Ninety-eight percent sounds like a solved problem. As of November 10, 2026, Kroger's radio-frequency identification pilot in its bakery division is reporting 98 percent read performance across roughly 1,500 stores, inventory counts nearly 70 percent faster and a 50 percent reduction in waste, while the chain shares computer vision data with suppliers to lift on-shelf availability. The number worth watching is not the 98 percent but the cost of the two percent that still reads wrong, because that is the part a shopper sees as an empty facing.
What 1,500 bakeries prove about the rest of the store
The pilot numbers come from Main Street's report on Kroger's investor day, where chief store operations officer Mark Ibbotson described the goal as less counting, better information, lower waste and higher availability. It is a useful reminder that availability is not a single system output. It is the end of a chain that starts with a tag being read, passes through a replenishment decision and finishes with a picker finding the item where the system said it would be.
A read rate is a counting promise
A 98 percent read rate says the technology recognised 98 of every 100 tagged items it passed. That is a statement about sensors, not about shelves. A store can read every tag perfectly and still hand a shopper an empty position, because the carton is in the back room, the facing was not restocked after the morning rush, or the item was scanned into the wrong location during delivery.
Where the aisle and the back room disagree
The gap between system stock and shelf stock is the number that decides whether an online order completes. Most retailers measure it once a quarter with an audit, which is too slow for a channel that promises delivery in under an hour. Turning a shopper network or a robot into a continuous count is the current answer, and it explains why shelf data is being sold as an operating product rather than a monthly report.
Suppliers get the signal before the shelf does
Sharing vision data with suppliers changes who acts first. In the old model the retailer noticed a gap at the audit, issued a replenishment order, and the supplier learned about the problem weeks later in a category review. In the new model the supplier can see a specific facing at a specific store on a specific day, which is useful and also uncomfortable, because it removes the argument that nobody knew.
The commercial question is who pays for that visibility. Kroger has not published a price for supplier access, and neither have the platforms selling shelf audits to brands. A mid-sized brand should test the data against its own store list before building a plan on it, because a national availability figure reflects the sample behind it as much as the state of the shelf.
Dark-store economics set the ceiling
Quick-commerce operators solved the same problem by removing the aisle. A dark store of 1,500 to 3,000 square feet carries 2,000 to 5,000 SKUs, sits within two to three kilometres of its customers and targets a pick under two minutes and a handoff under five, according to an operating breakdown of the model. The radius is not a marketing choice, it is arithmetic, because five kilometres pushes a two-wheeler past the thirty-minute promise.
Two kilometres is an operating rule
That arithmetic explains why store-based retailers cannot simply copy the promise. A supermarket with 20,000 SKUs and a 5 kilometre trade area has more to sell and further to travel, so its availability problem is bigger and its delivery window is longer. The realistic target is not ten minutes but a pick rate the store can sustain during its own peak, which is usually the same hour the online orders arrive.
| Measure | Kroger pilot figure | What it does not tell you |
|---|---|---|
| Tag read performance | 98 percent | Whether the item is on the shelf or in the back room |
| Inventory count speed | Nearly 70 percent faster | Whether the count happened during trading hours |
| Waste reduction | 50 percent | Whether the saving came from freshness or from ordering less |
| Rollout plan | More than 2,300 stores by 2029 | Cost per store and payback period |
Four numbers to keep after the pilot
None of these four numbers requires new hardware once tags are in place. They require a definition that survives a change of vendor, and a routine that runs on the worst day of the week rather than the calmest. The pilot is the easy part, because a 98 percent read rate is a technology result that arrives with the invoice; the four numbers above are an operating result that has to be earned every week.
FAQ
Does a 98 percent read rate mean availability is solved?
No. It means the technology recognised 98 of every 100 tagged items. Whether the item sits on the shelf, in the back room or on a pallet is a separate measurement, and that is the one a shopper experiences as a gap.
Why share shelf data with suppliers at all?
Because the supplier can fix the cause and the retailer can only fix the symptom. A detected gap at a specific store tells the supplier whether the problem is production, allocation or the delivery schedule.
Can store-based retailers match quick-commerce delivery times?
Not with the same promise. A supermarket with 20,000 SKUs and a 5 kilometre trade area has a longer pick and a longer ride, so the realistic target is a pick rate it can hold during its own peak.
What should a brand ask before buying shelf-audit data?
Three things: whether the store sample matches its distribution, how often each store is actually visited, and whether substitution is measured or only the empty facing.
