Instacart quietly let retail partners run algorithmic pricing tests on its platform through a tool called Eversight, which the company had acquired for $59 million back in 2022. The tool let retailers experiment with prices to see how shoppers would react. In practice, that meant two customers standing in the same virtual aisle, buying the same brand of the same product from the same store, could see different numbers on their screen.

Consumer Reports and the Groundwork Collaborative set out to test this directly. Volunteers shopped the same baskets at the same stores and screenshotted their carts without checking out. About three-quarters of the products checked were offered at different prices to different customers, with some products offered at as many as five different prices and variations ranging from as little as 7 cents to $2.56 per item. The testing wasn't limited to one chain either: investigators found evidence of price experimentation at Albertsons, Costco, Kroger, and Sprouts Farmers Market in addition to the original test store. This wasn't a rounding error or a stale cache. It was the system working as designed, just not in a way any shopper had agreed to or could see.

The dollar impact was real. CNBC reported that the total cost for the same basket of goods at a single store varied by about 7%, which could result in more than $1,000 in extra annual costs for customers. Instacart's initial defense was procedural: the company responded by saying that retailers determine prices listed on the app. That's technically true and also beside the point for a customer who just got charged more than the person shopping next to them for no disclosed reason. Once the Consumer Reports findings became public and federal regulators started asking questions, Instacart reversed course. The company said it would no longer allow retailers to run AI pricing tests on its grocery delivery platform after the practice drew widespread scrutiny, ending the use of its Eversight technology "effective immediately."

This wasn't happening in a vacuum. The FTC had already flagged the broader practice before the Instacart story broke, noting in a preliminary study that a handful of data intermediaries were working with hundreds of retail clients on similar personalized pricing schemes. The pattern across the pricing-AI failures we track is consistent: the technology isn't the problem, the lack of a disclosure and audit layer is. A pricing model that adjusts based on real-time signals is defensible. A pricing model that adjusts per-customer without anyone at the company being able to explain, on request, why shopper A paid $2.56 more than shopper B is not something you can defend to a regulator, a journalist, or your own customer service team.

If you're running any kind of dynamic pricing, personalization engine, or algorithmic test on customers, the action item this week is simple: pull a sample of transactions from the last 30 days and check whether you can reconstruct, in plain language, why any two customers paid different prices for the same thing. If you can't answer that question in one sentence, you have an unmonitored experiment running on your paying customers, not a pricing strategy. Fix the audit trail before you scale the tool, not after a consumer group finds it first.

Sources