Prodx · Retail Grocery Technology
Issue 05 · October 2026 · What runs beneath everything in grocery retail
On October 12, Greg Foran told Kroger employees the company would eliminate roughly 3,000 corporate roles over the next three years as it leans further into AI. “Technology is reshaping what’s possible,” he wrote. Reporting put the figure at something near 37% of the corporate workforce. The stated goal is to remove duplicative work, lower prices, and reinvest in store associates.
Kroger’s own framework is the place to start. At GroceryShop in September, CDO Yael Cosset described the company’s AI path in three stages, assisted, then augmented, then autonomous, and said the right level of AI depends on risk, repeatability, and the need for human judgement. That’s a more honest framing than most because it names the actual variable.
So here’s the thing, in grocery, human judgment has been quietly subsidizing product data for about thirty years.
“Those corrections never showed up in a system of record. They showed up as work.”
Somebody in the category knew the 12-pack was coded as a single. Somebody in merchandising knew the private label reformulation hadn’t made it into the attribute set yet. Somebody in e-commerce ops knew which three categories had unreliable sizes and compensated in the substitution rules. None of those corrections were in the data. They were in people, applied continuously, invisibly, and at real cost. Not free: that was senior category and merchandising judgment spent on record repair instead of assortment, negotiation, and margin. It just never appeared as a line item, which is why it was never challenged, and why the thing generating the errors was never fixed.
Assisted automation keeps those people in the loop, so the subsidy continues. Augmented automation thins it. Autonomous removes it entirely, by design. That’s the point of autonomous. And the defects those people were absorbing get promoted instead of resolved on the way out the door. They stop being someone’s unglamorous Tuesday and start being an input to a model that will act on them at scale, quickly, and without flagging that it had to guess.
This is not an argument against automating. The economics are what they are, and Kroger is hardly alone, for example, Albertsons announced its own expanded cost program the following day. It’s an argument about sequence. The window between announcing automation and reaching autonomy is the window in which the data either gets fixed or gets built on.
Which, as usual, sorts grocers into two groups. The ones who spend the next three years making the catalog good enough that judgment is no longer required to interpret it. And the ones who find out, somewhere around year two, exactly what those people had been doing all along.
On October 12, Ahold Delhaize disclosed a security issue and took systems offline as a precaution. Websites and apps for all five US banners went down and stayed down for days: Hannaford, Stop & Shop, Giant Food, Food Lion, The Giant Company. Stores remained open. Pharmacies and fuel stations kept running. And the banner sites, in their outage notices, pointed shoppers to third-party delivery marketplaces, where orders could still be placed. Read that again, because it’s the most interesting sentence in grocery this month: the first-party digital storefront was unavailable, and commerce continued on surfaces built from the same catalog, syndicated outward. We argued last month that your catalog had quietly become an API. October supplied the stress test. The front end turned out to be the fragile layer. The product data was the part that stayed standing, which is a good argument for treating it like infrastructure rather than like marketing collateral.
Albertsons’ Q2 told an uncomfortable story cleanly. Net sales were essentially flat at $18.9 billion, identical sales fell 0.8%, and digital sales rose 11%, with digital penetration now at 10.5%. In the same filing, the company attributes its gross margin decline primarily to higher delivery and handling costs tied to that digital growth, plus shrink. The one line that’s growing is the one eroding the rate. Albertsons’ response is a cost savings program raised to $2 billion over three years and a comprehensive review of nearly everything, and three new directors effective October 14, including Aligned Data Centers COO Chris Drumgoole, whose background is large-scale technology and AI transformation. Here’s the part the earnings call didn’t cover: a meaningful share of digital fulfillment cost is an avoidable error. The wrong item picked, the substitution the shopper rejects, the refund, the item that couldn’t be found because its attributes were wrong. You cannot cut your way out of delivery economics. You can stop paying for mistakes, and accuracy is the cheapest lever on that list.
Costco’s DoorDash partnership went live nationwide this month, putting roughly 4,000 SKUs from US warehouses into on-demand delivery. The warehouse model has always been able to get away with thin product content, and not by accident: the product sells off a pallet, in person, with the package doing the describing. There has never been much reason to invest in attribute depth for an item a member finds by walking past it. That changes the moment those SKUs have to rank in a marketplace search, resolve correctly against a dietary filter, and generate a sensible substitution when the item is out, competing against retailers who have been enriching catalogs for a decade. The format’s great operational virtue, minimal merchandising overhead, becomes a liability the instant the catalog has to perform as data rather than as a pallet.
Within forty-eight hours of each other, Kroger announced a 3,000-role reduction tied to AI adoption, and Albertsons raised its cost savings target to $2 billion over three years while launching a review of its cost structure, operating model, and banner portfolio. Both companies named digital as a growth engine in the same breath. Neither described technology investment as expansion.
That’s the vocabulary shift worth tracking, and it has a practical consequence for anyone who has to defend an infrastructure line item next year. In 2024 and 2025, data and AI work could be funded as enhancement of new capability, new experience, new revenue. In the budget cycle now opening, enhancement is the category getting cut. Cost avoidance is the category that survives. The same project described two ways gets two different outcomes: “enriching our product catalog to improve discovery” reads as discretionary, while “reducing substitution failures, refunds, and pick errors in digital fulfillment” reads as the cost program the CFO just announced. The work is identical. One of those sentences gets funded. It’s worth rewriting your 2027 asks before someone else frames them for you.
Prodx builds product data infrastructure for grocery retailers. We enrich, organize, and maintain the catalog attributes that power search, personalization, AI-assisted shopping, retail media, and digital shelf operations. Our customers include Hy-Vee, Wakefern, and Giant Eagle.