If you run loss prevention at a UK retailer, you have probably read the headline numbers from the BRC's most recent annual crime survey. £2.2bn lost to theft in 2023/24. Over 20 million theft incidents. £1.8bn spent on crime prevention. The numbers are bad and they are going in the wrong direction.
What the headlines don't break down, and what is more useful to a loss-prevention team trying to plan a year of investment, is the composition of shrinkage itself.
Shrinkage is not the same thing as theft
Across UK retail, stock shrinkage typically runs at 1.4% to 1.7% of turnover. That sits inside the £7.5bn–£9bn band across the sector, depending on whose estimate you use.
But shrinkage is a basket of four things, not one:
- External theft, shoplifting, organised retail crime, employee theft is sometimes counted here too.
- Internal / employee theft, under-rung tills, refund fraud, back-of-house leakage.
- Admin error, miscounts, misprices, mislabels, mishandled returns.
- Supplier fraud, short deliveries, invoice inflation, mis-scanned goods-in.
Different retailers carry very different shrinkage compositions. A premium supermarket carries different shrinkage than a forecourt or a high-street fashion store. A pharmacy carries different shrinkage from a cash-and-carry. The plan that works against one category does not work against the others.
Where AI loss prevention actually plays
Any honest framing of AI loss prevention should be specific about which component of shrinkage it addresses. QuantumEye is aimed at the external theft component, and a portion of the internal theft component where it crosses the till or a restricted-zone boundary that the cameras can see.
What AI loss prevention software does not directly fix:
- Admin error, that's a process and EPOS problem, not a camera problem.
- Supplier fraud, that's a goods-in audit problem.
- Self-checkout walk-away (the specific category increasingly in the press), that's a fixture, signage and lane-design problem first, with AI assist as a secondary control.
Repeat-offender concentration, the real story under the BRC number
What is genuinely new under the headline shrinkage number is the concentration of theft in a small offender population, people who work multiple stores across multiple operators. Estate-wide watchlist capability has gone from a nice-to-have to a basic table-stake against that population.
If your loss prevention only sees what happens in your own store, you are missing the pattern that ties yesterday's incident at one site to tomorrow's incident at another. The cross-site signal is what lets a chain identify the offenders worth focusing on, and the rest of the customer base benefits because the noise floor on alerts drops.
What we'd recommend you plan for
- Quantify your own shrinkage breakdown, by component, by category, by site cluster. The aggregate number is a board number; the operational planning needs the component split.
- Pick the component AI loss prevention can actually move, usually external theft and the till-adjacent slice of internal theft, and size the addressable loss against vendor cost.
- If you operate multiple sites, prioritise cross-store watchlist capability over per-site analytics. The estate is the unit; the store is a node.