Based on the upcoming book
Behemoth, Amazon Rising: Power and Seduction in the Age of Amazon
Amazon is the Behemoth. It dominates key parts of the economy like ecommerce and internet. Yet it’s still growing at an astounding rate – it will be the biggest company in America within a few years. And its tentacles are everywhere – online retail, groceries, books, publishing, logistics, satellites, security, adverting , entertainment, health care, even banking.
This explosive expansion reflects Amazon’s brilliant success for consumers. Usually ranked the most trusted brand in America (and often in the world), Amazon has transformed consumption. A single click of a mouse brings a thump on the doorstep next day. Subscriptions mean consumers don’t even have to actually order to receive their goods. And Amazon is the Everything Store; because of the Amazon Marketplace, goods from other sellers let it Amazon offer 350 million different items. Need a hardback, or a handsaw, or some halvah? It’s all there.
Beyond that, Amazon is the ardent suitor constantly showing just how much it loves you (assuming that you are like the vast majority of American households a member of its Prime consumer club). You cam feast on free fast shipping and returns, free movies and TV shows, free music, free gaming, free deals – endless deals culminating in Amazon’s own shopping holiday, Prime Day. Service that an emperor could not even imagine let alone match, and entertainment fit for a king, is your for $119 a year.
Amazon’s Information Black Hole
Yet all this comes at considerable cost. There are still terrible stories coming out of Amazon’s warehouses. We still don’t really know whether working at Amazon – even as a white-collar worker – is good or bad. And even though there are 2 million outside sellers (“third parties”) on Amazon’s platform, they too have plenty of bad stories to tell about Amazon’s secretive and impenetrable justice system and unfair competition from Amazon.
Amazon is among the most secretive companies in America. Its annual reports use data primarily to obfuscate. Beyond legally required disclosures, it publishes only a handful of cherry-picked statistics about its operations. It uses nondisclosure agreements to prevent employees and former employees speaking out, while forced arbitration ensures complaining sellers stay within the privately-run Amazon justice system. Even though Amazon will soon likely become the biggest employer in America, it reveals almost nothing about its employment practices. As Amazon grows into Behemoth, this silence becomes unacceptable. Balancing the power of the Behemoth requires that the closed and bolted door of company information be blasted open, as Amazon’s secrecy means that we cannot answer the most basic questions.
Sellers. The Marketplace is Amazon’s platform for outside sellers, and it is now much bigger than Amazon’s own retail operation, much more profitable, and growing faster. Amazon could tell us how many Marketplace sellers there are, where they come from, what sectors they are in, and how much they sell. It could explain how many sellers are subjected to disciplinary action, and what happens to them – both in terms of their status as sellers but also the impact on their business. Amazon has all that data readily available. Amazon could also show how Marketplace sellers interact with Amazon Retail: what are the factors that lead to winning the Buy Box, which largely determines who wins and who loses on Amazon’s platform. Amazon could also explain the finances of the Marketplace in more detail – its share of each transaction, how well shipping fees cover its own costs, and whether it’s possible to make money on Amazon without using Amazon advertising. It could show whether it treats sellers fairly, or just uses its strength and platform dominance to make and then defend the world it wants to live it.
Money. Amazon deliberately obfuscates its accounts by publishing quite different categories of revenues and expenses. Behemoth, Amazon Rising argues that Amazon is in fact losing a lot of money on its own retail operations – more than $30 billion annually by 2019. That has implications for investors and also for regulators, who may be concerned about predatory pricing. I’ve been told by insiders that this analysis is wrong. That is possible, though I doubt it. But the point here is that under current regulations, Amazon can use financial filings to help prevent outsiders from understanding its businesses.
Workers. Amazon claims vehemently that conditions in the warehouses are nothing like the picture painted by its critics. That warehouses offer good jobs at good wages. It’s possible that Amazon is right. But we don’t know, because Amazon publishes almost nothing about conditions in the warehouses. We don’t know how many workers are employed. How many are full time, how many are contractors not employees? What is the pay in each warehouse? What benefits are actually used by workers – how many in fact qualify for retirement benefits? What is the turnover for each warehouse? That’s a key metric because it’s the single best indicator of worker satisfaction. What is the health and safety record for each warehouse? Which warehouses are good, which are not? Amazon is among the most data-centric companies in the world. It tracks every one of those metrics internally.
Why antitrust won’t work
Antitrust has since the 1890 Sherman Act been the primary tool for managing big companies in the US. Over recent decades, though, its scope has narrowed and the definition of “harm” has increasingly focused tightly on consumer prices. But antitrust is now the wrong tool for managing the digital behemoths, and Amazon in particular. Specifically, there are technical reasons why antitrust won’t effectively restrain Amazon; problems with politics – customers love having a unified place to go; and practicalities – antitrust answers mostly the wrong questions, and imposes rigid answers on highly flexible environments.
To begin with, antitrust action against Amazon would probably fail. It’s not an accident that the Department of Justice is going after Google and Facebook first.
- Market definition. Amazon dominates ecommerce, but legally, that’s probably not the relevant market. It still holds only a small fraction – about 7% – of total retail. Even online, there are plenty of alternatives.
- Consumer prices. Amazon has pushed consumer costs down, not up; the “Amazon effect” is even deflationary for the whole economy. It is not systematically looting consumers. On the contrary.
- AWS is protected. Under current law, there is no reason why Amazon should spin off AWS: it operates in a highly competitive market with falling prices. And while the AWS-Amazon connection was strategically important from about 2006-2016, that’s no longer the case.
- Private labels don’t matter. The recent focus on private labels at Amazon is basically ridiculous. All retailers use private label brands – they account for around 30-35% of Walmart’s groceries, for example. And yes, big retailers use their access to sales data to pinpoint potentially profitable areas for private labels. Amazon is essentially no different. But Amazon’s private labels account for less than 1% of sales – barely an ant on the elephant’s behind. If necessary, Amazon could abandon private labels tomorrow without a backward glance.
Some limitations imposed by antitrust would be counterproductive. Antitrust hawks complain that big companies use control of one market to enter and then dominate adjacent ones. That’s true, especially for Amazon. But progressives also complain about industry concentration – that big companies increasingly dominate their industries. Yet outside its home turf in adjacent markets, Amazon is just a big well-funded startup. If it became a serious competitor in pharmacy, few would shed any tears for CVS, Rite Aid, Walgreens, and Walmart. A powerful new competitor like Amazon would – for regulators – be a feature, not a bug.
Antitrust hawks also overstate the case for limiting acquisitions. Over the years, Amazon has bought about 120 companies for a total of around $25 billion. Initially, several key purchases expanded its retail footprint and eliminated important competitors – the leading online bookstores in Germany and the UK, Zappos in shoes, diapers.com for baby products. More recently, acquisitions have been focused either on technology (Kiva’s robots for example, or Alexa Inc.’s voice technologies) or to get into adjacent markets (PillPack in pharmacy, for example, or Whole Foods in groceries and Audible in books). Those were gateway purchases, offering Amazon a significant toe-hold in a new market and the chance to leverage existing strengths. For AWS, buying the competition is impossible; Microsoft and Google are just too expensive, although AWS may buy other service providers to expand its portfolio of products. Limiting the acquisition of direct competitors is obviously appropriate. But beyond that, technology acquisitions and adjacent market acquisitions are protected not subject to antitrust concerns.
One component of Amazon’s empire is potentially vulnerable to antitrust. That’s the Marketplace, because there Amazon is indeed a monopolist for the sellers who need to use Amazon’s US platform. Amazon has exploited its position to raise prices for sellers (not consumers): Amazon’s overall take from third-party sales is now about 36%, up sharply in recent years, as advertising became effectively compulsory. Margins are 47%. Regulators may decide that sellers are actually Amazon customers, and that the online marketplace is the correct market definition for antitrust analysis. Amazon controls on the order of 85% of platform sales in the US. Amazon’s operating income from Marketplace (including advertising) was about $32 billion in 2019, with an operating margin was about 47%. Those are bright red flags.
But even in relation to the Marketplace, antitrust is also a poor tool in a fast changing world. Amazon might for example agree that it won’t apply predatory pricing. But what does that mean in an environment where it makes billions of price changes annually? Antitrust is a one-off investigation: regulators acquire gigabytes of data and then sift through them to make a specific case at a specific time. Sometimes that works, sometimes it doesn’t, but the consent decrees that may come from successful litigation are themselves quickly over-run by events. And antitrust would have little to say about the core problems of the Marketplace – Amazon’s mediocre policing system and its draconian impact on sellers. Antitrust is the Maginot line of state authority over against big companies.
There are other drawbacks to antitrust as well. It usually takes a long time – the Microsoft case (which the government essentially lost) took more than a decade. And even when the government wins and imposes structural change – such as the breakup of AT&T, outcomes don’t last. AT&T is regarded as a win for regulators, and led to the breakup of the Bell system. There’s no doubt this had positive results for consumers, and allowed a wave of innovation, but here we are with three remaining telcos, two of which just merged. It’s not as though Bell was replaced by small companies. Practicalities matter: long lags and uncertain results make structural change the nuclear option, rather than day to day constraint on companies.
Finally, antitrust could limit Amazon incredible record on innovation. This is the most innovative large company in the world, responsible for a strong of innovations that have transformed shopping, reading, and entertainment. Many of those innovations come from cross-fertilization between different parts of Amazon. The impact of breaking it up on innovation is simply unknown and hence risky.
Antitrust probably won’t work, won’t address key problems, and won’t generate better outcomes. Fortunately, there is a better way, one that would as a side benefit make antitrust enforcement much easier, faster, cheaper, and more effective.
An open data regime
We assume that company data is private unless otherwise required for financial filings of for health and safety reports. B8t that is not always true. In the US, New Deal legislation codified rules for “natural monopolies” like the electricity supply industry. The 1935 Public Utility Holding Company Act forced electric utilities to get permission from regulators for price changes. Much of the discussion of utility regulation since then has naturally been about price- when are increases justified, how to push back against company claims, etc. But that’s not the important point here. What matters is the disclosures. Utilities have to justify every increase (indeed every price). That means providing all the data and information that underpins the claim – which, it turns out, covers the entire range of operations of the business.
Utilities disclose far more of their business operations than other businesses. Key information included:
- Prices – what is the utility charging each customer?
- Costs – every dollar has to be reported, including labor and benefit costs, plant maintenance, operations (e.g. IT)
- Investments – every investment has to be approved, so each has to be explained, justified, and costed
- Strategic and operational plans have to be approved, because they involve money
For monopoly utilities, information plus regulation replaces competition: if competitors don’t exist, regulators must act instead. The sunlight imposed through utility regulation is the best way to manage Amazon, to disinfect it its anti-social tendencies.
Let’s start by specifying the data that Amazon would be required to release. Broadly, it would fall into two areas – the ecommerce platform, and the workforce. The platform is more specific to Amazon, but the current massive imbalance between worker and employee is starkly reflected within Amazon, where there are no recognized worker organizations (at least in the US), and where information about the workforce is secret.
So far as the Marketplace platform is concerned, Amazon would be required to publish data via API on all aspects of operations. And insofar as Amazon itself is a leading seller on its own platform, these requirements would cover Amazon’s own activities as well. Broadly, open data should cover:
- Listings and sellers. Data should break out listings by category (see below), and by seller characteristics (e.g. country of origin, years on the platform, number of items listed, aggregate annual sales). The latter should extend to individual categories (e.g. the number of sellers in laptop backpacks based in the US).
- Search algorithms. Outcomes on Amazon are overwhelmingly determined by search results and by the algorithms that drive selection of the Buy Box. Search algorithms and in particular those driving Buy Box selection need to be published and available for outside scrutiny. The same is true for other marketing programs at Amazon, such as Amazon Choice.
- Sales, at a granular but not individual seller level (except for Amazon which would report its own sales). That data should cover all sales on the platform, broken out by Amazon browse tree categories (developed by Amazon to reflect search and sales patterns) or similar. Published data should show (for each category) the number of items sold, average, median, and aggregate sales, sales breakouts by seller (deidentified except for Amazon). The objective is to provide detailed sales data without breaching seller privacy.
- Amazon’s private justice system. Amazon should report all disciplinary actions. Those should cover initial notifications, clearance rates for initial notification, subsequent actions – including suspension, but also including any other actions taken by Amazon to de-emphasize seller visibility for example). Reporting on downstream disciplinary action should cover appeals, results of appeals and the time they require for resolution, along with arbitrations and their outcomes. Amazon should also estimate economic losses of sellers due to disciplinary action (i.e. estimated lost sales caused by suspensions), and should provide breakouts for the causes of disciplinary actions – performance, IP issues, rules breaches, delivery etc. Amazon should also report expenditures on discipline.
- Amazon-specific data. Amazon’s own retail operations will require additional data disclosure, because they impinge so heavily on the Marketplace. In all sales categories, Amazon should provide comparative data showing first party, private label, and third party sales. Where Amazon-exclusive brands have gained traction (e.g. >1% of category sales), Amazon should provide separate data for these partnerships.
- FBA. Shipping and fulfillment are an integral part of the Amazon platform, and Amazon has clearly encouraged more sellers to use FBA for more of their goods. Using the categories described above, Amazon should provide detailed data on use of FBA, including as a percentage of all sales and of third party sales. Amazon should also disclose all charges related to shipping and fulfillment, including storage and disposal charges, as well as the actual cost to Amazon of these services. Along with data by category, Amazon should disclose data on FBA adoption by size of seller (by seller ranking).
- Advertising and product placement on search results pages. Advertising with Amazon has quickly become almost mandatory, and the placement of ads may have been used to tilt customer selections to Amazon’s benefit. This substantially affects platform operations, so Amazon should disclose auction prices for ads, its own use of advertising, and data on the distribution of ads.
Many of these items might emerge in the course of a long and aggressive antitrust investigation. The difference is that these data would be available in real time and via API, not through one-off investigator requests which, as previous big antitrust cases demonstrate very clearly, can be delayed or misdirected or simply stonewalled altogether.
Basic principles and a regulator for digital platforms
Amazon is immensely complex. The hundreds of pages in the US House of Representatives staff report barely scratch the service. To regulate platforms we will requires need professional platform regulators to ensure that the platforms meet basic principles, operate cleanly, and meet open data requirements. This requires a digital platform regulator (DPR).
The DPR could be entirely independent (like the Consumer Product Safety Commission) or could be part of the FTC or indeed the SEC. Congress will decide that. But either way, the DRP must be insulated from political and industry pressures, so it needs to be financed directly by fees on marketplace platforms; it must be professionally managed and run; it must be given a clear remit, annual reporting requirements, and an appropriate structure and resources; and it should defend clearly-defined principles. Those might include:
- Nondiscrimination. All items on the platform must be treated equally with regard to placement and sales. The platform must be open to all sellers who pass appropriate registration requirements.
- Nonbundling. Platforms may not require the use of bundled service. So for example, Amazon may not disadvantage items and sellers that are not delivered via FBA.
- Pricing freedom. All participants on the platform must be permitted to price as they like on and off the platform. Off-platform pricing cannot be allowed to affect search and sales results on the platform, so lower prices on a seller’s website or indeed Walmart+ could not be used to pressure sellers to lower prices on Amazon.com.
- Public disclosure of algorithms. All algorithms that affect search results, including both direct search pages and indirect influences like Amazon Choice badges, must be fully published and opened for analysis. Algorithms may not unfairly advantage items sold by the platform owner.
- Fair justice. Platforms must regulate their own markets efficiently and fairly. Complaints must include full identifying data that describes the problem, clear and rapid pathways to resolution, and an effective and unbiased third party appeals process. Arbitration clauses in operating agreements must reflect these requirements, and must permit the aggregation of seller complaints into classes.
- Incentives. To incentivize platforms to manage their justice systems better, sellers may use data on losses caused by justice actions as the basis for claims against the platform, if eventually the seller prevails. Triple damages may be available for proven negligence.
- No self-dealing. Companies like Amazon that participate in their marketplace must not tilt the market to their advantage. This covers actions including possible tweaks to the search and placement algorithms, justice actions, and pricing.
- No predatory pricing. In general, aside from very limited sales events (to be defined by the DPR), Amazon should not be allowed to sell online on its platform at below cost. The consequent market distortions are unfair to sellers with shallow pockets, and in the long run do not benefit customers.
- Communications. Sellers must be permitted reasonable access to their customers, which Amazon currently forbids. Sellers must be allowed to ask customers to opt in to future communications from them – including communications off the Amazon platform.
The key to successful regulation is that the regulator would have instant access – via API – to data that would otherwise take months or years of careful digging and repetitive data requests to find. It was precisely for this purpose that Jeff Bezos famously mandated that all internal communications between teams should go through an API. This replaced piecemeal inquiries by email with a largely automated electronic data flow. Exactly the same logic applies here.
That instant access would also make antitrust and other forms of regulatory action feasible. Instead of spending years fighting companies for disclosure – likely out of date by the time it was granted – data for successful regulation would be available in real time using APIs. In effect, regulators would have access to the company’s own data systems.
Insisting on data transparency and giving a DPR power through financial and other sanctions, while keeping the nuclear option of antitrust available, would be transformative. Almost all of the complaints and questions surrounding the way Amazon operates its platform would quickly be resolved by the disclosure of relevant data. Is Amazon tilting the playing field? We will see immediately. Is FBA tied to winning the Buy Box? The data will demonstrate. The key, critical, central point is this: unlike on-off antitrust investigations, mandatory disclosure would provide permanent and real-time monitoring of Amazon’s platform.
Similar access to real time data on Amazon’s labor force would permit effective action by labor regulators. More important, it would provide potential hires with exactly the information they need to decide whether Amazon is the right place for them to work. We already know that some employees love Amazon. The point is to protect those who don’t, and to make sure that all workers have sufficient information to make good choices. Asymmetric information in the labor market is neither efficient nor fair.
Amazon itself is neither a villain nor a hero. It just exploits its environment efficiently and ruthlessly. Today, that’s an environment of secrecy that’s ripe for exploitation, quite similar to those of the past. The early Industrial Revolution brought the exploitation of child labor and appalling conditions. Those were eventually tamed by child labor laws and health and safety regulation. The unlimited powers of management were circumscribed in the labor settlements of the New Deal and then after WWII. Today, digital platforms require further push back. As in the past, those who benefit from our existing cult of secrecy around company operations will resist. But we will find that the specific data that companies currently use for their own benefit is also their Achilles heel. Sunlight is indeed the best disinfectant.
Austan D. Goolsbee and Peter J. Klenow, “Internet Rising, Prices Falling: Measuring Inflation in a World of E-Commerce,” AEA Papers and Proceedings 108 (May 2018): 488–92. ↑
Matt Stoller, Goliath: The 100-Year War Between Monopoly Power and Democracy (Simon and Schuster, 2020). ↑
API means ↑
Amazon.com https://sellercentral.amazon.com/gp/help/external/200389080?language=en_US&ref=efph_200389080_cont_1701 ↑