Overview

shopreme Sentry AI is a machine-learning module that makes spot checks smarter. Instead of relying only on fixed probabilities and rule-based conditions, Sentry AI evaluates each Scan & Go purchase against a model trained on that store's own transaction history and returns a score that expresses how likely the basket is to contain a mistake.

That score is then used by the shopreme spot check logic as one more input when deciding whether a purchase should be marked for employee verification.

The goal is efficiency. A purely random selection, for example checking every tenth purchase, spreads employee attention evenly across all shoppers, including the overwhelming majority whose baskets are correct. Sentry AI concentrates the same number of checks on the purchases where a discrepancy is most likely, so stores find more faulty baskets without increasing the number of shoppers they stop.

How Sentry AI fits into the spot check logic

Sentry AI does not replace the spot check logic. It is an additional module inside it, in the group of custom modules that can be enabled per project.

The surrounding mechanism is unchanged and is described in Spot Check Logic for All Clients:

1. The shopper taps Pay in the Scan & Go app.

2. The backend runs through the spot check modules in a fixed order. Early exit modules (e.g. manual employee selection) can decide the outcome immediately.

3. The remaining modules raise or lower a probability value.

4. The Sentry AI module contributes its score to that calculation.

5. The resulting probability is compared against a random value to produce the final decision.

Because Sentry AI sits inside this chain rather than in front of it, all existing safeguards still apply. Theft-protected products still force a check, a paused store still pauses, a manual selection by an employee still takes precedence, and the minimum probability still keeps a residual chance of being checked on every purchase. The likelihood of being checked therefore remains unpredictable for the shopper.

What the model learns from

Sentry AI learns from the metadata of completed purchases in the store, aggregated at basket level: properties of the transaction as a whole rather than of each individual scanned item.

Two design decisions matter here:

• Basket-level aggregation: As some retailers have assortments running up to a million products, this could produce extremely large and sparse datasets that can be impractical to train and serve. Aggregating to basket level keeps the model small, fast and maintainable.

• A model per store. Shopping behaviour, assortment and layout differ from store to store, so the model is trained on the history of transactions for that store. What counts as an unremarkable basket in one location is not automatically unremarkable in another.

The model itself is trained offline, then loaded and served to the shopreme backend as an additional service.

Monitoring Sentry AI in the Management Console

Once Sentry AI is activated for a store, a dedicated Sentry page becomes available in the shopreme Management Console. It answers one question: Is the Sentry-powered targeted selection finding more discrepancies during a rescan than random selection would?

A green ML scoring active badge in the top right confirms that the module is live and scoring transactions.

Sentry Precision

This component shows how much more effective Sentry-selected checks are than random ones, expressed as a multiplier. The card compares two discrepancy rates side by side:

• the discrepancy rate for randomly selected baskets, and

• the discrepancy rate for Sentry-selected baskets.

The multiplier is the ratio between them. It answers: for the same volume of goods checked, how much more lost revenue did targeted selection recover?

Saved Spot Checks

The same result as Sentry Precision, but seen from the other direction. Rather than asking how many more errors the same number of checks finds, this card asks how many fewer checks are needed to find the same number of errors. It shows:

• the number of random checks that would have been needed to find the errors actually found in the period,

• the number of Sentry-selected checks that were needed, and

• the difference — the checks saved.

Checks saved translate directly into shopper convenience and employee time: the same number of faulty baskets caught, with fewer shoppers stopped on their way out.

Detection Rate in Spot Checks

This section displays a daily chart plotting Sentry Selection against Random Selection, with a line showing the estimated rate of suspicious baskets in the store overall.

That third line is the important one. A store only ever learns the outcome of the purchases it actually checked, and Sentry AI deliberately checks a non-random selection of them. This means that raw detection figures cannot be read as the store's true rate. The estimate corrects for that bias, giving a baseline that the two selection methods can fairly be measured against.

A falling detection rate is not a sign that the module is degrading. As the Management Console itself notes: once shoppers learn that effective controls are being carried out, fewer baskets are incomplete, and the detection rate falls with them. A declining curve alongside a maintained multiplier is the module working as intended.

Reading an individual decision

Alongside the aggregate page, the spot check calculation details for a single transaction show which modules contributed to the decision, including the Sentry AI score. A purchase that was not selected shows its score too, for example Sentry ML scoring: Chance increased by 27%, so any individual decision can be reconstructed after the fact.

Spot checks triggered by the Sentry AI module appear in the transaction data with the same spot check method as other automatically enforced checks, and the review outcome recorded by the supervisor in the Supervisor app is what the model is trained against.

Availability

Sentry AI has been live in Production systems since May 2026. If you are interested in adding Sentry to your spot check logic, reach out to your shopreme integration consultant for more information. ✨

Related articles

• Spot Check Logic for All Clients