users.

6. SCISSOR-LINE SUGGESTIONS

To illustrate how the idea of smart cuts would apply, we offer some scissor-line suggestions for the gatekeepers Alphabet, Amazon, and Apple, in organigram sketches of these hybrid platforms in three subsections. The companies’ organizations are depicted in square boxes, with the core recommender systems in white and other peripheral services and selling parts with commercial functions in green. Advertising subsidiaries are in yellow. Third parties that interact with the platform are indicated as ‘outside parties’. The demand for information by the users—the users’ search for queries or apps, use of the platform’s services, and search for products—is given by black arrows. The flow of money—payments by advertisers, third-party sellers, and app developers—follows the red arrows. The supply side—the provision of advertising space, physical distribution of products, and app distribution—is shown as blue arrows.

Alphabet

The organigram of Alphabet is sketched in Figure 1. It is an ecosystem of powerful Internet searches and convenient consumption. Users enter Alphabet’s services through Android or Chrome, or third-party browser, like Safari. They have access to a wide array of services, such as Google Search, Google Shopping, Google Maps, YouTube, and Gmail. With its integrated subsidiaries, Alphabet has many feelers for the collection of information—for example, via search queries, user clicks, location tracking, interactive speakers, and wearable health devices.87 It disseminates information in the form of search results, route and video suggestions, for example, and targeted advertising, including for products and services that generate revenue for Alphabet. On its platforms, Alphabet sells keyword advertising, which consists of sponsored positions primarily around the query search results, for which it controls the ad tech chain. The combination of these functionalities mingles Alphabet’s commercial objectives.

The main concerns with Google are search bias and self-preferencing. The key incentive problem that leads to such information distortion in the Alphabet case lies in the fact that the platform mixes the advisory function with the demand for advertising positions, including from its own subsidiaries seeking sales. This makes it profitable to subliminally rank certain commercial links before the better organic ones in Google Search. As a result, the incentive structure is such that Alphabet can benefit from rankings that are manipulated in favour of advertisers or hidden advertisements, which increase the click-through rate, the number of views per advertisement, and the follow-on purchases, quite possibly against the interests of the users. Google Search can do this because it is the dominant platform. It has superior information over its users and lacks competition from any alternative search engine, which is strengthened by its ability to make Search (and also Maps) the default option on Android and Chrome, as well as its deep pockets for deals with other hardware manufacturers and web browsers to make Google Search the default gateway for search on any device. The power over search allows Google to bias its recommender systems for money to a certain extent, without losing too many eyeballs.88

The smart cut suggestion is to insulate the recommender systems Google Search, Google Maps, and YouTube. Together, these three parts form Alphabet’s core advisory function, which could be a viable separate independent company. The scissor lines are clear. They would cut the monetization strategy based on the flow of eyeballs to Google’s subsidiaries and preferred advertisers. All three parts with a recommender function have been characterized as core platform services in the meaning of the DMA; they have many users and few close competitors. After the cut, Google Search just lists search results solely based on a continuously developing search algorithm that indexes the web. YouTube presents the most interesting videos without prejudice to their order. Google Maps provides location-based navigation services for the best available route.

The cutting is relatively straightforward since these three core divisions consist of relatively small teams, of no more than a couple of thousand people each, with dedicated tasks.89 The combined market share of competing search engines like Bing and Yahoo is negligible. In the creator video market, YouTube faces little competition, essentially only from Hulu and Vimeo. In the broader video streaming market, Netflix, Amazon Prime, and Disney+ are competitors—with Netflix being the closest, with 260 million active users, compared to approximately 2 billion on YouTube.90 The suggested separation would preserve the data-driven network effects generated within each recommender system, which can sell its advertising space. With Google Search as a separate entity, the ‘searchless’ Alphabet no longer has the ability to prioritize its subsidiaries, such as Google Shopping, over external competitors in the search results. Also, Chrome and Android would lose their incentives to present Google Search as the default option. The new search engine would not have any particular interest in favouring Alphabet’s DFP or AdX either and would instead be interested in seeing entry into the online advertising technology market. By breaking up the conglomerate, that is, the potential for anti-competitive practices would be curtailed, fostering a more open and competitive landscape. All the other parts of the Alphabet conglomerate can remain integrated.

Amazon Marketplace

Figure2 depicts Amazon, a customer-centric hybrid sales platform that facilitates hassle-free shopping. Its integrations create many benefits, such as on-platform competition between a wide variety of products and seamlessly integrated services.91 Here, the main contention is that Amazon gives prominent placement to certain sellers in the Buy Box, which then determines the purchasing decision of consumers. Sellers subscribe to Amazon Marketplace, and their products are listed, paying Amazon an average of about 15 percent of the products’ selling price in addition to a fixed fee. Sellers can choose to use various additional Amazon services as well, such as Amazon’s storage and distribution network, for which they pay fees—known as ‘fulfillment by Amazon’ (FBA).

The key information bias derives from Amazon having incentives to advance certain products over others. Amazon also acts as a retailer on its own Marketplace, selling its own first-party and private-label products while competing with third parties for placement in the Buy Box. Through Amazon Prime, it also offers a premium delivery and payment option, as well as video streaming, gaming, and music services. Users who subscribe to Amazon Prime have an incentive to use Amazon for all of their online shopping, which reduces their tendency to purchase via other (online) channels. Amazon balances an indirect network effect between buyers and third-party sellers while being able to use its own sellers to influence (pricing and exit) behaviour on the platform. One allegation is that if sellers make use of FBA, Amazon may display them in the Buy Box more than sellers who opt to do their own distribution, as Amazon treats FBA as a product quality aspect. In addition, when Amazon detects that certain products from third-party sellers sell well on its platform, the company has an incentive to enter these product markets and start selling similar products.92 This strengthens Amazon’s abilities and incentives to manipulate the rankings in favour of its own sellers, putting them in the Buy Box rather than the buyer’s true best match. It also harms third-party sellers and buyers because the probability of a sales match decreases, despite their product being the best fit for the consumer.

The smart cut suggestion isolates Amazon Marketplace as an independent market-maker, so that it becomes an insulated recommender system where suppliers and consumers are brought together.93 It would sever the connection between the Marketplace and Amazon’s proprietary sellers, so that Amazon would no longer have incentives to favour its first- over third-party sellers. Amazon Marketplace is a designated core platform service under the DMA. It should be viable on advertising revenues alone and no longer have hidden incentives to push for certain deals over others, because some commercial arm of Amazon happens to benefit from that deal. In particular, due to the separation, dealing with non-Amazon service providers would not affect the possibility of being recommended in the Buy Box. In addition, if broken up as suggested, Amazon no longer has an incentive to use information obtained from trades on its Marketplace—as it does on best-selling items—to push its own version of such high-demand products, to the point of even suppressing the vendors of the original best-selling item so that only Amazon’s own version is available.

The network externalities on Marketplace remain. A smart cut in Amazon should also be done with the caution necessary to preserve its benefits in integrated shopping and shipping. Yet wide product variety on offers, competition in the marketplace, and secure and fast delivery can all still be assured through independent links. Moreover, the remaining integrated Amazon structures can stay intact. A buyer continues, for example, after having chosen her best purchase, to be able to opt for Amazon Prime’s high-quality shipping service, since the new Amazon can continue to offer discounted bundles of Prime and Fulfillment. The key benefit is that now also other competing shipping options are available, such as FedEx, UPS, and DHL, all with their distribution offers tailored for the consumer and treated equally by the insulated Marketplace recommender system. Third-party distributors are free to also offer an enhanced shopping experience, be it with integrated video, gaming, or music services, or something completely different. Hence, we may see competition between same-day delivery by FBA/Prime or UPS/Netflix, or sustainable delivery by FedEx/Disney—or FedEx/Starbucks, for that matter, or any attractive bundle—all presenting slightly different offers to consumers that compete on price and different non-price dimensions. FBA/Prime may well continue to be the best option because of scale economies, but buyers and sellers would now have other options. To facilitate competing services, the new independent marketplace should satisfy certain open access requirements. Likewise, third-party sellers can choose among different analytics, brand development, and ad services. Hence, a great benefit of smart cuts is that other parties can enter and supply these services too, in competition.

Apple App Store

Finally, consider the thought experiment applied in Figure 3 to the structure of Apple, the producer of hardware devices with its user-friendly online ecosystem, which is designed to keep consumers inside—including by creating switching costs, such as the need to buy apps anew for any other platform. The App Store is the recommender system here: it ranks apps and advises users on their purchases, thus playing a central role in the company’s profit strategy. New apps are developed on Apple’s app development software. Before allowing a new app into the App Store, Apple checks it for safety and security. For these services, Apple’s fee is between 15 and 30 percent of the app’s purchasing or subscription price, as well as all in-app purchases that a user may make while using the app, which it collects through the obligatory use of Apple’s payment system. Apple has always disallowed the installation of apps from outside the App Store (so-called ‘side-loading’) and the use of alternative payment service providers, allegedly for security reasons. Yet, in order to comply with the DMA, the recent iOS 17.4 makes side-loading possible in the EU.94 This, together with Apple’s commitment to allow alternative payment services, opens up the ecosystem somewhat. 95 Still, the closed integration between the App Store and the other Apple services continues to give Apple distorted incentives to push certain software, and with it, certain purchases over others. Also, since the update, Apple levies a ‘core technology fee’ of €0.50 per installed app for designated apps and third-party app stores, for which it remains under investigation.96

Our suggestion for a smart cut is to divest the App Store into an independent entity that would process app demand and supply. Apple’s App Store is a designated core platform service under the DMA. Divesting it would cut the money flow from Apple apps to the App Store so that the store would no longer have incentives to disadvantage third-party app developers. Also, Apple would no longer have incentives to push the App Store as the default app distributor on its operating systems. There would be no disruption in the overall functionality of the platform, as indicated by Apple’s apparent ability to break up the European App Store in order to comply with DMA obligations that allow competing ways to buy and install apps on iOS.97 Network benefits would remain. The App Store alone is equally viable from the commissions of the apps it admits, without discrimination. In addition, it could sell identified advertisement space around its core functions, such as app distribution and rankings, with users who prefer to be shown advertisements over paying for their apps. Apple would no longer have commercial incentives to favour certain apps over others that users would prefer. Apple can retain all peripheral app store services, such as Xcode, the integrated development environment for iOS app development, and Apple’s Software Development Kit. It should continue to inspect all apps being installed on iOS for safety and security, for which the Apple ecosystem is renowned and appreciated. However, interoperability with iOS also allows other independent app stores to develop and offer software on Apple devices. This should create price competition so that fees will be lower.

7. ADVANTAGES OF SMART CUTS

The smart cuts approach has several advantages beyond its core objective of eliminating information bias while preserving network benefits. The concept suggests scissor lines between the intermediary and the businesses in the marketplace, reminiscent of earlier breakup proposals by Lina Khan and Senator Warren. Current breakup ideas are not necessarily ‘smart’, however, in that they do not originate from the ideal end structure. The FTC seeks a divestiture of Instagram and WhatsApp, not Facebook Marketplace.98 In Google Adtech (2023), the Commission considers a mandatory divestment of DFP and AdX, presumably allowing Alphabet to keep Google Search integrated with the rest of the company. This may break Alphabet’s stronghold over online advertising, which may invite entry into the advertisers’ software market and bring down marketing costs, possibly also lowering consumer prices. But Google Search would still be the dominant publisher, dictating the prices for advertising space. This kind of proposed breakup does not address the incentive structure that induces information bias by the search engine either.99 Setting Google Search apart, on the other hand, takes away its incentive to exclusively deal with AdX or Google Ads as well, as those will then become outside entities and profit streams. Similarly, the behavioural remedy in Google Shopping (2017), to set Google Shopping apart within Alphabet, did cut Google Search’s incentive to push its own shopping site, but it did not address any other information bias in search. It would have been smarter to cut out Google Search instead. The same is true for the DoJ’s suggestion that Alphabet spins off Chrome and Android.100 While this may reduce Alphabet’s ability to push Google Search as the default search engine and avoids the destruction of network benefits, it does not address the power of Google Search to bias its information in favor of advertisers. Likewise, neither unwinding past mergers, such as Google/DoubleClick or Amazon/Zappos/Whole Foods, nor blocking mergers, such as Facebook/Giphy, Amazon/iRobot, Booking.com/eTraveli, or Microsoft/Activision Blizzard, seem to fully get to the root cause of the problem, despite creating competition in some places.

Insulating the recommender system as we propose essentially makes it available to all peripheral services, including those separated under the parent company umbrella, where commercial transactions can be executed. Hence, it creates a level playing field that stimulates competition on the recommender system, so that users obtain the rents of network benefits, which smart cuts aim to preserve. In fact, without biases in favour of specific services on their platforms, recommender systems have an incentive to make sure that they are interoperable with as many third-party services as possible—rather than circumvent interoperability requirements with creative compliance.101 This makes their platform more attractive and leads to downstream competition. The more competing services and sellers they can offer on their recommendation platform, the more users the recommendation platform will attract, and the more eyeballs to present with advertisements. Smart cuts thus open up the dominant platforms’ ecosystems for competition around their recommendation systems.

Note that smart cuts do not directly address the market power of the core platform itself. Google Search, Amazon Marketplace, and the Apple App Store remain the dominant search platforms of choice—which they are due to their far better platform benefits than any smaller recommender system can offer. They should be expected to make good profit margins—as long as there are no deeper interventions, such as the nationalization of online network infrastructures, like Google Search, which insulation of the recommender systems as we suggest would facilitate.102 Neither do smart cuts necessarily address other social concerns, other than information bias, such as excessive pricing, including in the form of collecting large amounts of personal data to sell for targeted advertising, or lack of privacy. The data that the recommender system collects from its users can still be used to improve its (now truly) organic search results. After all, the gathering and processing of big data is what generates the network benefits—such as knowing which traffic jams to avoid. To that end, the recommender system can also be allowed to keep its feelers, such as Google Nest, Amazon Assistant, or Apple Watch, as long as they do not introduce commercial interests. A properly insulated recommender system is better, the more data it combines.

Note that the smart cuts approach itself, without further regulatory notice, does not require that the platform refrain from using the personal data it collected to inform advertisers directly either. Provided that the advisory function is shielded properly from the advertising function, selling advertising space around its organically generated rankings does not bias the intermediary itself. The recommender system is sustainable by selling advertisement space, yet now without preferential treatment. Likewise, smart cuts alone do not close off rabbit holes and echo chambers for which users have a taste to enter, nor stop fake news. It is primarily about taking away commercial incentives to promulgate such sources as information bias—which, of course, they are not if reinforced illusions happen to be the first best matches for a particular user’s query. But with well-executed smart cuts, at least no one should get sucked into false beliefs because they lead to certain commercial transactions for the recommender system, against the true preferences of the user.

Of course, our model can easily be complemented with stricter privacy regulations on what a platform may or may not collect and do with personal data, or information quality controls. Isolated, the recommender system can more easily be targeted and held to such flanking regulations for privacy and other concerns. Constraining the use of user information would probably reduce the strength of network benefits, but the trade-off would be more transparent. Likewise, should any social desire to address the dominance of the Big Tech recommender systems in intermediation be separately targeted, for example, with abuse control. For our primary objective to prevent search bias and obtain the best results, however, clean incentives and full transparency about what are sponsored links and what is truly organic advice suffice.

Moreover, if complemented with data portability requirements, isolating the core recommender system with smart cuts also helps to facilitate more competition between recommender systems. These would mostly need to be forced onto the recommender systems, as data portability fosters switching and therefore directly threatens their dominance. Data portability would allow users to more easily switch from one recommender system to another, taking with them their data, profiles, and previous software purchases. Once users can be more easily attracted away from the Big Techs, new entrant recommender systems will get the opportunity to more quickly build up a critical mass of users and offer matching platform benefits. This in turn disciplines the incumbents to improve their service, including possibly by offering more privacy or preventing the dissemination of misinformation. Allowing multiple app stores on iOS, for example, likely makes it easier for consumers to switch between different software providers. For instance, a user who buys an app on the Samsung app store on Android, with that app store also running on iOS, can install this app on an Apple device as well, and vice versa. If it is no longer necessary to repurchase apps in the case of a switch to a different platform, users can more easily be lured out of ecosystems, which in turn stimulates platform competition. The data portability requirements necessary for this are more easily imposed on isolated recommender systems than complete hybrid platforms.103

Lastly, smart cuts can greatly alleviate the workload on competition law enforcement agencies. When properly restructured, the platforms’ incentives to engage in many of the anti-competitive strategies they have displayed in the last decades will be taken away, and with it the need for minute control of their behaviours or acquisitions. After all, the competition concerns in antitrust cases against the Big Techs have mostly been about anti-competitive ways in which they have tried to protect and grow their ecosystems in order to keep users within. These behaviours are discouraged by insulating recommender systems. It would simply be forbidden for them to re-extend into hybrid businesses. The remaining parts of the Big Techs would no longer have the platform monopoly basis on which to build a plethora of peripheral commercial products and services. Which ones they do choose to acquire is much less of a concern for competition authorities. Without Marketplace, for example, Amazon can safely be allowed to integrate with new products and services. Such acquisitions would no longer carry the concerns of adding to the information bias problem. Regulators would not need to be concerned with every detail of Amazon’s business strategies outside its advisory functions. Essentially, only strict adherence to the separation needs to be monitored, including the code that advertising space is clearly marked as such. This is relatively straightforward to do, certainly compared to controlling behavioural requirements such as the prohibition of self-preferencing. Structural separation can thus help to reduce regulatory red tape and relieve the burden of competition authorities and regulators. Once properly trimmed, plucked, and shaved, the platforms should be controllable with relatively light oversight.

8. CONCLUDING REMARKS

If there are going to be structural interventions in the Big Tech business models, we propose smart cuts: precise insertions to insulate the core platform functions with minimal damage to positive network effects. To address information bias, the recommender system is separated from peripheral services that give it commercial interests in the non-digital economy with incentives to direct online traffic and sales there. In addition, this structural intervention allows for focused complementary further regulation, if so desired. Setting apart Google Search, Amazon Marketplace, and the App Store are examples for which the scissor lines are relatively clear-cut. Yet, the concept of smart cuts can be extended to other platforms in which an intermediary operates under the same umbrella company as peripheral firms. Other potential smart cuts would avoid, for example, Booking owning both meta-search platforms and online travel agencies, as well as having stakes in hotels; Netflix owning video producers besides the streaming platform; or Spotify recommending and producing podcasts. Smart cuts could also be applied to other platforms, such as the social media website X/Twitter, which was accused of bluntly altering the algorithm and artificially boosting the owner’s posts by a factor of 1,000, thereby also promoting affiliated companies such as Tesla and SpaceX.104 Or to ByteDance, of which TikTok’s video recommendation system is rumoured to induce addiction, with user retention as the ultimate goal. While one should, of course, always be cautious about the specifics of the market and the business model, which may be importantly different, these practices appear similar to those prosecuted.

It is essential that the isolated recommender system continues to generate positive network externalities, as it still brings all users together. Scale economies in the other parts of the company, which, in the case of smart cuts, can stay fully integrated, also carry over. Acquisitions of new products and service divisions no longer need to raise self-preferencing concerns. The insulated recommender system makes its money from selling designated advertisement space or requesting a fee from apps or sellers that are listed. This incentive structure induces the recommender system to offer objective advice in order to attract and keep the largest number of eyeballs and be attractive to third parties. Hence, users are given objective, uncompromised organic search results, content, and recommendations on which they can rely for their purchasing decisions.

Smart cuts have a conceptual basis and may go beyond unwinding recent acquisitions; they may open the ecosystem structures up for competition on information platforms and, between them, can alleviate competition law enforcement. They are more than divesting Instagram, AdX, or Chrome. Of course, to obtain the benefits sketched, it is essential to cut with care, so as not to cut up the positive network effects. This will be a specialist job, as the parts of the hybrid platforms that we suggest for a smart cut can be heavily intertwined with the peripheral services, and may therefore be hard to disentangle. Making mistakes can be costly, and caution should therefore be observed. The adage ‘measure twice, cut once’ applies.

Smart cuts are the axe to the root of the information bias problem. Once properly executed, the approach we have laid out should create a healthy structure for digital markets that does away with much of the need for antitrust enforcement. Competition cases have been a great rapid-response first-aid care for the mounting market power problems in information technology, but they are not the structural first-best long-term cure for the tendency of digital markets to grow dominant gatekeeper platforms. As a series of tease punches in a sparring match, the competition cases have probed the tech platforms, so as to find out how to handle their novel business strategies. Now, if the Big Tech business models need structural intervention, we suggest clean surgery to isolate the core of the ecosystem, insulate it from information bias incentives, allow all services to compete on it, and force an open framework to allow competition between information platforms. This would reduce the risk of overburdening the competition law enforcement system, which needs to keep its capacity to identify and pursue the next morphology of competition concerns. Even though smart cuts are first aimed at providing consumers with objective information, ultimately they will also benefit third-party sellers, advertisers, and market oversight.

2

Meta, Alphabet, Microsoft, ByteDance, Booking, Apple, and Amazon are designated as gatekeepers under the DMA. See Commission, ‘Commission Designates Booking as a Gatekeeper and Opens a Market Investigation into X’ (Press Release, IP/24/2561). X’s rebuttal is still under investigation. See Commission, ibid. Netflix and Spotify are specialized platforms with fast growing user bases and own productions.

3

See, eg, DS Evans, Platform Economics: Essays on Multi-Sided Businesses (Competition Policy International 2011).

4

This is stressed in F Marty and T Warin, ‘Multi-sided Platforms and Innovation: A Competition Law Perspective’ (2022) 27 Competition & Change 184.

5

Seminal analyses of this self-enforcing effect are given in B Caillaud and B Jullien, ‘Chicken & Egg: Competition among Intermediation Service Providers’ (2003) 34 The RAND Journal of Economics 309; F Zhu and M Iansiti, ‘Why Some Platforms Thrive and Others Don’t’ (2019) Jan-Feb Harvard Business Review 118 <https://hbr.org/2019/01/why-some-platforms-thrive-and-others-dont>; J-C Rochet and J Tirole, ‘Platform Competition in Two-Sided Markets’ (2003) 1 Journal of the European Economic Association 990. For an accessible survey, see P Belleflamme and N Neysen, Platform Strategies: A Guidebook for Entrepreneurs in the Platform Economy (Routledge 2023).

6

See J Prüfer and C Schottmüller, ‘Competing with Big Data’ (2021) 69 Journal of Industrial Economics 967.

7

See C Wu and others, ‘The Economic Value of Online Reviews’ (2015) 34 Marketing Science 739.

8

See M Peitz and P Belleflamme, The Economics of Platforms: Concepts and Strategy (Cambridge University Press 2021).

9

See JPH Dubé, GJ Hitsch and PK Chintagunta, ‘Tipping and Concentration in Markets with Indirect Network Effects’ (2010) 29 Marketing Science 216; A Pavan, B Jullien and M Rysman, ‘Two-sided Markets, Pricing, and Network Effects’ (2021) CEPR Discussion Paper, DP16480.

11

See LM Khan, ‘The Separation of Platforms and Commerce’ (2019) 119 Columbia Law Review 973.

12

For an overview of the literature of biased recommendations, see E Calvano and M Polo, ‘Market Power, Competition and Innovation in Digital Markets: A Survey’ (2021) 54 Information Economics and Policy 100.

13

We use the following taxonomy for the various parts of the platform. A hybrid platform describes the complete integrated business, composed of recommender systems and peripheral services. The recommender systems are the parts of the (hybrid) platform that provide primary information to users, including search results, rankings, and location-based information. These latter functions are designated in the DMA as core platform services in the categories of Intermediation, Video sharing, and Search. However, not every core platform activity is a recommender system, because we exclude services such as ads, browsers, and N-IICS. We refer to these services as peripheral services, which are all other services offered by the platforms, such as first-party sellers, distribution services, and advertising services.

14

See IM Koning, RJJM van den Eijnden and HGM Vossen, ‘From Greenwashing to Screenwashing? How the Tech Industry Plays Around with Children’s Future’ (2024) 13 Journal of Behavioral Addictions 1.

15

See M Schaake, The Tech Coup: How to Save Democracy from Silicon Valley (Princeton University Press 2024).

17

Google (2024): [United States v Google LLC, No 20-cv-3010 (D.D.C. filed 24 January 2024)] and [United States v Google LLC, No 1:20-cv-03010-APM (D.D.C. filed 20 October 2024)].

18

Microsoft (1994): [United States v Microsoft Corp, 253 F.3d 34 (D.C. Cir 2001)]; Apple (2024a): [United States v Apple Inc, No 2:24-cv-04055, (C.D. Cal. 21 March 2024)].

19

Microsoft (2007): [Microsoft Corp v Commission of the European Communities (Judgment of the Court of First Instance (Grand Chamber) of 17 September 2007) Case T-201/04, ECLI:EU:T:2007:289, [2007] ECR II-03601.]; Google Shopping (2017): Commission, ‘Antitrust: Commission fines Google €2.42 Billion for Abusing Dominance as Search Engine by Giving Illegal Advantage to Own Comparison Shopping Service’ (Press Release, IP/17/1784); Google Android (2018): Commission, ‘Antitrust: Commission Fines Google €4.34 Billion for Illegal Practices regarding Android Mobile Devices to Strengthen Dominance of Google’s Search Engine’ (Press Release, IP/18/4581); Apple (2024b): Commission, ‘Commission Fines Apple over €1.8 Billion over Abusive App Store Rules for Music Streaming Providers’ (Press Release, IP/24/1161).

20

See Apple (2024c): Commission, ‘Commission accepts Commitments by Apple Opening Access to ‘tap and go’ Technology on iPhones’ (Press Release, IP/24/3706).

21

See Commission, ‘Commission Seeks Feedback on Draft Antitrust Guidelines on Exclusionary Abuses’ (Press Release, IP/24/3623).

22

For an in-depth discussion of data portability and interoperability, see J Crémer, Y-A De Montjoye and H Schweitzer, Competition Policy for the Digital Era (Publications Office of the European Union 2019).

23

See Commission, ‘Commission Sends Preliminary Findings to Apple and Opens Additional Non-compliance Investigations Against Apple under the Digital Markets Act’ (Press Release, IP/24/3433).

24

See Commission, ‘Commission opens non-compliance investigations against Alphabet, Apple and Meta under the Digital Markets Act’ (Press Release, IP/24/1689).

25

See Commission (n 2).

28

See LM Khan, ‘Amazon’s Antitrust Paradox’ (2017) 126 The Yale Law Journal 710; LM Khan, ‘The Separation of Platforms and Commerce’ (2019) 119 Columbia Law Review 973; T Wu, The Curse of Bigness: Antitrust in the New Gilded Age (Columbia Global Reports 2018); R Van Loo, ‘In Defense of Breakups: Administering a “Radical” Remedy’ (2020) 105 Cornell Law Review 1955. See also S Galloway, ‘It’s Time: Break Up Big Tech’ (YouTube, 21 December 2017) <https://www.youtube.com/live/1ebKI4x_k8A?app=desktop> accessed 8 November 2024.

33

See T Knapstad, ‘Breakups of Digital Gatekeepers under the Digital Markets Act: Three Strikes and You’re Out?’ (2023) 14 Journal of European Competition Law & Practice 394.

34

See Commission, ‘Antitrust: Commission sends Statement of Objections to Google over Abusive Practices in Online Advertising Technology’ (Press Release, IP/23/3207). The US Department of Justice also prosecutes Google for monopolization of the ad exchanges market, but without asking for structural remedies (yet). See US Department of Justice, ‘Justice Department Sues Google for Monopolizing Digital Advertising Technologies’ (U.S. Department of Justice, 24 January 2023) <https://www.justice.gov/opa/pr/justice-department-sues-google-monopolizing-digital-advertising-technologies> accessed 16 October 2024.

36

Possibly, this article makes concrete what may have been meant by ‘smart cuts’ in the ABBA parody ‘Breaking up is Never Easy’ on YouTube at https://www.youtube.com/watch?v=HWWMO5fjpUs.

39

See S Brin and L Page, ‘The Anatomy of a Large-Scale Hypertextual Web Search Engine’ (1998) 30 Computer Networks and ISDN Systems 107.

40

Google’s stakes in the restaurants that create distortion incentives may be direct commissions, but can also be more indirect. For example, Google Maps may display restaurants that use Google Maps to process orders, or that gather a large number of good reviews within its review system or use Google’s reservations system more prominently. For all we know, Google Maps may be selling traffic diversions away from wealthy neighborhoods, where residents are willing to pay for quiet streets. See also B Barrett, ‘An Artist Used 99 Phones to Fake a Google Maps Traffic Jam’ (Wired, 3 February 2020) <https://www.wired.com/story/99-phones-fake-google-maps-traffic-jam/> accessed 15 October 2024.

42

Behavioral remedies were imposed in December 2022. See Commission, ‘Antitrust: Commission Accepts Commitments by Amazon Barring it from using Marketplace Seller Data, and Ensuring Equal Access to Buy Box and Prime’ (Press Release, IP/22/7777).

45

See Commission (n 19).

47

See Commission, ‘Commission sends Statement of Objections to Meta over Abusive Practices Benefiting Facebook Marketplace’ (Press Release, IP/22/7728).

53

U.S. House of Representatives, Committee on the Judiciary, Investigation of Competition in Digital Markets, House Prints 117–8 (Government Publishing Office 2020). See <https://democrats-judiciary.house.gov/issues/issue/?IssueID=14921> for the agenda of the hearings. Video recordings of the full testimonies on 29 July 2020 is available at <https://youtu.be/WBFDQvIrWYM> accessed 12 September 2024.

59

See PR Enia, ‘A Continental Rift? The United States and European Union’s Contrasting Approaches to Regulating the Monopolistic Behavior of Gatekeeper Platforms’ (2022) 16 Brooklyn Journal of Corporate, Financial & Commercial Law 249; C Zakrzewski, ‘Klobuchar and Cotton Introduce Legislation to Regulate Big Tech Acquisitions’ The Washington Post (Washington D.C., United States, 5 November 2021) <https://www.washingtonpost.com/technology/2021/11/05/klobuchar-cotton-tech-competition/> accessed 16 October 2024.

61

J Kwoka and T Valletti, ‘Unscrambling the Eggs: Breaking up Consummated Mergers and Dominant Firms’ (2019:1295) 30 Industrial and Corporate Change 1286.

62

See M Usman, ‘Breaking Up Big Tech: Lessons from AT&T’ (2021) 170 University of Pennsylvania Law Review 523. The author does point at the difference between separating physical and intangible assets, noting that the latter is more difficult.

63

See M Watzinger and M Schnitzer, ‘The Breakup of the Bell System and its Impact on US Innovation’ (2022) CEPR Discussion Paper, DP17635.

64

F Poege, ‘Competition and Innovation: The Breakup of IG Farben’ (2022) Boston University School of Law Research Paper Series, No 22–24.

66

See for some of their discussions, All-in Podcast <https://allin.com/episodes>, #192, from 5:28 and #36 from 7:02.

67

See respectively, F Etro, ‘Product Selection in Online Marketplaces’ (2021) 30 Journal of Economics & Management Strategy 614; Pavan, Jullien and Rysman (n 9); F Zhu and Q Liu, ‘Competing with Complementors: An Empirical Look at Amazon.com’ (2018) 39 Strategic Management Journal 2618.

70

EM Fox and DI Baker, ‘Antitrust and Big Tech Breakups: Piercing the Popular Myths by Cautious Inquiry’ (2021) 2 Competition Policy International.

72

A Hagiu, T-H Teh and J Wright, ‘Should Platforms be Allowed to Sell on their Own Marketplaces?’ (2022) 53 The RAND Journal of Economics 297.

73

C Shapiro, ‘Antitrust: What Went Wrong and How to Fix It’ (2021:42) 34 Antitrust 33.

74

J Tirole, ‘Competition and the Industrial Challenge for the Digital Age’ (2023) 15 Annual Review of Economics 573. See A Schrager, ‘A Nobel-winning Economist’s Guide to Taming Tech Monopolies’ Quartz (New York, United States, 27 June 2018) <https://qz.com/1310266/nobel-winning-economist-jean-tirole-on-how-to-regulate-tech-monopolies> accessed 15 October 2024. See also J Tirole, ‘Competition and Industrial Policy in the 21st Century’ (2024) 3 Oxford Open Economics i983.

75

See Kwoka and Valletti (n 61). The comparative advantage of structural over behavioral remedies in light of incentive effects are advocated also in J Kwoka, ‘Merger Remedies: An Incentives/Constraints Framework’ (2017) 62 The Antitrust Bulletin 367. The authors follow up with examples of post-merger enforcement cases in which structural separation led to viable firms, although with an uncertain impact on competition in J Kwoka and T Valletti, ‘Confronting Consummated Mergers: An Inquiry into Policy and Practice’ (2024) Working Paper.

76

H Hovenkamp, ‘Structural Antitrust Relief Against Digital Platforms’ (2024) 7 Journal of Law & Innovation 58.

82

Facebook’s currency was launched as Libra in 2020 and was shut down named Diem in 2022. See BBC, ‘Facebook-funded Cryptocurrency Diem Winds Down’ BBC (London, 1 February 2022) <https://www.bbc.com/news/technology-60156682> accessed 16 October 2024.

85

As articulated in particular in Khan (2017) and by Senator Warren; Oscar Borgogno and Giuseppe Colangelo argue that the broad set of (proposed) behavioural interventions for app stores in the DMA, such as the neutrality obligations, essentially turn them into public utilities. See Khan (n 28); Warren (n 51); O Borgogno and G Colangelo, ‘Platform and Device Neutrality Regime: The Transatlantic New Competition Rulebook for App Stores?’ (2022) TTLF Working Papers No 83.

86

On the difficulties of monitoring the efficacy of firewalls, see Khan (n 11) 1079–1080.

89

See Google CEO Sundar Pichai testifies before the House Judiciary Committee on ‘Transparency and Accountability: Examining Google and its Data Collection, Use and Filtering Practices’. (YouTube, 11 December 2018) <https://www.youtube.com/watch?v=prdxra7B5H4&t=11202s> accessed 28 October 2024 at 3:07:00.

91

Zhu and Liu (n 66); A Hagiu, T-H Teh and J Wright, ‘Should Platforms be Allowed to Sell on Their Own Marketplaces?’ (2022) 53 The RAND Journal of Economics 297; F Etro, ‘Product Selection in Online Marketplaces’ (2021) 30 Journal of Economics & Management Strategy 614.

92

Numerous allegations against Amazon deploying this strategy are presented in the 2020 US House Antitrust Hearing, in which it was said that competitors referred to Amazon as ‘a drug dealer’ for third-party sellers, because ‘you find out that this person who is seemingly benefiting you, making you feel good was just ultimately going to be your downfall’. See CEOs of Facebook, Amazon, Google, and Apple face Congress in antitrust hearing (YouTube, 29 July 2020) <https://www.youtube.com/watch?v=T0gJYFX8WVc> accessed 28 October 2024 at 3:41:40.

93

See also Mitchell and Knox (n 51).

95

See Commission (n 20).

96

Apple is the first gatekeeper charged by the European Commission for non-compliance with the DMA. See Commission (n 23).

99

In addition, it would introduce a need for micromanagement of online advertising markets, which are highly volatile, by regulatory agencies that may not be equipped for that task. See T Cowen, ‘European Regulatory Transformation-A Case Study: Competition, Remedies, and Google’ (2024) 12 Journal of Antitrust Enforcement 213.

100

United States of America et al v Google LLC (Plaintiffs’ Proposed Remedy Framework, Case Nos 1:20-cv-03010-APM and 1:20-cv-03715-APM, DC Dist Ct, 2020).

101

Eg, when Apple implemented a technical solution to allow Android users to send messages to iMessage, Android users’ messages were displayed in green text bubbles, in-between the blue messages of Apple users. Quickly, the green messages were seen as ‘uncool’ and led to bullying behaviour as much among teens as professional basketball players. See T Higgins, ‘Why Apple’s iMessage Is Winning: Teens Dread the Green Text Bubble’ The Wall Street Journal (New York City, United States, 8 January 2022) <https://www.wsj.com/articles/why-apples-imessage-is-winning-teens-dread-the-green-text-bubble-11641618009> accessed 15 October 2024; A Krishnamurthy, ‘Jarrett Allen Reveals He Had To Get An iPhone Because The Cavaliers Wouldn’t Let Him Be Part Of The Group Text: "They Wanted All Blue Messages."’ (Yardbarker, 24 February 2022) <https://www.yardbarker.com/nba/articles/jarrett_allen_reveals_he_had_to_get_an_iphone_because_the_cavaliers_wouldnt_let_him_be_part_of_the_group_text_they_wanted_all_blue_messages/s1_16751_37244484> accessed 15 October 2024. Due to the peer pressure, more users switched to the dominant platform. In another example, when automakers in Massachusetts were forced to allow data sharing with independent mechanics in order to break their monopoly, they quickly redesigned their cars with wireless technology that was not covered by the law. See C Doctorow, ‘Tech Monopolies and the Insufficient Necessity of Interoperability’ Medium (25 May 2021) <https://onezero.medium.com/tech-monopolies-and-the-insufficient-necessity-of-interoperability-aafba94f1eb3> accessed 15 October 2024. For a story of creative compliance in the case of Google Shopping (2017), see P Marsden, ‘Google Shopping for the Empress’s New Clothes–When a Remedy Isn’t a Remedy (and How to Fix it)’ (2020) 11 Journal of European Competition Law & Practice 553.

102

This is part of the concerns of the European Parliament, formulated by MEP Paul Tang as: ‘The commission should take direct aim and at least dismantle the perverse business model of these tech giants: monetising personal data via advertising.’ See J Espinoza, ‘EU Warns that it may Break up Big Tech Companies’ Financial Times (London, 15 December 2020) <https://www.ft.com/content/15bf2e24-284f-4819-89ff-2520676e29ed> accessed 15 October 2024.

103

Note that subsequently, once there is actual competition between information platforms, privacy conditions can improve too, as new entrants can commit to a subscription with a fee that does not collect personal data. Under pressure of government intervention, Facebook reluctantly introduced paid for versions of Facebook and Instagram that shows less advertisement—and promises not to collect information on its paid-subscription users. See Meta, ‘Facebook and Instagram to Offer Subscription for No Ads in Europe’ Meta Newsroom (Menlo Park, United States, 30 October 2023) <https://about.fb.com/news/2023/10/facebook-and-instagram-to-offer-subscription-for-no-ads-in-europe/> accessed 15 October 2024.

Acknowledgements

We thank John Kwoka, Nicholas Shaxson, Marleen Mitzman-Wessel, Sander Onderstal, Sean Ennis, Ariel Ezrachi, and an anonymous reviewer and participants of EARIE 2024 and CLEEN 2024 for useful comments and discussions that helped us sharpen our argument. All opinions and any remaining errors are our own.

© The Author(s) 2024. Published by Oxford University Press.