AI Pricing Risks Surfacing
- Analysts warn AI-driven pricing systems could enable personalised prices that exploit consumers beyond regulators' reach. - The concern centers on adaptive, opaque pricing decisions that vary by user features and context. - Engineers building pricing or ranking systems may need stronger audit trails, logging, and fairness mechanisms. (earth.com)
Online prices are starting to act less like price tags and more like negotiations run by software. Regulators and consumer advocates say artificial intelligence can now change what one shopper pays based on personal data the shopper never sees. (ftc.gov) The Federal Trade Commission said on January 17, 2025 that its surveillance-pricing study found companies can use data such as precise location, browser history, shopping history and even mouse movements to tailor prices for individual consumers. The agency said the same product can carry different prices or promotions depending on the buyer, the time and the sales channel. (ftc.gov) The Federal Trade Commission opened that study in July 2024 using its 6(b) authority and sent orders to Mastercard, Accenture, PROS, Bloomreach, Revionics and McKinsey & Co. The agency said those firms sit in the middle of the pricing system, selling tools that retailers use to algorithmically adjust offers. (ftc.gov) The basic practice is simple: a pricing model predicts what a shopper is likely to pay, then changes the offer in real time. The Organisation for Economic Co-operation and Development said pricing algorithms can improve efficiency, but they can also be used in ways that harm consumers or restrict competition. (oecd.org) That has moved the issue beyond theory. A December 2025 investigation by Consumer Reports and Groundwork Collaborative found some Instacart grocery prices differed by as much as 23 percent per item between customers shopping at the same time. (consumerreports.org) Consumer Reports said roughly three-quarters of the products it checked on Instacart showed different prices, with gaps ranging from 7 cents to $2.56 per item. Instacart later stopped offering the tool that let retailers charge different shoppers different prices for the same groceries at the same time. (consumerreports.org) States have started writing rules around the practice. New York said its Algorithmic Pricing Disclosure Act took effect on November 10, 2025 and requires businesses to tell consumers when a price was set by an algorithm using personal data. (governor.ny.gov) California lawmakers have also pushed tighter limits. Assembly Bill 446 in the 2025-2026 session proposed banning prices based in whole or in part on personally identifiable information gathered through electronic surveillance technology. (legiscan.com) The engineering problem is not just the model but the record around it. If a company cannot show what data went into a price, what rule changed it, and which customers were affected, auditors and regulators have little chance of proving whether software crossed the line from dynamic pricing into hidden discrimination. (ftc.gov) The result is a market where two shoppers can see the same item and get different numbers, with no visible reason why. The Federal Trade Commission’s study is still ongoing, and the fight is shifting from whether this pricing exists to how much of it companies will have to explain. (ftc.gov)
