Computer Vision for a Single Store: Is It Worth It Under 10 Locations

The economics are about amortisation

Shelf-monitoring computer vision has a large fixed cost — training and validating a detection model for your fixtures and products, building the alerting and dashboard, integrating with store systems — and a small per-store cost once that exists. For a single store, or a handful, there is almost nothing to spread the fixed cost across, and the maths usually does not work. This is a build-versus-buy question before it is a technology one.

What the reference case assumes

Our computer vision project ran across 150-plus stores. The low per-store hardware and integration cost — a portable runtime, a low-power edge accelerator, a structured event instead of an image — is what made it scale past a pilot. Those choices reduce the marginal cost of store number 151; they do nothing for store number one's share of the build.

For one store, the alternatives are strong

A single store's staff can see most of the floor most of the time. Disciplined replenishment routines, a short scheduled walk of the key aisles, and good back-room organisation address a large share of what vision would catch. The gap between "staff walking the aisles" and "cameras watching them" is much smaller at one location than at a hundred.

Off-the-shelf products for small estates

Where a small operator does want automation, a packaged shelf-monitoring product — subscription-priced, pre-trained on common categories, with its own app — spreads a vendor's build cost across all their customers. It will fit less precisely than a custom system, but it does not ask one store to fund a bespoke model.

The signals that change the answer

A custom build starts to make sense for a small estate when the format is unusual enough that no packaged product fits, when the few stores are very large, when shrink or availability losses on specific lines are severe and measurable, or when the operator is on a clear growth path and building now avoids rebuilding at scale.

Piloting without over-committing

If a single store wants to test the idea, the right scope is a narrow pilot — one or two high-value aisles, an off-the-shelf tool or a tightly limited custom build — with a pre-agreed measure of success. Spending a full custom-build budget to find out whether it helps is the mistake.

A worked example

A three-store grocery operator is quoted a custom shelf-monitoring build. The fixed cost, divided by three stores, works out to more per store per month than the staff time it would save on aisle checks. They instead adopt a subscription shelf-monitoring app on their top two aisles per store, get most of the availability benefit, and keep the option to build custom if they reach fifteen or twenty stores.

The rebuild risk cuts both ways

Building custom too early wastes budget on amortising a fixed cost across too few stores. But adopting a packaged tool and then outgrowing it means a migration — new hardware, new integrations, staff relearning a different system. The judgement is about how firm the growth plan is. A vague ambition to expand should not drive a build; a funded, dated plan to reach twenty or thirty stores within a couple of years reasonably can, because building the reusable core now is cheaper than a disruptive switch later. Absent that certainty, start packaged and keep the option open.

Where this stops being right

• An unusual format that no packaged product supports can justify a custom build even at low store counts.
• Very large single stores — a flagship or a hypermarket — have enough shelf to monitor that the per-store maths improves.
• A fast-growing operator may rationally build ahead of the store count to avoid a disruptive migration later.

FAQ

Why does store count matter so much? Because the cost is mostly a one-time build. Across many stores it is cheap per site; across one or two it is a large fixed cost with almost nothing to amortise it against.

What should a single store do instead? Tighten replenishment routines and scheduled aisle checks, and if automation is wanted, use a subscription shelf-monitoring product that spreads the vendor's build cost across many customers.

When is a custom build justified below ten stores? An unusual format no product fits, very large stores, severe measurable losses on specific lines, or a clear growth trajectory that makes building now cheaper than rebuilding later.

ISTRALLEN builds custom computer vision where the store count justifies it, and will say so when an off-the-shelf tool is the better call — see AI for Retail.