Data Brokers Part 2: Surveillance Pricing

Charging different buyers different amounts for the same good is not an invention of the data economy. It is the original condition of commerce, the natural state of every bazaar and bartering floor on which a price was the outcome of a negotiation and therefore particular to the two people conducting it. The actual anomaly is the arrangement most of us mistake for the natural order. The single posted price — fixed, public, and identical for everyone — is barely more than a century and a half old, a deliberate invention of the early department stores: Wanamaker in Philadelphia and his imitators, who replaced haggling with a marked ticket partly as an operational efficiency and partly as a moral claim, the price made honest by being made universal.

The previous article was an intro into the inner workings of data brokers and how they obtain, aggregate, and model your personal information. This piece is about what those predictive data models are being used for: the restoration of per-person pricing after the brief civilizing interlude of the price tag. In the old negotiation, the buyer at least saw the seller’s face and could walk away. In the new one, the haggling has moved entirely to the seller’s side, conducted by a model holding full information against a buyer holding none.

Older dynamic pricing prices the situation: the off-season room, the matinee that costs less than the same film at night, each a function of conditions, so that the price, however it moves, remains a statement about the world. Surveillance pricing prices the person. Its argument is not the hour or the weather or the remaining inventory but you, the individual, resolved to the absolute most a model believes you will surrender before you walk away. The moment this becomes the method, the price ceases to describe the good at all and becomes instead a measurement of your urgency, your alternatives, and your capacity to absorb a figure.

The history

Arthur Pigou — a Cambridge economist, the successor to Alfred Marshall and one of the founders of welfare economics — gave the practice its modern account in his 1920 book The Economics of Welfare. He sorted price discrimination into three degrees, and the sorting is still the one in use.

The third degree charges different prices to broad, observable groups: the student rate, the senior discount, the higher markup in a wealthier ZIP code. The second degree prices by quantity or version and lets buyers sort themselves: the bulk discount, the first-class cabin, the “pro” tier. And the first degree — what Pigou called perfect discrimination — is the logical extreme: a separate price for every individual, each set at the exact maximum that person would pay, so that the seller captures the whole of the gain from the trade and leaves the buyer none of it.

Pigou treated that first degree as an instructive abstraction rather than a possibility, because reaching it would require a seller to see into each buyer’s mind, and no merchant could. [1] The entire apparatus this article describes is most precisely understood as the long, expensive, and now nearly finished project of reaching that limit: of manufacturing, out of data, the omniscience Pigou had assumed was permanently out of reach, so that first-degree price discrimination passes quietly from a textbook hypothetical into a deployed capability.

The history of dynamic pricing is largely the history of its renamings. It first showed up in any advanced form in the airlines. An airline seat is perishable: once the plane leaves the gate, an empty seat is worth nothing. The airline wants to fill every seat, but not by handing a cheap fare to the business traveler who would have paid full price. Yield management is the first name, and the system that arose. It watches how a flight is filling up — how fast seats are selling compared with past flights, how many are left, how many days remain — and decides, continuously, how many seats to sell at each fare and when to shut the cheap ones off, holding a variable number back for last-minute travelers who will pay the most, and releasing discounts only on the seats that would otherwise fly empty. It prices the flight, not the passenger — the situation, not the person.

Robert Crandall built the first full system at American Airlines in the 1980s, and American credited it with roughly a billion and a half dollars in added revenue over three years. The same machinery produced the airline’s Ultimate Super Saver fares — cheap on just enough seats to match the budget carrier People Express where it hurt, while the rest of the cabin still sold at full price. People Express could not answer it and was bankrupt by 1987; its founder blamed the Super Savers by name. [2]

From there, hotels in the 1990s — Marriott prominent among them — called it revenue management. Then the web arrived and it became dynamic pricing. When the data turned individual, it became personalized pricing. Each renaming carries the practice a step further from the blunt word Pigou used: discrimination. And it is no accident that the one term in the sequence with any sting in it — surveillance pricing — is the only one coined by the people on the receiving end rather than the people selling the software.

The three conditions

Charging each person the most they will pay sat mostly idle for decades, because it needed three things to be true at once. It needed permission, it needed computation, and — the one everyone forgets — it needed concealment. Only lately have all three arrived together.

Permission is the right to charge two people different amounts for the same thing. In the United States that right is nearly absolute. No general law says a seller must offer everyone the same price, or disclose that a price was set for you in particular, or let you see the price it showed someone else. Airlines got their permission from deregulation in 1978. For most goods there was never a rule to remove. The permission was always just the silence of the law, and that silence has held.

Computation is the ability to know what each person will pay. This is the part Pigou thought was impossible, and for most of history he was right — a merchant could read a stranger’s face and little else. That barrier is gone. An entire industry now sells the stranger’s file: hundreds of data brokers who assemble a record of income, debts, habits, and location on nearly every American, tied to your name and your devices by the identity graphs from the first half of this series. Feed the file to a model and it returns a number — what you, specifically, will pay. The thing Pigou called impossible is now a purchase the company makes. [3]

Concealment is the one that actually mattered. Permission and computation were both within reach years ago, and still the personal price did not take over. The reason is that every time a company tried it in the open, it got caught, and every time it got caught, it backed down.

In 1999 Coca-Cola described a vending machine that would raise the price of a can as the temperature rose; the backlash killed it in days. In 2000 Amazon was caught charging different customers different prices for the same DVDs, and apologized. The scene has repeated ever since. Wendy’s floated “dynamic pricing” on its menu boards in 2024 and spent a week insisting it meant discounts. Kroger’s digital shelf labels drew a Senate letter. Delta said an algorithm would set individual fares, then denied the fares were individual. In every case the technology worked. What failed was secrecy. The practice could survive anything except being seen. [4] [5]

So the real obstacle was never permission or computation. It was the posted price that you and anyone around you could see changing. A posted price is the simplest form of accountability in commerce. It lets two strangers find out, without a word, that they were charged the same, and it lets either of them notice when they were not. To hide a personal price, you first have to take the posted price apart. That is what is happening now, one piece at a time.

The price tag

The posted price was never just a number. It was shared infrastructure, the thing that let a market be a place where strangers could check that they stood on equal terms.

Start with visibility. The price existed before you asked — that is now optional. More and more, the number does not appear until you have identified yourself: the item with no price until it is in your cart, the site that asks you to sign in to see it, the “quote,” the “contact sales,” the airline fare that is no longer published but generated for you the instant you search. A price that appears only for you is a price no one else can stand beside.

Then stability. A static price was the same this hour as last. Electronic shelf labels end that. A paper tag could only be changed by hand; a digital tag changes instantly, from a central server, without notice. Online it moves faster still. Amazon changes prices millions of times a day, so even a public price is never still long enough to compare. [6]

Then uniformity. The price now varies by where you are (ZIP-code pricing), by the device you hold (Mac users are comparatively charged more than PC users), by the loyalty app you’re logged into, by whether the site remembers you, and finally by the whole profile attached to your name. [7]

And then wholeness. Drip pricing breaks the static price from below: the headline trailed by “resort” and “service” and “processing” fees that surface only at the last step. Shrinkflation breaks it from the other direction, holding the price still while the package shrinks. Tiers and bundles ensure that no two offers are ever quite the same thing. And steering handles what remains — arranging what you see so the cheaper option sits below the fold, or never appears at all. [8]

One at a time, these are small annoyances, easy to blame on clutter or fees. Together they remove a public fact. The personal price did not win an argument against the posted price. It methodically destroyed the conditions under which a posted price could function.

The loyalty trap

The defense and the surveillance are the same act. Every loyalty program works the same way: you are paid a small, certain discount in exchange for the information used to find the largest price you will tolerate. The card, the app, the account are sold as the way to save money, and they do lower the price — today. They are also the richest stream of behavioral data a seller collects about you, the raw material that makes the personal price possible in the first place. The instant you log in to claim the member price or to check out, you reattach to the whole record. The act of collecting the discount is the act of being identified.

For a century the clever shopper could beat the store — wait for the sale, haggle, clip the coupon, know the going rate. That edge is gone. Every move that once marked a smart buyer now marks a known one. The harder you work to win, the more legible you become to the thing you are trying to win against. And the system does not only read you; it shapes you, nudging you toward the habits that make you easier to price next time.

The countermeasures that remain all attack the signals of the moment: clear your cookies, open a private window, log out, use a VPN, leave an item in the cart to summon a hesitation discount (aka. exit intent). These work, briefly and partially, because they strip the history that marks you as a returning, committed buyer. But they are evasions, not solutions, and each becomes less effective as the identity graphs described in the first half of this series close the gaps between your sessions. There is no longer such a thing as a smart shopper. There is only a more or less measured one.

Who wins

The usual defense of personalized pricing is that it helps the poor. Charge each person their maximum and you also charge the price-sensitive shopper less, so the struggling family gets the cheaper fare while the executive makes up the difference. There is a real study behind this. When the economists Dubé and Misra ran a controlled pricing experiment, personalized pricing raised the seller’s profit by about 19 percent, and more than 60 percent of customers paid less than they would have under a single price. [9]

But look at the other number from the same experiment. Total consumer surplus — the collective benefit buyers get from a market — fell by 23 percent. Both things are true at once: most people pay a little less, a few pay a lot more, and on net more money moves from buyers to the seller. The reason it feels fair is that no one experiences the average. You see your own price, decide you got a deal or got gouged, and never learn that the system as a whole moved wealth quietly upward.

All of this falls unevenly. A surplus-extracting price lands hardest on the people with the least room to refuse, and those people are, overwhelmingly, the poor. They are the ones who cannot wait for the sale, cannot buy in bulk, cannot pay the yearly rate instead of the monthly one, cannot drive to the cheaper store, cannot afford to walk away — and every one of those constraints reads, to a pricing model, as willingness to pay. The sociologist David Caplovitz gave this its name sixty years ago, in a book called The Poor Pay More: the poverty premium, the long-documented fact that the same goods cost more for the people who can least afford them. [10]

The impossible price

Pigou held that perfect discrimination “very likely does not exist anywhere because of the enormous transaction costs required to get each customer to pay his reservation price,” and that attempting it “would produce fraud in bargaining.” [11] His impossibility was not a law of nature. It was a temporary information shortage, and it has been solved.

A generation after Pigou, the asymmetry of information — one side of a deal knowing more than the other — became a field of its own. George Akerlof won a Nobel for the case of the used car, where the seller knows the quality and the buyer does not, and showed the gap can be corrosive enough to wreck an entire market: the good cars withdraw, the lemons remain, trust drains out. Joseph Stiglitz won a Nobel for showing that once information is asymmetric, markets are not efficient in a clean sense. [12] Economics has worried over that imbalance for fifty years.

Surveillance pricing does not perfect the market. It builds, on purpose, the exact asymmetry that economics spent half a century learning was a market failure, and sells the result as efficiency. What it installs is closer to a tax — a private one, charged transaction by transaction, set to whatever each person can least afford to refuse, collected by companies that answer to no electorate. A levy that takes most from those with least, returns nothing, and stays invisible to the person paying it is not a refinement of the market. It is a slow draining of it. [13]

The modern theory of pricing in an information economy was written, in large part, by Hal Varian, whose 1998 book Information Rules taught a generation of firms how to turn what they knew about a customer into a price. In 2002 Varian joined Google, where he became its chief economist. [14] The theory and the machine were never separate undertakings. The people who showed that perfect discrimination was possible and the people who built the infrastructure to deliver it were, often enough, the same people — a fact that says less about any individual than about the structural continuity between academic economics and the industry it was supposed to describe from the outside.

And the machine, once built, produces its own undoing. The first thing that happens is that buyers learn. Once a person understands that revealing himself raises his price, he stops revealing himself — masquerading as the poorer customer, manufacturing a lower willingness to pay. Economists call this the ratchet effect, [15] and it is fatal to the promise: perfect discrimination assumed buyers would show their hand, and buyers, finding an honest hand is punished, begin to hide it. The market fills with evasion, and the clean efficiency dissolves back into the distortion it was supposed to abolish. The seller’s omniscience produces, in the end, a population working to become unreadable.

The second thing is that trust drains out. Researchers find that personalized pricing is felt as unfair even by the people it offers a better deal, because it breaks the plain expectation that a price is a price; trust falls, and the willingness to take part at all falls with it. [16]

And the third thing is the oldest worry in the discipline. Consumer surplus is not only the buyer’s good luck — it is the slack that lets the buyer keep buying. The underconsumptionists of the nineteenth century, and Keynes after them, kept circling the same trap: an economy that squeezes its buyers too well strangles the demand it runs on. [17] You cannot perfect the extraction of the consumer and still have a consumer.

References

1. A. C. Pigou, The Economics of Welfare (1920), classifying price discrimination into first, second, and third degree; first-degree (“perfect”) discrimination — charging each buyer their exact maximum willingness to pay — as the theoretical limit no seller could reach without seeing into each buyer’s mind.

2. “Yield management”: Littlewood’s rule (British Overseas Airways, 1972); Robert Crandall’s system at American Airlines through the 1980s; the Ultimate Super Saver fares and the 1987 collapse of People Express; INFORMS Edelman Prize recognition (~$1.4 billion over three years).

3. The data-broker and identity-graph apparatus is documented at length in Part 1 of this series. See also Federal Trade Commission, “FTC Surveillance Pricing Study Indicates Wide Range of Personal Data Used to Set Individualized Consumer Prices,” January 2025: intermediary firms use personal data — location, demographics, browsing and purchase history — to set targeted, individual prices.

4. “Coke’s price may rise with the temperature,” Deseret News, October 28, 1999 (CEO M. Douglas Ivester’s weather-based vending-machine pricing, abandoned after backlash); “Amazon apologizes for random DVD price test,” CNN, September 28, 2000 (different prices on 68 DVDs; 6,896 customers refunded an average of $3.10; Jeff Bezos disavowed demographic pricing).

5. “Wendy’s pushes back after outrage over surge pricing prospect,” Bloomberg, February 28, 2024; Sens. Warren and Casey, “letter to Kroger on digital price tags and surge pricing,” August 2024; “Delta plans to use AI in ticket pricing draws fire from U.S. lawmakers,” CNBC, July 2025, and Delta’s subsequent denial of “individualized prices based on personal data” (Skift, August 2025).

6. “Walmart digital price labels are coming to every store shelf in U.S. by end of 2026,” CNBC, March 21, 2026. Amazon adjusts prices roughly 2.5 million times a day (Profitero analysis), against roughly 50,000 changes a month at Best Buy or Walmart.

7. Aniko Hannak et al., “Measuring Price Discrimination and Steering on E-commerce Web Sites,” Northeastern University, IMC 2014 (Staples ZIP-code pricing; Home Depot device-based steering); “Report: Orbitz steers Mac users to pricier hotels,” CBS News, June 26, 2012 (reporting Dana Mattioli’s Wall Street Journal story).

8. Federal Trade Commission, “Federal Trade Commission Announces Bipartisan Rule Banning Junk Ticket and Hotel Fees,” December 2024 (effective May 2025), on drip pricing; “Hidden Fees, Drip Pricing, and Shrinkflation,” Chicago Booth Review, on shrinkflation as a concealed price increase.

9. Jean-Pierre Dubé and Sanjog Misra, “Personalized Pricing and Consumer Welfare,” Journal of Political Economy 131, no. 1 (2023). A randomized pricing field experiment: personalized pricing raised firm profit by roughly 19 percent and left more than 60 percent of consumers paying less than under a uniform price, even as total consumer surplus fell about 23 percent.

10. David Caplovitz, The Poor Pay More: Consumer Practices of Low-Income Families (Free Press, 1963), the founding study of the “poverty premium” — the documented tendency of low-income consumers to pay more for the same goods and credit.

11. A. C. Pigou, The Economics of Welfare (1920); Pigou’s view that first-degree discrimination “very likely does not exist anywhere” because of transaction costs, and would “produce fraud in bargaining,” is discussed in Herbert Hovenkamp, “The Coase Theorem and Arthur Cecil Pigou,” Arizona Law Review 51, no. 3 (2009).

12. George A. Akerlof, “The Market for ‘Lemons’: Quality Uncertainty and the Market Mechanism,” Quarterly Journal of Economics 84, no. 3 (1970) (Nobel Memorial Prize, 2001); Bruce Greenwald and Joseph E. Stiglitz, “Externalities in Economies with Imperfect Information and Incomplete Markets,” Quarterly Journal of Economics 101, no. 2 (1986), showing that under asymmetric information markets are not constrained Pareto efficient (Stiglitz, Nobel Memorial Prize, 2001).

13. “AI-Driven Dynamic Pricing: Erosion of Consumer Welfare, the Invisible Hand, and the Rise of Platform Quasi-Taxation,” German Law Journal (2026), framing personalized algorithmic pricing as a private, individualized levy on consumers.

14. Carl Shapiro and Hal R. Varian, Information Rules: A Strategic Guide to the Network Economy (Harvard Business Review Press, 1998); Varian joined Google in 2002 and served as its chief economist.

15. On the “ratchet effect” — buyers withholding or distorting information once they learn that revealing it raises their price, which undermines the efficiency case for first-degree discrimination — see “Big data and first-degree price discrimination,” Bruegel, and Drew Fudenberg and J. Miguel Villas-Boas, “Price Discrimination in the Digital Economy” (2012).

16. Jura Liaukonyte et al., “Personalized Pricing and Price Fairness,” International Journal of Industrial Organization; see also “The Perils of Personalized Pricing,” Yale Insights — personalized pricing is perceived as unfair and erodes trust and willingness to transact even among consumers who receive lower prices.

17. On underconsumption and the dependence of aggregate demand on consumer purchasing power — from Sismondi and Malthus through Marx’s “realization problem” to Keynes — see “Underconsumption” and John Maynard Keynes, The General Theory of Employment, Interest and Money (1936).