Solution · Custom computer vision
Catch the shrink your cameras already see but never flag.
When packaged video analytics cannot see your store's specific loss, Tec-Tel builds the computer vision that can. POS-exception pairing, sweethearting, self-checkout loss, and concealment detection, running on the cameras you already own.
- NDAA-compliant
- Platform-agnostic
- 1,000+ deployments over 15 years
Retail shrink detection is custom computer vision Tec-Tel builds when off-the-shelf VMS analytics cannot see your store's specific loss. The model runs on the cameras you already own, reading existing feeds over ONVIF or RTSP. It pairs POS exceptions to the right clip and flags sweethearting, self-checkout loss, and concealment as they happen, then routes structured events into your VMS, exception reports, and alert channels. Built and tuned per store, then standardized across the chain. Book a free consultation.
§01 What the model detects
From recorded footage to caught incidents.
A record-and-review camera only helps after someone knows where to look. Custom computer vision surfaces the everyday leak as it happens: the misring, the sweetheart deal, the self-checkout walk-off, the concealment in the high-theft aisle. Not every store needs every detection, so the consultation picks the two or three that cost you most.
§02 The problem
The shrink that an off-the-shelf VMS never sees.
Most retail cameras are a record-and-review system. The footage exists, but it only earns its keep after an incident, when an investigator already knows where to look. The everyday leak, the misring that happens forty times a shift, the self-checkout walk-off, the friend at lane four, runs underneath the recording and never surfaces on its own.
Packaged video analytics help with the generic cases: a line crossed, a person loitering, a plate on a list. They are built to sell to every store, so they stop where your floor gets specific. They do not know your register layout, your high-theft SKUs, or the exact move your loss keeps taking.
That gap is where Tec-Tel builds. When the off-the-shelf product cannot see the operational problem, we build the computer vision that can, and run it on the cameras already in your ceiling.
§03 How it works
Custom computer vision on the cameras you already own.
A camera produces video. Computer vision turns that video into structured events: an item crossed the belt with no scan, a cart left through an unstaffed lane, a void fired with nobody at the register. The events show up where your team already works, paired to the exact clip.
The model reads existing feeds over ONVIF or RTSP and runs inference on a separate server or in the cloud. It does not touch the cameras. Most fleets built in the last several years clear the baseline of roughly 1080p at 15 fps, so the build runs on what you have. Where a lane is too dark or a camera too old to carry detection, we say so and phase that fix in, rather than gate the whole project behind a hardware refresh.
This is the difference between buying a product and getting a solution. The detection is trained and tuned to your stores, not a license you switch on and hope fits.
§04 The ROI
Why a fraction of a percent is worth building for.
The NRF National Retail Security Survey 2023 put U.S. retail shrink at $112.1 billion in 2022, about 1.6% of sales, with 63% of it tied to internal and process causes rather than outside theft. For a single $10M store, 1.6% is $160,000 a year walking out the door, and most of it is the quiet, repeatable loss a record-and-review camera never flags.
Recovering even a slice of that changes the math on the system. A custom detection model that surfaces misrings, sweethearting, and self-checkout loss as they happen turns the cameras from an after-the-fact archive into a daily loss-prevention tool. What matters is incidents your team actually catches, not detections on a slide.
§05 How Tec-Tel builds it
Scope, build, tune. Then expand.
We do not quote shrink CV off a floorplan. The build follows a sequence, and a pilot store proves it before the chain commits.
- → Scope on site. We walk the store with your LP and store-ops leads, read your shrink data, and find the two or three patterns that cost you most. That is what the model gets built for.
- → Confirm the cameras carry it. We check resolution, framerate, and angle on the lanes and zones that matter, and flag the few that need an upgrade before detection will hold.
- → Build and integrate. The detection model is trained for your floor and wired into your VMS, your POS exception stream, and your alert channel, so events land paired to the clip.
- → Tune before alerts go live. We run the model quietly against real footage, calibrate thresholds, and cut false positives down before anything routes to your team.
- → Pilot, then expand. One store proves the lift. Then we standardize the build across the chain on one playbook, not five.
Questions buyers ask us
FAQ
- How is this different from the analytics that came with our VMS?
- Packaged VMS analytics cover generic cases sold to every store: line crossing, loitering, basic people counting. They do not know your register layout, your high-theft SKUs, or the specific move your shrink keeps taking. Tec-Tel builds custom computer vision for the operational problem your off-the-shelf product cannot see, trained and tuned to your floor, and runs it on the cameras you already own.
- Do we have to replace our cameras to get this?
- Usually no. The detection model reads your existing feeds over ONVIF or RTSP and runs inference on a separate server or in the cloud, without touching the cameras. Most fleets from the last several years clear the baseline of roughly 1080p at 15 fps. Where a specific lane or zone has a camera too old or poorly placed to carry detection, we flag it and phase that fix in, rather than gate the whole project behind a refresh.
- What kinds of shrink can a custom model actually detect?
- The common builds are POS-exception clip pairing for voids, refunds, and discounts, scan-avoidance and sweethearting at staffed lanes, self-checkout loss like ticket switching and walk-offs, concealment and cart-out behavior, and dwell or removal patterns in high-theft aisles. We pick the two or three that cost you most and build for those first, then expand.
- How big is the shrink problem, really?
- The NRF National Retail Security Survey 2023 reported $112.1 billion in U.S. retail shrink for 2022, roughly 1.6% of sales, with 63% tied to internal and process causes rather than outside theft. For a $10M store, 1.6% is about $160,000 a year, and most of it is the quiet, repeatable loss a record-and-review camera never surfaces on its own.
- How long before the system is catching real incidents?
- After integration we run the model quietly against your actual footage, calibrating thresholds and cutting false positives until the queue is trustworthy enough to route to your team. A pilot store proves the lift before the chain commits. The goal is a filtered, trustworthy queue your LP team actions, not a firehose of detections they learn to ignore.
- Will this hold up with our loss-prevention and privacy obligations?
- We design around how stores are actually run. Detection focuses on transactions and movement, not on identifying individuals, and we default away from biometrics unless you have a documented, consent-based program. Where state law sets notice or consent requirements for employee monitoring, we scope to them. Camera retention in the cardholder data environment follows PCI-DSS Requirement 9.
- Can this run across a whole chain, not just one store?
- Yes, and that is the point of the sequence. We prove the build in a pilot store, then standardize it across the chain on one playbook: one detection model family, one integration pattern, one runbook. Five stores from five remodels get one consistent system instead of five one-off setups.
Book a walkthrough
Want a read on where your shrink is hiding?
The free consultation walks your stores, reads your shrink data, and scopes which detections a custom model would actually move, plus which cameras can carry it today.
- Tell us how many sites you run and what's already in place. We'll show you what a build or upgrade looks like.
- Straight answers from the team that does the work. We're platform-agnostic, so you get the system that fits your sites, not one brand's catalog.
Since 2010 · 1,000+ deployments nationwide · ISN-accredited
Or send the details
How can we help?
What you're looking for, plus any details. We review it and follow up, usually the same day.
Related from Tec-Tel
AI & analytics hub
Every detection, footage-search, and identity capability in one place.
Read on IndustryRetail security
The full retail picture: ORC, POS-to-video, PCI scope, and multi-store rollout.
Read on SolutionVision Intelligence
AI analytics layered onto the cameras you already own, across every vertical.
Read on PartnersTechnology partners
The camera, VMS, and analytics platforms Tec-Tel builds and integrates on.
Read on