Will You Pay More Than Your Neighbour for the Same Thing? AI and the End of the Single Price

THE THESIS. The single price will not die because of a technical decision. Whether it dies or survives will be a political decision. AI already makes it technically possible to charge each customer based on their profile. What is still undecided is whether that will be legal, whether it must be disclosed, and whether consumers will have their own AI to fight back. Our reading is one of nuanced optimism. In the short term, the risk is real: opaque prices that squeeze customers using their own data, and algorithms that push prices up in a coordinated way. In the long term, the same technology can work in the consumer's favour. It can lower prices for those who have less and put a negotiating agent in every buyer's pocket. Which of these happens depends on the rules written now.

THREE THINGS THAT GET MIXED UP. Before forming an opinion, we need to separate three different practices. The first is dynamic pricing: the price changes with demand and is the same for everyone at that moment. That is how it works when a flight gets pricier as seats run out, or when Uber raises fares at rush hour. The second is personalised price discrimination, known in the US as «surveillance pricing». Here the price depends not on the moment but on who you are: your location, your browsing history or your device. The third is algorithmic collusion: several competitors use the same software, or algorithms that learn not to compete, and prices rise together without anyone signing a deal. Public debate tends to blur all three. Oasis is the clearest example. Britain's competition watchdog, the CMA, investigated ticket sales for the band's 2025 tour and found no algorithmic dynamic pricing. It found a tiered system: when the cheap tickets sold out, a pricier tier opened. Ticketmaster agreed to give 24 hours' notice of tiered pricing, without admitting any wrongdoing. The anger was legitimate, but the popular diagnosis was not accurate.

WHAT WE ACTUALLY KNOW. Evidence on personalised pricing exists, but it is patchy. In January 2025, the US consumer regulator, the FTC, published preliminary findings from its study of eight intermediaries that sell pricing tools. According to the agency, data such as precise location or browsing history is used to set different prices for different people for the same product. According to those summaries, the tools could raise company revenue by 2% to 5%. The most high-profile case came in December 2025. An investigation by Consumer Reports, Groundwork Collaborative and More Perfect Union, with 437 volunteers in four cities, found that Instacart showed different prices for the same item, in the same store, at the same time. It affected about 75% of items, with gaps of $0.07 to $2.56 per item. Instacart said these were A/B tests (showing different versions to different groups to see which works best) at 10 retailers. It said it did not use personal or demographic data, and it called «completely wrong» the researchers' estimate that a typical family paid $1,200 more a year. Even so, the company ended the tests. Delta, for its part, said in 2025 that Fetcherr's AI influenced 3% of its fares and that it aimed for 20%. In writing to several senators, it denied using individualised prices based on personal data. What this episode shows is that the technology is being deployed. What it does not prove is that individual prices are already the norm.

THE REGULATORY RESPONSE. Regulators are moving faster than in earlier tech waves. Since November 2025, New York has enforced the first law of its kind. It requires the notice «this price was set by an algorithm using your personal data», with fines of up to $1,000 per violation. On 19 August 2026, the FTC proposed an enforcement policy statement (a document that explains how it will apply the law). It does not ban personalised pricing, but it calls for disclosure when customers expect a uniform price and an explanation of what data is used. On 28 October 2026, it sent information orders to nine companies: Delta, United, JetBlue, Uber, Lyft, DoorDash, Instacart, Target and Walmart. They have 45 days to respond. This is a study, not an accusation, and Delta and JetBlue deny personalising fares. In the European Union, the Omnibus Directive already requires a notice when a price has been personalised by automated means. The Commission also plans to present the Digital Fairness Act this quarter, which, according to early analyses, would include targeted limits on personalised pricing. On collusion, the precedent is RealPage, software that recommended rents to landlords. In November 2025, it settled with the US Department of Justice. It paid no fine and admitted no guilt. But it can no longer use competitors' private data in real time, and it can only train its algorithm on data at least a year old. It will also be overseen by an independent monitor for three years. Critics such as the American Economic Liberties Project call the deal weak.

PIGOU WAS HALF RIGHT. In 1920, the economist Arthur Pigou defined three degrees of price discrimination. In the first degree, each customer pays exactly the most they would be willing to pay. In the second, customers choose for themselves from a menu, as with bulk discounts or economy and business class. In the third, the price varies by group, as with student or pensioner discounts. Until now, the first degree was purely theoretical because nobody could know each person's willingness to pay. AI brings it closer to reality. The theory says something uncomfortable for both sides. Discrimination can improve overall welfare if it lets sellers serve people who would be priced out at a single price. That is what happens with cheaper medicines in poor countries or free software for students. But in pure first-degree discrimination, all of the consumer's savings go to the seller. More gets sold, but the customer gains nothing. So the key question is not whether prices differ. The key question is who pays less, and why. If AI charges less to those who have less, it can be progressive. If it charges more to those in a hurry, those who don't compare or those who can't push back, it is extraction dressed up as efficiency.

MACHINES VERSUS MACHINES. This is the new element that changes the game. Shopping agents (AI that searches, compares and pays on your behalf) could hand power back to consumers. An agent never gets tired of comparing and does not give in to urgency. It can also spot when another profile is being offered a better price. But early evidence calls for caution. In Microsoft Research's Magentic Marketplace simulation, with 100 buyer agents and 300 seller agents, buyers showed a strong bias toward the first offer they received. According to the paper, this gave whoever answered first a 10- to 30-fold advantage. Several models were also open to manipulation. On the seller side, a study by Fish, Gonczarowski and Shorrer found that GPT-4-based pricing agents reached above-competitive prices on their own, without being told to. Changing seemingly harmless phrases in their instructions was enough to change how much they colluded. In other words, collusion can emerge without any human agreement, which makes traditional antitrust law much harder to apply.

OUR READING AND THE IMPLICATIONS. First, we reject the doom narrative: there is no solid evidence that you systematically pay more than your neighbour today because of your data. We also reject complacency: a 2% to 5% revenue boost spread over millions of purchases is a lot of money leaving consumers' pockets without their knowledge. Second, the most underrated risk is collusion, not personalisation. Personalisation harms individuals. Collusion raises prices for everyone at once, as in rental housing, the most sensitive household cost. Third, transparency is the floor, not the solution. A notice in New York does not tell you what others are paying. More useful rules would be different: ban the use of sensitive data (health, financial situation, vulnerability), give people the right to ask for the non-personalised price, and audit pricing algorithms the way we audit company accounts. Fourth, watch who controls the shopping agent. If the platform that sells also controls the agent, the conflict of interest is obvious. This ties into our thesis on agents: what matters is whom they serve, with what permissions, and who can revoke them. In the long run, the desirable scenario is plausible. A market where everyone has a tireless negotiator on their side, and those with less pay less, looks more like abundance than plunder. But that future will not arrive on its own. It requires independent shopping agents, clear rules against collusion, and regulators who measure before they ban.