Prices in the Machine

This article appears in the December 2025 issue of The American Prospect magazine. Subscribe here.

Earlier this year, a slightly balding man in spectacles, a black T-shirt, and bright high-top sneakers gave a presentation about how his computer can predict what you want to buy. His name is Dr. Uri Yerushalmi, co-founder and chief artificial intelligence officer of Fetcherr, an Israeli-based pricing consultant used by a half dozen airlines around the world.

“While most of us in the AI community have been focusing on building models that are generating either text or image, in Fetcherr we have been focusing during the last five years on building large market models,” Yerushalmi said. Trained on years of market data—prices, orders, competitors, regulation, stock prices, even the weather—Fetcherr’s “business agents” aim to simulate market dynamics, assist pricing analysts, and even automatically price tickets.

More from David Dayen

One presentation slide stuck out. It showed price fluctuations before the use of Fetcherr’s system and after. In the before graph, prices are relatively static and straight. After Fetcherr, the jagged lines pulse with staccato rhythms. “The dynamic is much more similar to dynamics in NASDAQ or capital markets where the prices change much more frequently,” said Yerushalmi, “because every time something in the market changes, there is an immediate response.”

Yerushalmi once ran AI for a high-speed stock trading firm. He approaches pricing like a science experiment, engaging in constant real-time testing and tweaking to maximize corporate profits. Fetcherr boasts of delivering annual “revenue uplift” of over 10 percent. The guinea pigs for these tests, the ones being separated from their cash, are you and me.

AI can depict you as an anime character. It can respond half-intelligently to questions about the Franco-Prussian War or concentrations of sulfur in the upper atmosphere. It can delight and distract and maybe help you get work done. But none of that is as prized by corporate America as its data-driven approach to the previously conjectural world of pricing.

“That’s the use case they don’t want you to talk about. That’s why we’re building all these data centers,” said Lee Hepner, senior legal counsel for the American Economic Liberties Project. “As we have built social media platforms that shape the flow of information across society, now we are building the platforms that control the flow of money.”

Technology-fueled pricing is more widespread than once thought, presenting serious policy challenges in an age where affordability is on everyone’s minds. AI bots are colluding with one another, anticipating consumer choices, and accumulating surplus—that is, transferring wealth—for the businesses that employ them. It’s part of why corporate profits hit record highs after the pandemic and have stayed there.

The public recoils at the thought of seeing prices tick up based on when they like to get lunch or what device they’re using. When Delta, which employs Fetcherr as a pricing agent, announced on an earnings call this summer that up to 20 percent of its flights would be priced using AI by the end of the year, the outcry was so intense that the company claimed its critics were peddling “misinformation.”

Policymakers and advocates have united to crack down on some tactics, winning real legislative victories from coast to coast. But every step government takes can be countered by AI price-setters. How can consumers keep up?

“What worries me is that it will become hyperefficient to price discriminate and to charge more in ways that ordinary people will not be able to combat,” said Doha Mekki, the top deputy at the Justice Department’s Antitrust Division under Joe Biden. “And it will make a very bad cost-of-living crisis even worse.”

A YEAR AGO, I WROTE ABOUT WHAT I TERMED surveillance pricing: companies offering individualized prices based on personal data. The idea was that a business equipped with granular information about demographics, purchase history, social and financial interactions, or even medical status could exploit a customer’s willingness to pay. Uber could charge more when a rider booked on a company credit card; people aren’t as price-sensitive when someone else is paying. Delta could jack up fares after learning that a traveler needs to attend a funeral; desperation could translate into opportunity.

I couldn’t be certain how many businesses used surveillance pricing. The proliferation of third-party pricing consultants touting “digital pricing transformations” seemed like a strong indicator. But concealment was key to the strategy, to minimize the anger generated by charging different prices to different people. For instance, when a reporter at SFGate logged in to hotel booking platforms using his regular IP address in high-wage San Francisco, he received a quote of $829 a night. But when he used a virtual private network set up to originate from Phoenix or Kansas City, prices were more than $500 less. Without a public price, it takes research to understand if someone is being ripped off.

A month after my report, Lina Khan’s Federal Trade Commission announced an investigation into surveillance pricing, seeking information from eight third-party consultants. The agency only had a short time to conduct the study before the changeover of power in Washington. But days before Donald Trump’s inauguration, it issued “research summaries” of the work done thus far. (Trump’s FTC has yet to finish the study.)

Stephanie Nguyen, the FTC’s chief technologist under Khan, told me those eight pricing consultants were working with over 250 clients, suggesting broad reach across the economy. Services included individually targeted pricing, segmentation of customers based on their profiles, and ranking tools that alter what products people see atop a web page or search. A customer clicking on fast shipping, for example, suggests an urgency that would lead them to tolerate higher prices.

The key to surveillance pricing is data. Nguyen and Sam Levine, the FTC’s former head of the Bureau of Consumer Protection, recently wrote a paper about how customers deliver that data when they sign up for loyalty programs. Enticed by promised discounts and concierge treatment, customers consent to data collection that allows companies to build intricate social graphs; one customer profile created by the grocery chain Kroger stretched to 62 pages. The discounts often aren’t maintained or are curtailed, loyalty card fees expand over time, and the more loyal a customer, the more data is collected and the more they pay over time, according to the report.

One case study is the McDonald’s app, which has 185 million users and provides seamless access on smartphones, our personal data-spewing machines. According to a recent earnings call, after downloading the app, a customer goes from 10.5 annual McDonald’s visits on average to 26. McDonald’s clearly wants its customers on the app. Its popular Monopoly game gives out stickers on the packaging of items like Big Macs and fries; you collect the right ones to win big prizes. But this year, instead of filling out a physical game board, Monopoly pieces now must be scanned into the app.

“This is about taxing people who don’t turn over their data, and manipulating people who do,” Levine said. “Do we really want a world in which you will have to pay a premium if you want to shop anonymously?”

Surveillance pricing is just one technology that companies use to set prices. Surge pricing accelerates supply and demand to raise prices when more people want the product or service. Subscription pricing offers products for a monthly fee, a steady stream of revenue from absent-minded customers that can feature deceptive sign-ups and impossible cancellation policies. Trump’s FTC sued Uber for a subscription service that takes as many as 23 screens and 32 actions to cancel; Amazon Prime’s cancellation policy, the subject of a separate FTC settlement that cost the company $2.5 billion, was internally nicknamed “Iliad Flow,” like the tortuous war Odysseus waged in the Greek epic.

Electronic shelf tags at Walmart and other grocers have triggered fears of endlessly changing prices while you shop. FIFA has confirmed use of differential pricing for World Cup events based on demand, something New York City Mayor-elect Zohran Mamdani has criticized. “They can become a tool for pure extraction and leave everybody paying what they shouldn’t,” said Kevin Erickson of the Future of Music Coalition, which monitors event ticketing.

One of the most insidious uses of technology is on the wage side. An issue brief from the Washington Center for Equitable Growth focused on how companies are using personal data to set worker pay, something first seen in the ride-hailing industry. Twenty AI consultants offer surveillance wage services to clients in logistics, manufacturing, retail, finance, education, transportation, technology, and health care, which is “the new Uber in a lot of ways,” said Veena Dubal, a law professor at the University of California and co-author of the report.

The AI systems can adjust worker pay in real time, and offer different wages to people doing the same work. They can adjust wages based on performance data, customer feedback, and even factors off the job, such as “predictive analytics that attempt to determine a worker’s potential tolerance for low pay,” the report notes. Most of these wage-setting practices are opaque to workers or their bargaining representatives.

“This is becoming normalized,” Dubal said. “It fundamentally undermines the well-worn and very American idea that hard work should result in higher pay.”

The result of these strategies is that supply and demand no longer solely determines price, as in textbook economics. AI-based pricing has become more critical than sale volume or product quality. Customers seeking a fair deal are simply outworked, unable to avoid being targeted. “There used to be moments where you really blew it, you had to buy a last-minute airline ticket and they’ve got you,” said Tim Wu, former competition policy chief in the Biden White House. “That’s really daily life now.”

Consumers used to benefit from what economists would call imperfect information. The uncertainty of defining the optimal price, and the ability in open markets for new businesses to undercut competitors, gave consumers a fighting chance to get a deal, or just to manage their life without being tracked and prodded. Technology eliminates that information inefficiency. “What these technologies are about is eliminating all risk for the shareholder,” Hepner said. “There’s no more error. It is a well-oiled extraction machine.”

EVEN ADVOCATES HAVE BEEN SURPRISED by the furiousness of the response to technology-aided pricing. As post-pandemic inflation swelled, AI trickery was a tangible, easy-to-understand depiction of an economy rigged against ordinary people, and the lack of transparency and unpredictability vexed consumers. “It was hiding in plain sight and everyone has found it and you just can’t unsee it,” said Lindsay Owens, executive director of the Groundwork Collaborative.

Politicians looking for anti-inflation messaging rode the wave of constituent anger. Sen. Ruben Gallego (D-AZ) challenged Delta’s palpable glee over surveillance pricing in an earnings call, and when Delta responded that they wouldn’t actually target customers using their personal data, Gallego didn’t let up. After all, Delta’s president Glen Hauenstein said on the earnings call that “to get [travelers] the right offer in your hand at the right time” is the “Holy Grail,” which sure sounds like surveillance pricing. “Delta is telling their investors one thing, and then turning around and telling the public another,” Gallego said in a statement.

House Democrats proposed a bill to ban surveillance pricing and surveillance wage-setting. So-called “drip pricing,” where junk fees are added during a sale, was effectively banned in event ticketing and hotel stays by an FTC rule finalized in May. But with GOP control of Washington, the real action has moved to the states. Surveillance pricing bans were introduced in California, Colorado, Georgia, Illinois, and Minnesota; surge pricing bans for grocery stores and restaurants were introduced in New York and Maine. By July, 24 states had seen bills introduced over some form of technology-aided pricing, according to a tracker from Consumer Reports.

In a major win this year, the nation’s two largest blue states explicitly banned algorithmic price-fixing, a tactic famously used by RealPage, which aggregated data from landlords across a city and recommended coordinated rent hikes. Joe Biden’s Justice Department sued RealPage, reaching settlements with corporate landlords vowing to end the practice. But data aggregators like Agri Stats and PotatoTrac already do this for meat and produce, and states wanted to prevent tech-enabled collusion from expanding.

Advocates were pushing on an open door; housing is the biggest monthly expense people have. “When RealPage came out, it quickly mapped onto something people feel viscerally,” said Samantha Gordon, chief advocacy officer at TechEquity, which lobbied on the California tech pricing bills. Moreover, since price-fixing is illegal when human beings enter a smoke-filled room and collude, extending that to algorithmic collusion made sense to lawmakers.

That was the concept behind California’s AB 325, which states that “common pricing algorithms,” whether informed through public or nonpublic competitor data, violate the Cartwright Act, California’s antitrust law. The law lowers the evidence standard to prove an algorithmic pricing conspiracy. It also prohibits anyone from distributing a pricing tool that coerces the setting of a recommended price, protecting small businesses along with consumers, explained Hepner, who was active in drafting the bill. “We saw mom-and-pop businesses swept up into these pricing schemes,” he said. “If you want to get the buy box on Amazon, you have to use their smart pricing tool. You have to give up independent decision-making authority to participate in the economy.”

After seven rounds of amendments, AB 325 passed and Gov. Gavin Newsom (D-CA) signed it in October. Days later, Gov. Kathy Hochul (D-NY) signed S.7882, another prohibition on coordinated price-fixing via software, though limited to rental housing. The New York law also blocks landlords from using algorithms to set “lease renewal terms, ideal occupancy levels, or other lease terms and conditions.” Cities like Jersey City, Philadelphia, Minneapolis, and Seattle have enacted similar citywide collusion bans, and 51 algorithmic price-fixing bills were introduced nationwide in just the 2025 legislative session. It’s “a potentially very powerful first step in tilting some of this economy back in favor of consumers,” Hepner said.

Surveillance pricing bills have not seen the same success, though their failure revealed critical information. In California, AB 446, which would have banned pricing based on personal data, was held back, giving sponsors time to build support next year. That’s because businesses across the economy, even ones not suspected of surveillance pricing, swarmed Sacramento to defend their practices. “It was fascinating over the course of that legislative process to see the opposition come out and openly say, ‘Hey, we’re actually doing this, you mean we can’t do this anymore?’” Hepner told me.

Some economists allege that tech pricing crackdowns threaten discounts and other potential benefits for consumers. But it’s hard to believe that companies would hire the most sophisticated engineers to figure out how to reduce their revenues. As Nguyen told me, the argument to preserve discounts boils down to this: “You can have privacy or low prices, but you can’t have both.”

A modest breakthrough came in New York, which passed a disclosure law requiring that relevant transactions include a pop-up stating: “This price was set by an algorithm using your personal data.” The National Retail Federation sued to block the law, but in October, a federal judge dismissed the case, ruling that the disclaimer “is plainly factual.”

Advocates believe that the legal victory will give policymakers permission to revisit the topic. But the fierce pushback showed that AI-based pricing was no longer a speculative harm. “So many companies and industries trying to kill [surveillance pricing bans] told us everything we need to know on why everything is getting more and more expensive,” Gordon concluded.

RealPage got the message as well. As the price-fixing bills worked their way through statehouses, the company announced it was acquiring Livble, a property management software tool. With Livble integrated into their platform, RealPage said, residents could subdivide rental payments into installments based on real-time cash flow landlords can see. In other words, RealPage was trying out surveillance pricing.

THESE TECHNOLOGICAL SHIFTS FROM ONE SCHEME to the next make it difficult for policymakers to stay ahead of the curve. For example, Lyft, one of the early adopters of surge pricing, reacted to rider anger about the service being more expensive precisely when they needed it most by moving to something called Price Lock, a subscription-based pricing tool that caps fares on targeted routes—for a monthly fee.

Another tactic involves redefining basic terms. When Delta claimed it would not use personal data, it added that it was merely “enhanc[ing] our existing fare pricing processes” with AI. “The lack of regulations about this and the way people’s data can be collected and used allows them to play around with what they are calling personal data and use,” said Ben Winters, director of AI and privacy at the Consumer Federation of America.

This forces policymakers to think more broadly, even on the outer edges of the possible. “I do worry that we’re always going to be chasing the companies,” said Sam Levine. “I think companies need to advertise a price publicly that’s available to all consumers … there should be a basic principle to advertise a price and it’s deceptive to charge more than that.”

Levine’s testimony to Congress this summer hits on another option: limiting collection of personal data, rather than just its usage. “‘Surveillance pricing’ is only possible because of how companies collect, share, and weaponize our personal data,” he told lawmakers, while arguing that relying on outdated tools to safeguard privacy risked further abuse.

Consumer protection and anti-monopoly experts are warming to this idea of cutting the data off at the source. Companies, after all, provide a full road map of what they collect in the privacy policies almost nobody reads. Levine and Nguyen’s loyalty program paper documents privacy policies, finding that rental car giant Hertz admits to collecting demographic and behavioral data, Home Depot tracks Wi-Fi usage inside their stores, and Macy’s captures customer driver’s licenses. Much of this data is sold to third-party data brokers, who share it with other businesses.

A data minimization standard could prevent use of personal information for pricing, Winters suggested. Mekki added that antitrust law recognizes that antitrust remedies often call for disgorgement of whatever gives companies an unfair advantage in the marketplace. Applied to surveillance pricing, that would argue for destroying the data allowing them to price-set. “If you can be forced to give up money or property, it stands to reason that you need to give up data,” she said.

The bipartisan American Privacy Rights Act introduced last year would have limited collection of certain kinds of data and given people the ability to opt out of having their data used to target them or sold to third parties. But this is at least the third major comprehensive privacy bill in the internet era that’s gone nowhere in Congress; too many companies are invested in the surveillance economy to let it be legislated out of existence.

“This is a textbook example of people not getting what they want,” said Wu, whose recent book The Age of Extraction highlights Big Tech platform-ization multiplying across the economy. “Is there anyone in America who wants less privacy or thinks it’s fine, or really wants targeted ads? The constituency is zero … In 20 years of privacy laws, there hasn’t been a single vote. This is not even a conspiracy. The tech industry lobbyists specialize in one thing: killing privacy laws.”

Some states have fared better, with a dozen passing limitations on data collection or letting consumers opt out of targeted ads or sale of personal data. California updated its policy this legislative session by passing AB 566, which allows internet users to opt out of data collection at the browser level, rather than having to go website by website.

But practically all privacy legislation has carve-outs for “bona fide loyalty programs,” giving companies an escape hatch under the guise of offering discounts. Even given that, however, loyalty programs are not exempt from “excessive” collection under these statutes that is not reasonably necessary.

Maryland recently finalized a privacy law that prevents companies from conditioning loyalty programs on the sale of data, despite industry pushback.

The best way to get at this activity might not be to go after the data, but rather those who receive, share, and process it. That includes companies acting as data brokers by selling and trading data, and the third-party pricing consultants they share the data with. Data sales could be considered unfair or deceptive under state consumer protection laws. And even when customers consent to having their data collected, they may not be consenting to transferring that data to third parties. “State enforcers should be thinking, under what circumstances is it lawful for companies to share personal data with pricing consultants?” Levine said.

Under Lina Khan, the FTC made headway by looking at these agglomerations of data as unfair practices, and subsequently prohibiting the sale of sensitive data, like when someone visits an abortion clinic. Today’s FTC has walked away from this enforcement, but their past work outlines a path for states to argue that targeted pricing is similarly unfair.

AI PRICING WON’T LOOK THE SAME IN FIVE YEARS as it does now, but we have hints of where it’s going. That’s why the role of Fetcherr is more interesting than Delta’s meandering explanations for how it uses AI. Airlines have been at the forefront of every pricing innovation of the past 30 years, from junk fees to differential pricing to loyalty programs. What’s the next frontier?

Fetcherr, which closed $90 million in funding in 2024 and another $42 million this year, tells you right on its website: “hyper-personalization at scale.” (Or at least, it used to tell you that, before it scrubbed this section.) “We’re talking about understanding each customer as an individual, optimizing every interaction for maximum value,” the now-redacted section states, detailing how its AI model uses data points like “customer lifetime value, past purchase behaviors, and the real-time context of each booking inquiry” to give each passenger their own product bundle at a bespoke price. The goal isn’t just a pleasant customer experience—it’s “revenue growth.” Robby Nissan, a Fetcherr co-founder, has said publicly that its AI systems can “manipulat[e] the market in order to gain more profit.” There’s not a lot of mystery here.

There are other airline pricing consultants, like Peter Thiel–backed FLYR and PROS and Air Price IQ. But the pitch is the same: maximizing travelers’ willingness to pay by analyzing reams of data. If you want to see the dystopia of perfect information accurately predicting exactly how much you’ll pay for a service, just book a flight this Christmas. “It’s a very real example of how AI-based pricing schemes put consumers into economic silos,” Hepner said. “You cannot comparison shop anymore. You cannot predict what a price is supposed to be anymore.”

This is facilitated by a concentrated economy where markets aren’t as open to alternatives to AI-optimized schemes. But AI pricing may shrink choice even further. For example, Delta recently stopped flying from LAX to London Heathrow Airport, despite the popularity of the route. Virgin Atlantic still makes the LAX-Heathrow trip; it’s a joint venture partner with Delta, and perhaps as important, both of them use Fetcherr. Was canceling the route an AI-enabled suggestion to benefit another client? We will never know.

Large market models running millions of price tests with thousands of input signals could short-circuit restrictions on surveillance pricing or data minimization. If it’s all just simulations, after all, is it really using personal data?

Another future is glimpsed in the rise of AI agents that shop for you. ChatGPT users will soon be able to link their PayPal accounts for instant purchases based on chatbot recommendations. Walmart has partnered with OpenAI for e-commerce as well. Autonomous bots with a linked bank account could even shop on their own, something a new AI agent lobby is trying to facilitate in Washington.

There’s little transparency on how products are recommended, how prices are derived, or whether AI agents will act in the interest of consumers or the retailers they’re partnering with. “You’re delegating purchasing power to one algorithm to interface with another,” said Ben Winters. “You’re getting people signed up into these systems where they have no control.”

But there’s also a brighter future possible, one where the backlash to nonstop data collection and pricing subterfuge accelerates, and people simply demand a fair price. Polling shows that the public is deeply skeptical that AI will work in their favor. Consumers may not have much control, but they do hold the dollars companies need to thrive, and they can withhold them from companies that treat them poorly.

“I think it’s possible that this is headed back to cost-based pricing,” said Owens, whose book Gouged about the new era of pricing releases next year. “It’s the way companies priced for decades … There is a world where the outcome is transparent, public, predictable pricing.”

This article appears in Dec 2025 Issue.