According to Engadget, the Federal Trade Commission has sent Instacart a civil investigative demand, seeking information about its AI-powered pricing tool. This action follows a recently published pricing experiment study that found the online grocery delivery app gave different users different prices for the same items from the same store at the exact same time. Some testers saw prices up to 23 percent higher than others, with an average price difference of around 7 percent for the same list of items. The study suggested those higher prices could cost customers over $1,000 more per year. Instacart previously claimed the variances were due to retail partners conducting “limited, short-term and randomized tests” to understand consumers, enabled by its AI tool called Eversight. The FTC, while not commenting directly, told Reuters it was “disturbed by what we have read in the press about Instacart’s alleged pricing practices.”
How the AI pricing tool reportedly works
Here’s the thing: the core of this issue is a tool called Eversight, which Instacart acquired in 2022. Basically, it’s an AI platform that lets retailers run what they call “randomized pricing tests.” The idea, from a business perspective, is pretty standard: you show slightly different prices to different groups of people to see how they react. It’s a way to figure out the optimal price point without alienating your entire customer base at once. But the key claim from Instacart is that these tests aren’t using personal data—they say it’s not based on your demographics, your purchase history, or even real-time supply and demand. So, if that’s true, the algorithm is presumably segmenting users in a more anonymous, group-based way. The problem, of course, is that the outcome looks identical to personalized, or “discriminatory,” pricing from the customer’s perspective.
The very thin line between testing and gouging
Instacart’s defense hinges on the word “randomized.” But let’s be real. In the world of big data and AI, is anything ever truly random? Or is it just random within a carefully controlled segment that the algorithm has determined has a certain tolerance? The study found discrepancies as high as 23%—that’s not a rounding error. That’s a significant financial impact, adding up to potentially over a thousand bucks a year for a regular user. The FTC’s involvement suggests they’re skeptical of the “it’s just testing” explanation. They’re probably digging into whether the “randomization” somehow consistently disadvantages certain groups or neighborhoods, even if it’s not using explicit demographic data. Location data, for instance, can be a very powerful proxy.
Why this matters beyond groceries
This isn’t just about whether you pay $4.99 or $5.99 for a box of cereal. It’s a precedent-setting case for how AI-driven pricing will be regulated. Every major online platform, from ride-shares to travel sites, uses some form of dynamic pricing. The question regulators are now asking is: where does market optimization end and unfair deception begin? If an AI is constantly running thousands of micro-experiments on consumers without their knowledge, what’s the collective cost? Instacart might be the first target, but they certainly won’t be the last. The outcome of this probe could define the rules of the road for algorithmic pricing across the entire digital economy. And that’s a cart full of implications way bigger than just groceries.
