What POS Data Can Tell You—and What Requires Video
A cancelled sale, a refund or an open cash drawer can be entirely legitimate. It can also be part of a theft.
The POS record tells a loss prevention team that an action occurred. Video helps explain what physically happened. Connecting the two makes investigation faster. Adding AI video analysis changes which events need human attention in the first place.
For retailers, this creates an opportunity to examine more activity across more stores with the team they already have.
POS data provides a useful starting point for identifying suspicious activity. Unusual refund volumes, rising cancellation rates and frequent cash-drawer openings can help teams decide where to investigate.
But a transaction record rarely tells the whole story.
Consider a cancelled sale. A customer may have changed their mind. Alternatively, an employee may have accepted cash, cancelled the transaction after the customer left and kept the money. Both situations can produce the same POS event.
Other theft methods may involve merchandise handed over without being scanned, fictitious returns or employees colluding with customers. These can happen even when no cash changes hands.
Video provides the physical context needed to assess the event. This also makes timely review valuable: once footage is overwritten, reconstructing what happened becomes much harder. The loss a retailer can prove may cover only part of the activity that actually occurred.
Why Loss Prevention Teams Have Relied on Sampling
Retail chains generate more POS events than a small loss prevention team can manually examine.
Teams therefore prioritize. Stores with unusual refund activity, significant cash exposure or increasing exception rates receive more attention. Investigators then sample events and review the associated footage.
This is a practical response to limited capacity. Every event takes time to locate and assess, including the many events that turn out to be legitimate.
The limitation is coverage. A store whose overall figures look normal may receive little attention, even when individual transactions deserve investigation.
Reducing the time spent on each event helps. Reducing the number of routine events requiring human review can change the scale of the work.
Connecting POS Events to the Relevant Video
A system that links POS events to CCTV footage allows an investigator to move directly from a transaction or action to the associated video.
In our work at iCape, we measured a reduction in the time needed to reach the relevant footage from ten minutes to five seconds.
That measures retrieval time, rather than the duration of a complete investigation. Investigators still need to assess what happened and develop evidence where necessary. But removing a repeated search step frees time for that work.
POS video integration makes footage easier to reach. AI video analysis can then help determine which footage deserves attention.
How AI Helps Select Events for Review
Return to the cancelled sale.
Instead of randomly reviewing cancellation videos, a team can focus on cancellations where video analysis indicates that cash was received before the transaction was cancelled.
The POS event identifies the action. Video analysis adds the context that makes it worth examining.
At one retail chain, a month of activity included approximately 486k cash-drawer openings. The system retained 109 events for human review. The loss prevention team agreed that every one warranted examination.
Those were 109 events requiring review, not 109 confirmed thefts. Investigation outcomes were not shared with us. The operational result was a much smaller queue that the team considered relevant.
This changes how investigators spend their time: fewer routine events to screen, more attention available for events that need judgment.
From a Few Unusual Stores to Broader Network Coverage
Store-level indicators remain useful. AI video analysis adds another way to direct investigative effort: a specific event can bring a store into the review queue even when its overall figures look ordinary.
The same enabled scenario can be examined across the deployed store network.
At a fashion retailer we work with, the team previously reviewed five anomalous stores each month using sampled checks. Today, the same team conducts regular, focused checks at 45 stores. The retailer plans to complete deployment across all 130 stores by the end of 2026.
The achieved benefit is already clear: more stores receive focused attention without expanding the team.
Human Review Still Matters
An event selected by AI needs assessment. The reviewer must establish what the footage shows, whether the activity has a legitimate explanation and whether further investigation is warranted.
Teams should also check selected events that the system did not flag.
At the chain with approximately 486k drawer openings, we reviewed a random sample of 300 non-flagged events across stores and shifts. All were found to be legitimate. That sample does not prove that every excluded event was legitimate, but it demonstrates a useful practice for checking the screening process.
Broader coverage should go hand in hand with ongoing checks of what the system may have missed.
How iCape Supports This Workflow
iCape connects POS events with footage from a retailer’s existing CCTV and uses AI video analysis to help focus human review on relevant events.
The value comes from combining faster access to evidence with a smaller screening burden. Investigators can examine more stores, follow up sooner and spend more time understanding suspicious activity.
For a retailer concerned about theft at the till, the practical question is how much of the network its team can meaningfully examine—and how quickly it can act while the evidence is still available.
Talk to iCape about connecting your POS data and existing CCTV to expand your loss prevention team’s coverage. Get in touch