Retail store loss prevention with AI video: a practical guide

Retail employee theft and POS fraud: where store routines open gaps

In a shop, every item has a price and a barcode, and every sale is supposed to leave a line in the POS. The openings for employee theft sit where that rule meets the routines of the store: returns and exchanges, discounts, cash payments, change and the drawer itself. Each kind of store has its own version.

Fashion boutiques: returns and exchanges

Returns and exchanges are part of selling clothes: the wrong size, a change of mind, a gift swapped after the holidays. They are entered at the same register as sales, often by the same person, sometimes against a receipt and sometimes without one. A refund entered when no customer is there and nothing comes back looks, in the log, like any other return.

Luxury boutiques: few sales, high value

A luxury boutique may close only a handful of sales a day, each for a high amount, and staff may know their clients personally. A single voided sale, refund or under-rung item is a large amount on its own, and a discount for a regular client is a normal part of service — which is why one improper discount does not stand out.

Souvenir shops: seasonal staff and tourist cash

Souvenir shops may take much of their payment in cash from visitors who are passing through, may not speak the language and are unlikely to come back to question a receipt. In season, temporary staff may join who are new to the register routines, and logins or manager codes may be shared to keep the queue moving. Then the employee ID on an entry says little about who was at the register.

Bookshops: many items, low prices

A bookshop basket can hold several items at low individual prices. One unscanned book among five barely changes the total the customer pays. The missing item surfaces later as a stock-count difference, with no register, hour or cashier attached.

Self-checkout lanes, where present

Where a store has self-checkout lanes, the customer does the scanning. Sweethearting by a cashier does not apply there in the strict sense, but missed scans do: an item goes from the basket into the bag without crossing the scanner, and the transaction simply has fewer lines than items.

Several stores, one team

An operations or loss-prevention lead responsible for several stores often sees them through reports: void, refund, no-sale and discount counts per store and per cashier. A store with more refunds than the others is visible; what happened at the counter during each one is not, and nobody can watch the footage of every register in every store.

None of this is suspicious on its own. The difficulty is that the POS log records only what was keyed in. Whether a customer was at the counter, whether goods or cash crossed it, and which product actually went into the bag is on the camera above the register.

Three ways cash and stock go missing at a store register

Each of these has its own detailed guide. Here is what it looks like in a store.

Void and refund fraud: refunds with no customer, sales voided after payment

In a boutique, the typical case is a cash refund entered between customers — against an old receipt or as a no-receipt return — with no one at the counter and nothing coming back across it. A variation is refunding a real earlier sale to a card the employee controls. The other form is a void after payment: the customer pays, takes the goods and leaves, and the closed sale is voided later, so the drawer still balances at close. With high-value items, one such void is a large amount. The check is the order of events — payment and handover on camera first, a void in the POS after — and, for a refund, whether a customer was at the counter and whether anything came back. An exchange that the POS records as a return is checked the same way. More in void and refund fraud.

Cash skimming and no-sale drawers

A customer pays cash and leaves with the item, but no sale is rung up; the money goes into a pocket, into a drawer left open from the previous sale, or into the drawer behind a "No Sale" opening, to be taken out later. Because the POS never expected that cash, the drawer still balances. In a souvenir shop with many cash sales, or a busy bookshop register, a few extra no-sale openings can pass unseen. Cash also moves with no entry at all: a drawer opened with a key, an unlogged payout, or a change-fund movement at the safe in the back office. More in cash skimming and no-sale drawers.

Sweethearting and missed scans

Sweethearting is a cashier letting a friend, relative or accomplice take goods free or cheap. In a bookshop it can be a book moved around the scanner and straight into the bag; in a fashion boutique, a leather jacket rung up under the code of a basic T-shirt; in a souvenir shop, a manual discount keyed on a sale for a familiar customer. At self-checkout, the same missed scan happens without a cashier. No cash leaves the drawer and the drawer balances; the loss shows up later as stock shrink. More in sweethearting and missed scans.

In all three, the POS log either has an entry with no context — a refund, a void, a no-sale, a discount, a cheaper product — or no entry at all. The context is on the camera at the same moment.

What to look for in a solution for retail stores

Whichever vendor you talk to, these are fair questions to ask about a system meant to catch employee theft and POS fraud in a store:

• Works on your existing cameras. Can it use the cameras already above your registers and any self-checkout lanes, and any that cover the safe, or does it need new hardware?
• Integrates with your POS transaction log. Refunds, voids, no-sales and discounts live in the log. Ask which POS exports it can read, how returns and exchanges are handled, and whether it can flag an item or a cash payment that has no line in the log — not only events that are in it.
• Compares products, not just counts. For under-rings, ask whether it compares the item on camera with the product on the POS line.
• Produces evidence usable in HR processes. A time-stamped clip together with the matching transaction line, so a finding can be reviewed and discussed fairly — not a bare alert.
• Respects privacy. Is there an on-premise option where footage never leaves the site? Is anonymization applied, is analysis limited to the checkout and cash-handling zones rather than the sales floor and fitting rooms, and does it meet GDPR and other data-protection requirements?
• Can be piloted on your own footage. Your registers, camera angles, products and POS — including a busy season, not a demo video.
• Separates single events from repeating patterns. Many refunds, voids and discounts are honest, and barcodes fail. Ask how findings are reviewed by a person and how single events are separated from patterns that repeat.
• Gets alerts to the right role, across stores. An instant alert to the store manager, a digest for loss prevention, a dashboard for area and head-office teams across all stores — each person sees what they need to act on.

How Primarch's Retail Fraud Expert helps retail stores

This section is about our product. Primarch builds on and enhances vision-language models (VLMs), which read video and language together: they understand the action in the scene, not just the objects in it. The Retail Fraud Expert is the Primarch expert module for the checkout. It recognizes actions at the register — scan, skip, void and refund — and cross-checks each one against the POS record in the same second. It works in any store with a register — fashion and luxury boutiques, souvenir shops, bookshops — and on self-checkout lanes. In a store it:

• Checks whether a customer is at the counter when a refund is entered, and flags a return entered with nothing on the counter.
• Recognizes on video that a sale was paid and handed over before it was voided.
• Matches no-sale drawer openings, back-to-back voids and discount-key anomalies with the motion on camera, telling cash going in from cash coming out and whether it was handed across the counter. A drawer opened with a key or left open with no POS entry is flagged.
• Flags cash handled when the POS shows nothing at all: unrung sales, unlogged payouts or pickups.
• Recognizes an item passed around or over the scanner and checks in the same moment whether the POS has a matching line; spots under-rings by comparing the item on camera with the product on the POS line.
• Applies the same camera + POS check to self-checkout lanes.
• Checks safe drops, change-fund movements and pickups away from the register against their records, where a camera covers the area.
• Chains events by store, register, cashier and hour into 30-day patterns: each single event is alerted for review, and the POS-validated 30-day pattern shows what repeats.

Putting it to work

1. 01

   Connect

   Connect your existing RTSP cameras and the POS transaction log export; no new cameras or special hardware. It is designed to work with any common POS system that can export transaction logs; the integration is validated together during the pilot.

2. 02

   Pilot on your own footage

   Run it on your own registers, camera angles and POS data, and review the findings together.

3. 03

   Roll out

   Extend to more registers and stores, with central monitoring across the chain.

• Alerts, digests and dashboards. Instant notifications, periodic digests and role-based dashboards, so each finding reaches the person who acts on it — from the store manager to the area and head-office teams.
• Ask the past in plain language. Findings are written into a queryable event memory: "Which no-receipt refunds were flagged at register 1 last week?"
• Evidence for HR. Every finding carries its time-stamped clip and the matching transaction line.
• Privacy by design. With on-premise deployment, footage never leaves the facility. Analysis focuses on the checkout and cash-handling zones with privacy-compliant anonymization; the goal is action–transaction consistency, not tracking people.

For counters, drive-thrus and delivery cash in food service, see restaurant and café loss prevention.

Industry figures: what fraud data shows for retail

There is no reliable public figure for register fraud in boutiques, souvenir shops or bookshops on their own. The closest published figure is the overall estimate from the ACFE's 2024 study of occupational fraud, which is not limited to retail or to the register.

Refunds, voids, no-sales and discounts are already in the POS log of every store; each can be checked against the footage of the register at that moment. Missed scans and unrung cash sales leave no line at all, so the camera is the only place they can be seen as they happen — and repeats at the same register and hour stand out as a pattern.