Retail loss prevention systems in 2026: the five layers, and what each one catches
A retail loss prevention system is a stack of five layers, not one product: cameras that capture the store, detection that recognizes risky behavior, deterrence that interrupts it, case management and evidence that turn incidents into cases, and reporting that shows where loss concentrates. Most chains already own the cameras and some of the reporting, so the gap sits in the three layers between them. Spot AI works on the cameras a retailer already owns and adds detection, deterrence, and case-ready evidence on top, so one connected workflow detects, deters, investigates, and helps resolve incidents.
The pressure behind that gap is about patterns more than volume. Surveyed retailers saw shoplifting incidents fall 12.4% on average in 2025 against 2024, yet 50% reported higher rates of repeat offenders and 40% reported higher rates of incidents tied to organized retail crime (ORC) (Source: National Retail Federation). A camera that only records cannot connect a repeat offender across stores.
Key takeaways
- A retail loss prevention system has five layers: cameras, detection, deterrence, case management and evidence, and reporting. Most chains own the first and part of the last.
- Detection and deterrence pay back together. Detection alone documents a loss after the fact, and deterrence alone means a staffed post at every door.
- Spot AI is camera-agnostic and connects IP cameras that support ONVIF or RTSP, so the cameras a retailer already paid for become the base of the stack.
- Spot AI's Intelligent Video Recorder (IVR) keeps full-resolution video in the store and sends only metadata across the network, the architecture question IT and security reviews tend to ask first.
- Case management decides whether small incidents ever add up to a case law enforcement will take.
Key terms
- Detection layer: software that flags the event that matters in video or transaction data, such as loitering after close or a refund with no customer present. Spot AI's detections are context-aware, weighing location, time, and movement rather than firing on any pixel change.
- Deterrence layer: the response that interrupts an incident while it is still happening. Spot AI triggers strobes, horns, and a natural-conversation talkdown automatically when a detection crosses a threshold the retailer sets.
- Exception-based reporting (EBR): analysis of point-of-sale (POS) data that flags unusual voids, refunds, or no-sale drawer opens for review.
- Case-ready evidence: timestamped clips and incident details packaged against a case record, so an investigator hands over a finished file instead of assembling one.
The five layers of a retail loss prevention system
The table shows each layer's job, the cost of skipping it, and whether a typical chain already has it.
Layer | What it does | What skipping it costs | Usually owned today? |
|---|---|---|---|
1. Cameras | Capture the entrance, sales floor, registers, stockroom, receiving door, and lot | Every layer above goes blind where a camera is missing, misaimed, or offline | Yes, almost always, in a mix of brands and install years |
2. Detection | Flags behavior and transaction exceptions as they happen | Loss surfaces at the next inventory count or in a chance footage review | Partly: POS exception reports and motion alerts, rarely behavior detection |
3. Deterrence | Interrupts an incident with strobes, horns, talkdown, or a routed alert | The store records losses it could have interrupted, or pays someone to stand at the door | Rarely automated: EAS gates, plus guards at the highest-risk stores |
4. Case management and evidence | Builds cases with clips, a timeline, and a record of who handled the evidence | Small incidents are never linked, so they stay below the thresholds law enforcement acts on | Sometimes: case software or spreadsheets kept apart from the video |
5. Reporting | Shows loss by store, zone, hour, and type, and whether the program pays back | Budget follows the last incident, and the program cannot prove its value | Partly: shrink from inventory counts and POS reports, rarely trends from video |
Before talking to any vendor, score the five layers for one representative store as owned, partial, or missing. The missing layers are the shortlist, and the partial ones are often an integration job rather than a purchase.
Our comparison of retail loss prevention software then scores the products that fill the gaps.
Layer 1: cameras, the layer most chains already paid for
The camera layer covers the entrance, sales floor, registers, stockroom, receiving door, and parking lot. Most chains own it already, in a mix of brands and install years. Our guide to AI security cameras for retail chains covers coverage planning by store format.
Few chains skip this layer. It fails quietly instead. A camera blocked by a new fixture or offline since last month leaves a hole in every layer above it, which makes camera downtime a loss prevention problem and not only an IT ticket. At Don Franklin Auto, a 30-location dealership group in Kentucky, proactive camera monitoring and an easier interface save the IT department 5 hours a week on system management (customer-reported).
Spot AI is camera-agnostic. It connects any IP camera that supports ONVIF or RTSP, including Avigilon, Axis, Hanwha, and Pelco units, and brings legacy analog cameras in through the IVR. Where a lot or perimeter has no usable cameras or power, VigilanteX, Spot AI's deployment partner, deploys pole-mounted units that report into the same Spot AI dashboard.
Layer 2: detection, where most stacks thin out
Detection turns video and transaction data into an alert while an incident is still happening: loitering after close, entry to a stockroom, a refund with no customer present, voids clustered on one shift. Exception-based reporting covers the transaction side, and video analytics covers the rest.
Chains usually own the transaction side and rarely the video side. Across 47 video analytics in six retail environments, a 2024 survey of 70 retailers found an overall utilization rate of 10%, and 4% on the sales floor itself (Source: ECR Retail Loss). Self-checkout draws the most planned activity: 49% of those retailers were planning or trialing non-scan detection at self-checkout lanes (Source: ECR Retail Loss).
Without detection, loss is discovered rather than caught, at the next count or in a weekly report. Motion alerts are no substitute: a rule that fires on any movement cannot tell a delivery driver from someone casing the lot, so teams learn to ignore it.
Spot AI runs 15+ pre-trained Video AI Agents with context-aware detections and ties each POS exception to its register clip. Through Iris, a team can describe a new detection in plain language and have it running in about eight minutes.
Layer 3: deterrence, the difference between a record and an interruption
Deterrence acts while the person is still there: strobes and horns, a natural-conversation talkdown, and an alert routed to whoever can respond. Most chains own a narrow version: EAS gates at the exit, guards at the worst-hit stores, and little automated deterrence at the stockroom door or in the lot.
Without it, a stack is a very good record of losses it watched happen. The traditional alternative is a person. The median annual wage for security guards was $38,020 in May 2025 (Source: U.S. Bureau of Labor Statistics). One post staffed around the clock, 8,760 hours a year, takes about 4.2 full-time schedules, so wages alone come to about $160,000 before benefits, overtime, or an agency's margin.
Spot AI deters in seconds when a detection crosses the retailer's threshold, with a talkdown that can call a person out by what they are wearing and escalate if they stay. Spot AI then routes the incident and auto-resolves routine ones without human review, so people spend their time on the incidents that need them.
Layer 4: case management and evidence, where incidents become cases
The case layer gathers clips, the timeline, POS data, and notes, links related incidents, and records who handled the evidence. It turns a bad night into a file a prosecutor can act on.
Skip it and small incidents stay small. In the same NRF study, 63% of retailers report fewer than half of store theft incidents to law enforcement, and 60% cite low-dollar losses or not meeting felony thresholds as a reason (Source: National Retail Federation). A repeat offender who takes $200 from each of five stores is five unreported incidents in a chain with no case layer, and one reportable pattern in a chain that links them, which is how organized retail crime cases get built. Export quality matters too: NIST notes that the lack of a common export standard for CCTV video often forces conversions that degrade image quality, lose metadata, and add costly delays (Source: NIST).
Larger chains often run standalone case software, and smaller ones use spreadsheets. Either way the case lives apart from the video, so every file starts with manual exports.
Spot AI clips the relevant footage, reconstructs the timeline across cameras, and organizes it into case-ready evidence that can be shared in a click. At Don Franklin Auto, where organized thieves took six vehicles worth $130,000 each from a showroom floor, video reached the responding officers' phones within four minutes of the alarm and five of the six vehicles were recovered within the hour (customer-reported).
"Within four minutes of the alarm going off, Spot AI gave us video footage of the incident on the responding officers' phones. We had five recoveries within the hour."
Josh Bowlin, IT Director, Don Franklin Family of Dealerships
Layer 5: reporting, the layer that pays for the other four
Reporting turns detections, cases, and store data into trends by store, zone, hour, and loss type. Most chains own two sources already, shrink from physical inventory counts and POS reports, but both arrive late and neither shows where in the store or at what hour the loss happened.
Without this layer the program cannot prove itself, a gap that reaches well beyond loss prevention: 75% of retail and consumer-products executives call AI a top strategic priority, but only 16.5% can quantify a return (Source: Deloitte). A program that can show its own results store by store wins the budget argument over one that waits for the next inventory count.
Spot AI's Agents Reporting turns detections into trends across days, weeks, and months, broken down by agent, camera, and location. The same cameras can count people and map traffic by aisle, which gives store operations a reason to share the cost.
Which layer catches which kind of loss
The table maps common retail loss types to the layer that catches them first and the layer that closes them out.
Loss type | Layer that catches it | Layer that closes it | What Spot AI does |
|---|---|---|---|
Repeat and organized theft across stores | Detection of repeat vehicles and patterns | Case management, linking incidents across stores | Spot AI flags license plates of interest and keeps cases from every store in one system |
Walkouts and push-outs | Deterrence at the door, alongside EAS | Case management, with the door clip attached | Spot AI adds the door camera's clip and the timeline to the case |
Register fraud: no-customer refunds, voids, no-sale opens | Detection through exception-based reporting | Case management, with the matching clip | Spot AI ties each POS exception to its register clip, cutting review from hours to minutes |
Sweethearting and skip-scanning | Detection at the register | Reporting by associate and shift | Spot AI surfaces the exception pattern and the video behind it for coaching |
Self-checkout walk-throughs | Detection at the lane | Reporting on the lanes and hours that repeat | Spot AI flags walk-throughs at self-checkout for review |
Stockroom and back-door theft | Detection of unauthorized or after-hours entry | Case management | Spot AI alerts on unauthorized entry at the stockroom or receiving door |
After-hours break-ins | Detection | Deterrence, then a routed alert | Spot AI triggers strobes, horns, and talkdown, then notifies the on-call contact |
Lot loitering, car-casing, and vehicle break-ins | Detection | Deterrence | Spot AI detects loitering and vehicle break-ins and deters in seconds |
Process loss: receiving errors, skipped closing steps | Reporting | Operations follow-up | Spot AI checks opening and closing procedures against the store's standard |
Our guides to register fraud and skip-scanning and to parking lot loitering go deeper on those rows.
When the cameras are already installed and the gap is everything above them
This is the most common starting point: cameras in every store, POS exception reports, cases in spreadsheets, and nothing that acts. The work is integration rather than construction, and it runs in this order:
- Inventory the estate. Record each camera's model, resolution, install year, and whether it speaks ONVIF or RTSP. Our guide to adding AI to a camera network you already own covers the checks in detail.
- Map cameras to loss zones. Use the loss-type table above: entrance, registers, stockroom, receiving door, and lot. A zone with no camera, or a camera facing the wrong way, is a coverage gap that no software fixes.
- Connect what exists. Feed POS data into detection so exceptions arrive with video, and decide whether cases stay in the current system or move to one that holds the video.
- Decide where video lives. With Spot AI's IVR in the store, full-resolution video stays on site and only metadata crosses the network, so the store's connection carries events rather than streams.
- Set rules by zone and hour. The lot after close gets strobes and talkdown, the stockroom door alerts the manager on duty, and the sales floor during store hours gets detection with no audio.
- Pilot, then scale in waves. Spot AI's usual path runs through a demo on the retailer's own video, a technical feasibility check, and a proof of value of commonly about three months. Sites commonly go live in days, so store access and network readiness set the pace of the rollout.
Write the pilot's success criteria down before it starts, using numbers the stack itself can report: incidents interrupted, days from incident to closed case, and alerts dismissed per week. A pilot judged on shrink alone rarely shows a result inside three months, because shrink is only known after an inventory count.
What IT and security teams check before approving the stack
The IT or security lead often sets the timeline on this purchase, and these questions decide it:
- Camera compatibility. Ask for the answer as a list per store: which cameras connect as they are, which analog units need the IVR, and which zones have no usable camera at all.
- Network load. Measure it at the pilot store rather than debating it on paper, by comparing the store's upstream traffic before and after the install.
- PCI scope. Keep the camera network segmented from the POS network, and confirm the video stack never needs access to cardholder data. Spot AI's edge-first design is built to keep deployments PCI-clean.
- Security documentation. Spot AI is NDAA-compliant, follows SOC 2 practices, and is built on zero-trust principles. Ask for the documents before the review meeting rather than during it.
- Data access. Spot AI offers open APIs, webhooks, and a Model Context Protocol (MCP) endpoint that lets an AI assistant query Spot AI data with read-only permissions.
Where a layered system falls short
No stack removes the need for judgment. Plan for these limits:
- Coverage. Every layer is bounded by what the cameras can see, and lighting, angle, and resolution decide what detection can read.
- Tuning time. The first weeks produce more alerts than the steady state, and rules settle as teams dismiss and adjust them.
- Deterrence policy. What a talkdown says, when it escalates, and when associates stay out of it are store-policy decisions, not software settings.
- Law enforcement response. A finished case improves the odds of action, but response still differs from one jurisdiction to the next.
- Seams between vendors. Each separate tool for a layer adds an integration, a login, and a handoff where evidence can go missing.
Process and training carry the rest of the program, and our guide to retail loss prevention strategies covers that side.
If you are the loss prevention lead building the case, or the IT lead asked to vet it, the fastest test is your own footage. Book a demo with a camera list from one store, and Spot AI will show which of the five layers those cameras can carry today.
Frequently asked questions
What are the main types of retail loss prevention systems?
Most products sold as a loss prevention system cover one or two of the five layers. EAS tags and gates deter at the exit, a video management system (VMS) stores video, exception-based reporting software reads POS data, and case management software holds investigations. Video AI platforms such as Spot AI span detection, deterrence, and case-ready evidence on installed cameras, so compare products by the layers they cover.
Can a loss prevention system cover stores with no loss prevention staff on site?
Spot AI is built for that case, because acting on an incident does not depend on someone watching: deterrence fires on its own, routine incidents close without review, and the rest go to whoever runs the store. The design work is routing rather than headcount. Decide which incidents reach the manager on duty, which escalate to a district lead, and which wait for the next business day, then review that split after the first month.
Is EAS still worth keeping once video AI is in place?
For most chains, yes. EAS gates deter at the exit, while video AI covers what a gate cannot see: the aisle, the register, the stockroom, and the lot. An EAS alarm says a tagged item crossed the door, and the clip in the case record shows who carried it and where they went. Keep EAS where the tag program already pays back, and add detection where loss happens away from the door.
Which metrics show a loss prevention system is working?
Read the numbers layer by layer instead of waiting for shrink to confirm the trend. Detection is settling when alerts dismissed per week fall, deterrence is working when repeat visits to the same zone drop, and the case layer is working when days to close fall and law enforcement accepts more of the cases. Spot AI breaks detections down by agent, camera, and location, so each of those checks can run store by store.
How do you build the business case for a retail loss prevention system?
Start from the losses the stack can reach and the labor it changes, not from a vendor's average. Pull 12 months of incident logs and exception reports by store, then price the current response: guard hours, investigator hours per case, and write-offs on incidents that never became cases. Compare pilot stores with similar stores that do not have the system, so the result is a range from your own estate, not a borrowed percentage.
About the author
Joshua Foster is an IT Systems Engineer at Spot AI, where he focuses on designing and securing scalable enterprise networks, managing cloud-integrated infrastructure, and automating system workflows to enhance operational efficiency. He is passionate about cross-functional collaboration and takes pride in delivering robust technical solutions that empower both the Spot AI team and its customers.