As of 8 November 2026, Instacart has begun rolling out two in-store software products that turn its shopper network into a continuous shelf audit: Inventory Intelligence for retailers and the Store Excellence Portal for CPG brands. The company says roughly 600,000 active shoppers visit participating stores more than 15 times per store per day, generating over 10 million unique data points daily across more than 2,200 banners and nearly 100,000 stores. For brands that still buy availability data by the quarter, the offer is a different unit of measure: an empty slot counted at store level, priced in lost basket value, refreshed continuously rather than recounted monthly.
What Instacart put on the shelf this week
The retailer product renders a virtual walkthrough of any aisle and flags shelf gaps, high-velocity items at risk of depletion and planogram changes before they happen. The brand product goes further: it estimates lost sales revenue caused by out of stocks every day, at store level and at UPC level, and tracks whether a shopper who met an empty shelf switched to a competitor SKU or simply chose a different pack size of the same product.
That second question is the one most availability reports never answer. A stockout rate tells a brand how often a product was missing; it does not tell the brand where the demand went. If shoppers substituted within the brand, the loss is smaller than the rate implies. If they walked to a competitor's facing, the loss compounds, because the substitution may persist after the shelf is refilled.
Six hundred thousand shoppers as a sampling frame
The measurement engine is not a new sensor network. It is the existing labour force. Shoppers already walk every aisle to pick orders, and each visit produces scans and shelf imagery, augmented by Caper Cart smart carts that use computer vision from Arpalus, a company Instacart acquired to build this capability. The claim is that the data is a by-product of work that was happening anyway.
What the scan cannot see
A scan shows whether a facing is occupied. It does not show whether the product behind it is in the back room, whether a replenishment task is already queued, or whether the gap exists because the item was delisted. Those distinctions matter for a brand deciding whether to escalate to a buyer or to a distributor, and they are the reason availability data still needs a human check before it becomes a claim.
Store coverage is the second limit. More than 15 visits per store per day sounds dense, but density is unevenly distributed: high-volume stores in dense metros generate far more passes than a suburban store with a thin order book. A brand reading a national availability number should ask whether the store mix in the sample matches its own distribution footprint.
Lost sales priced per store and per UPC
Turning an empty shelf into a currency figure requires assumptions about what the shopper would have bought. Instacart anchors those assumptions in more than 1.6 billion lifetime orders, which gives it substitution patterns by category, pack size and price tier. The output is an estimate, not an invoice, and the difference matters when the number travels from a dashboard into a trade negotiation.
Advantage Solutions, one of the early users, said access to the portal lets its teams prioritise store visits by sales opportunity and send field resources to the shelves most likely to recover revenue. That is a operational change: field visits stop being allocated by rotation or by store size, and start being allocated by estimated recoverable value.
For a brand, the same logic applies inside its own sales force. If the portal can rank stores by lost revenue rather than by stockout percentage, the visit plan changes. A store with a modest stockout rate on a high-margin line can outrank a store with a severe rate on a low-value line, which is the opposite of how most availability reports are sorted today.
Why periodic counts kept failing
Traditional inventory systems rely on estimates and periodic counts, as Instacart's chief commercial officer put it, which leaves both grocers and brands with an incomplete picture of what is actually on the shelf. Electronic shelf labels, handheld scan guns and point-of-sale depletion models each capture a partial view, and none of them is refreshed continuously.
The labour trade behind availability
Continuous shelf data does not remove the labour cost of fixing a gap; it moves the cost from detection to response. Once a gap is visible within hours rather than weeks, the constraint becomes whether a store has someone to fill it. That is why the same data can improve one retailer's numbers and leave another unchanged: the difference sits in replenishment capacity, not in measurement.
For suppliers, this shifts the conversation with retailers. Arguing about a stockout rate is a dispute about measurement. Arguing about recoverable sales in specific stores is a dispute about execution, and it usually ends with a joint store visit rather than a spreadsheet exchange.
Building an availability record that survives the quarter
The practical question for a brand team is what to keep. Continuous data is only useful if it is stored in a shape that can be compared across periods, because the value comes from the trend rather than the snapshot. A record that keeps the store, the UPC, the date, the observed state and the response is enough to build that trend.
| Field to keep | Why it matters | Review cadence |
|---|---|---|
| Store and banner | Separates format effects from brand effects | Weekly |
| UPC and facing count | Distinguishes delisting from a gap | Weekly |
| Estimated lost sales | Ranks stores by recoverable value | Weekly |
| Substitution outcome | Shows whether demand left the brand | Monthly |
Kept this way, the record answers questions that a quarterly audit cannot. It shows whether a gap was fixed within a day or within a month, whether the same stores recur, and whether substitution is drifting toward competitors. None of those answers require new technology once the fields are fixed; they require the discipline to keep collecting them after the launch attention fades.
FAQ
Does continuous shelf data replace physical store audits?
No. It changes their purpose. Continuous scans show where gaps appear and how long they last, while physical audits confirm whether a gap is a replenishment failure, a delisting or a back-room stock problem. The two answer different questions and are most useful when paired.
How should a brand read an estimated lost sales figure?
As a ranking tool rather than an accounting number. The estimate depends on substitution assumptions drawn from order history, so it is reliable for comparing stores and categories, and less reliable as a single absolute figure to put in a contract.
What is the main sampling risk in shopper-generated shelf data?
Uneven store coverage. High-volume stores generate far more passes than thin-volume ones, so a national availability rate can be dominated by a subset of stores. Brands should check whether the store mix in the sample matches their own distribution footprint before acting on the number.
Why does on-shelf availability still lag distribution coverage as a metric?
Because distribution coverage counts doors won, which is easy to report at launch, while availability counts positions selling, which requires continuous observation. A listing across hundreds of doors can still underperform if facings are empty when shoppers arrive.
