What does a 35 percent drop in out-of-stocks actually prove? DeCicco & Sons said as of 10 November 2026 that out-of-stocks had fallen 35% within 60 days at pilot stores, and that it would run Simbe's shelf-scanning platform across its New York and Connecticut stores. The same release reports up to a 28% cut in the average time an item stayed out of stock. For a brand team, the number to copy is not the 35%. It is the gap the case exposes between what an inventory system believes is on the shelf and what a shopper can actually pick up (PR Newswire).
Thirty-five percent is a measurement result, not a shelf result
Tally, the autonomous scanner, records shelf conditions several times a day and pushes that data through APIs into the retailer's existing systems, where it informs e-commerce availability, supplier decisions and store operations. Simbe says more than 75 retail banners across nearly a dozen countries use the platform, with more than 3,000 stores under contract. The mechanism matters more than the percentage: the retailer is not adding a new dashboard, it is feeding shelf truth into systems that already drive replenishment.
| Reported metric | Value | What it actually measures |
|---|---|---|
| Out-of-stock reduction | 35% in 60 days | Change at pilot stores against their own baseline |
| Time out of stock | Up to 28% lower | Speed of correction, not depth of the gap |
| Banners using the platform | 75+ | Adoption, not performance |
| Stores under contract | 3,000+ | Committed footprint, not live coverage |
None of those four numbers tells a brand what its own on-shelf availability is. A 35% improvement from a weak starting point can still leave the shelf worse than a competitor that never measured. The percentage describes a direction of travel. The level is what a category team has to establish locally, store by store, before it can negotiate anything.
System stock and shelf stock are two different numbers
The distinction is easy to see in a worked example. Take a beverage brand with 12 listed SKUs in a mid-size grocery store. Nine are on the shelf in the right place, one sits in the backroom, one is on a neighbouring shelf, and one was never delivered. On-shelf availability is 9 divided by 12, or 75%. The system in-stock rate is 11 divided by 12, or about 92%. The two figures differ by roughly 17 percentage points, and two of the three missing SKUs need a store task rather than a delivery (Goods Checker).
That gap has a name. Phantom inventory is stock the system counts but the shopper cannot buy. In the same example the phantom rate is 2 divided by 11, or about 18%. A brand that manages only to the system number will keep hearing that availability is fine while shoppers substitute, delay or leave.
Why the weighted number matters more than the average
Aggregation hides the same problem one level up. If 10 large stores generate half of a brand's sales and 500 small stores generate the rest, and availability falls to 80% in the large stores while holding at 96% in the small ones, the simple average is about 95.7% but the sales-weighted figure is 88%. Ten of the twelve missing points come from the large stores. Reporting the simple average is the fastest way to make a structural problem look like noise.
Live shelf data is turning into an input, not a report
The second shift is on the demand side of the same data. Instacart is rolling out Inventory Intelligence for retailers and a Store Excellence Portal for CPG brands, built on scan data from roughly 600,000 shoppers who visit stores more than 15 times per store per day on average, generating over 10 million unique data points daily, plus shelf imagery from Caper Carts. Combined with more than 1.6 billion lifetime orders, the system is designed to separate a product that is selling through from one that was never properly stocked (Restaurant Hospitality News).
For brands, the interesting part is the substitution view. Knowing that a shopper who could not find a SKU switched to a competitor rather than a different pack size is a different fact from knowing the SKU was missing. The first supports a distribution argument, the second supports a pack-size argument. Most legacy reporting cannot tell them apart because it only records what entered and left the store.
What a brand team should ask for before signing off
Three asks cover most of the risk. First, the level, not the delta: current on-shelf availability by store and by SKU, measured by visit rather than by system. Second, the phantom rate, so backroom and misplaced stock are counted as missing rather than as available. Third, the weighted figure, so the stores that carry the volume carry the weight in the number. Without those three, an improvement claim is unfalsifiable.
A fourth ask is worth adding where the retailer will share it: the correction time. A shelf that is empty for two hours and one that is empty for two days produce the same availability score on the day of the audit, but they cost very different amounts of lost sales.
FAQ
Does a 35% reduction in out-of-stocks mean availability is now good?
No. It means availability improved by 35% against that retailer's own starting point. If the start was 62%, the end is roughly 84%, which is still a material gap. Always ask for the level and the denominator before accepting the delta.
Why do system in-stock and shelf availability differ so much?
Because the system counts units it believes it holds, including backroom cases and misplaced stock, while the shelf counts only what a shopper can reach in the SKU's own facing. In a 12-SKU example the two figures were 92% and 75%, a 17-point gap.
Is robot scanning necessary to get this data?
No, but frequency is. Manual audits several times a year cannot separate a two-hour gap from a two-day gap. Continuous capture is what makes correction time measurable, whether it comes from a robot, a shopper network or a cart-mounted camera.
What should a brand do when the retailer will not share SKU-level data?
Fall back to visit-based audits on a fixed sample of stores and SKUs, and publish the methodology with the number. A transparent sample is more defensible than an opaque system figure, and it gives the retailer something concrete to respond to.
