Key Takeaways

  • Groundwork Collaborative, Consumer Reports, and More Perfect Union recruited 400 secret shoppers across four US markets to audit Instacart orders simultaneously.
  • 75% of items had different prices for different shoppers buying from the exact same store at the exact same second.
  • Individual item prices varied by as much as 23%, creating an estimated annual penalty of up to $1,200 for identical family grocery baskets.
  • Instacart's acquisition of AI pricing platform Eversight gave grocery chains automated tools to conduct continuous A/B pricing tests on consumers without disclosure.

The 400-Shopper Safeway Test

For over a century, retail commerce rested on a simple contract: the price on the shelf is the price everyone pays. That contract is dead online.

Lindsay Owens and her team at Groundwork Collaborative partnered with Consumer Reports and More Perfect Union to measure what happens when identical consumers open the Instacart app at the exact same moment. They recruited 400 secret shoppers across four US locations, directed them to open the exact same store catalog, and logged the prices.

“What we found: 75% of the items were differently priced for different shoppers, same time, same place,” Owens said.

This was not regional pricing. It was not a surcharge for someone ordering milk in Manhattan versus rural Ohio. Two shoppers standing in Washington, D.C., looking at the exact same Safeway at the exact same minute, saw wildly divergent numbers.

“The differences were as big as 23%,” Owens explained. “So you could be charged 23% more for a dozen Lucerne eggs than I was charged from the same Safeway in Washington, D.C. at the exact same time.”

Over fifty-two weeks of household shopping, small gaps compound quickly. “We estimated that because of these differences, some families could end up paying as much as $1,200 more in a given year,” Owens noted.

Eversight and the Turnkey A/B Pricing Machine

Grocery margins are notoriously thin, hovering around 1% to 2%. To squeeze out more margin, retail platforms moved away from static price tags and turned to continuous automated experimentation. Instacart accelerated this transition by acquiring Eversight, a retail pricing software company.

Owens pointed directly to that deal: “Because Eversight was an AI company that built a platform to allow retailers to run pricing experiments on their own platforms.”

Software teams run A/B tests on button colors, onboarding flows, and headline copy every day. Eversight gave enterprise grocers that exact same capability for shelf prices. Instead of changing a storewide price tag on Tuesday morning, an algorithmic pricing engine can silently test price elasticity on a live user base in real time.

If the engine detects that a user buys a specific brand of olive oil every Thursday, or that an account orders from a neighborhood with higher average incomes, it nudges the checkout price up. The user never sees the baseline. They only see what the algorithm decided they were willing to stomach.

Algorithmic Price Discrimination Destroys Retention

For digital platforms, dynamic personalization feels like pure margin optimization. In practice, it breaks the core trust that makes recurring software and commerce work.

When a consumer discovers that their co-worker paid $4 less for the exact same carton of eggs from the same physical aisle, the platform loses its utility status and becomes an adversary. The transaction stops feeling like a helpful service and starts feeling like an extraction trap. Regulators are already circling these surveillance pricing practices, examining device signals, battery levels, and location tracking.

Optimizing short-term basket yield through opaque algorithmic splits burns the goodwill you need to keep users around for five years. Once buyers realize you are running live extraction experiments on their essentials, they stop trusting every checkout screen you show them.

What to Do With This

Audit your pricing architecture this week. If your checkout flow shows different prices to different users based on tracking signals, device data, or hidden A/B tests rather than explicit volume tiers or public discount codes, kill the experiment. Replace it with transparent, published pricing tiers that you would feel comfortable defending publicly to your most skeptical customer.