As of November 13, 2026, the case for choosing fulfillment stores by footfall is getting harder to defend. Instacart's Inventory Intelligence and Store Excellence Portal turn shopper scans and cart cameras into a continuous, aisle-level reading of what is actually on a shelf, and that reading is the input a store-selection model has been missing. A store that looks strong on sales can still be the wrong place to promise a thirty-minute order.
Footfall is a proxy; the shelf is the measurement
A trade area file usually holds postcodes, a delivery radius, competitor density and a sales history. None of those fields says whether the store can pick an order. The new shelf tools add a live field instead: what share of the top items is on the shelf at eleven in the morning, and whether a gap is selling through or was never stocked at all (Instacart shelf-intelligence launch).
The scale behind that field is the reason it is credible. Roughly 600,000 shoppers walk these stores more than fifteen times per store per day, generating over ten million unique data points daily, and shelf imagery comes from both those shoppers and camera-equipped carts. Combined with more than 1.6 billion lifetime orders, the system can separate a fast seller from an unstocked facing.
| Store type | Signals used today | What shelf accuracy changes | Cost of ignoring it |
|---|---|---|---|
| Dense urban discount store | Sales per square foot, footfall | Whether narrow aisles can be picked without blocking shoppers | Late orders and pickers colliding with customers |
| Suburban supermarket | Basket size, delivery radius | Which aisles can carry an online basket end to end | Substitutions on the items that define the basket |
| Rural small format | Distance to next store | Whether the shelf can support a promise at all | A promise made where no substitute exists |
What a shelf map adds to a trade area file
It converts a geographic question into an operational one. Two stores can sit inside the same radius and score very differently once on-shelf availability is measured weekly on the items an app is most likely to surface. That is the difference between a high-potential store and a store that merely happens to be near the customer.
A picking store is not a busy store
Store selection has leaned on traffic because traffic was the only signal available at scale. A shelf score inverts part of that logic. A store with heavy foot traffic and a narrow aisle footprint is a poor picking location: pickers block shoppers, both slow down, and the order that was supposed to take twelve minutes takes twenty. The cost lands in the delivery window, not in the sales report.
The thirty-minute claim is a radius, not a target
Instacart advertises delivery in as fast as thirty minutes where available, and the retailer network already covers about 20,000 discount-format stores across 48 states (nationwide rollout). That coverage is a density argument, not a quality argument. Density decides whether the radius is reachable; shelf accuracy decides whether the order arrives complete.
Scoring stores on what they can ship
The score that survives contact with operations has three inputs: shelf accuracy on the top hundred online items, pick time per order under real aisle conditions, and the share of orders completed without substitution. All three are measurable weekly, and all three change faster than a trade area does.
A fourth input belongs beside them, and it is the one most networks skip: what the store would have sold anyway. Counting orders a store would otherwise have missed requires a comparison group of similar stores without the channel, held for at least a quarter. Without it, a launch looks like growth when it is only movement between channels.
Putting the score inside the store file
The practical change is small and unglamorous: the store file gains a shelf field that is refreshed weekly, and the field has an owner. Once it exists, a high-potential store list stops being a static ranking and becomes a queue that reorders itself when a shelf degrades.
None of this needs a new system. It needs the shelf number to sit in the same file as the sales number, updated on the same cadence, with a named owner who can move a store out of the promise. Everything else is a dashboard.
FAQ
Why is footfall a weak input for store selection?
Because traffic says how many people pass a store, not whether the store can pick an order. A busy store with narrow aisles can miss a delivery window while a quieter store with a clean shelf fills the same order in twelve minutes.
What should a shelf score measure?
On-shelf availability on the hundred items that define an online basket, median pick time under real aisle conditions, and the share of orders completed without substitution. Refresh it weekly.
Does a national rollout prove the model works?
No. Coverage across 20,000 stores in 48 states shows density, not quality. Whether the channel adds orders or moves them needs a comparison group of similar stores held for a quarter.
