Computer Vision for retail: why your CCTV doesn't catch shrinkage

One of the most common questions we get in Mexican mid-market retail: “We have 200 cameras installed. Why do we still have 3% shrinkage?” The short answer: because recording is not the same as preventing.

Traditional CCTV is post-mortem

90% of LATAM retailers run passive CCTV: cameras record, disks fill up, and footage is only reviewed after something has already happened. By the time you discover the theft in your weekly or monthly inventory, 7-30 days have passed. Finding the exact moment in hours of footage is a needle in a haystack.

The typical outcome:

The incident is identified in 15-25% of cases
The person responsible is rarely identified
The loss is already booked as an operating expense
And since no one prevented it, it happens again the following month

Applied Computer Vision changes the model

The difference is that cameras stop being recorders and become active observers. AI watches the feed in real time and triggers alerts when it detects specific patterns:

Real use cases in LATAM retail

Shelf concealment detection: AI identifies the “grab product + hide under clothing” or “double take” movement in critical zones. It notifies the security team before the suspect crosses the checkout line.
Aisle heat maps: where do your customers stop? Which shelves get traffic and which get ignored? This is merchandising intelligence you can’t extract from the POS. A Mexican electronics chain raised conversion 22% in its top aisle after reordering categories based on the heat map (case under NDA at /en/industrias/retail/).
People counting and dwell time: how many enter, how many buy, how long they stay in each section. Cross-referenced with sales, this gives you the real conversion rate per section — not per store.
PPE in the warehouse/distribution center: it automatically detects workers without a hard hat, reflective vest or safety gear. Regulatory fines and incidents drop to nearly zero.
Automatic license plate recognition (LPR) in the parking lot: integrated with an internal database or a list of suspect vehicles for early alerts.

The typical result in 90 days

In the operations where Migura deploys Computer Vision over a LATAM mid-market retailer (between 20 and 100 stores), the numbers in the first 90 days are consistent:

Metric	Before	After 90 days
Shrinkage event detection	18-25%	+74% over baseline
Incident review time	4-6 hours	under 10 minutes
Identification of the responsible party	under 30%	over 75%
Shrinkage measured in inventory	100% (base)	–25% to –35%
Live event coverage	0%	over 90% of critical zones

The stack we use

We don’t sell boxes. What a LATAM mid-market retailer needs is:

Professional IP cameras with enough resolution for AI analytics: Hikvision (we’re a Premium Partner), Dahua or AXIS depending on budget and use case
A central management platform: HikCentral, proprietary platforms or an open stack depending on the case
AI models trained for the specific use case: this isn’t “install and done” — the models are tuned to your store’s layout and behavior
Integration with your ERP/POS to correlate events with real transactions
NOC monitoring or an executive dashboard depending on the operating model

Full technical details at /en/seguridad/.

Investment and ROI

A pilot in 1-3 stores (coverage of 30-50 AI cameras with a central platform) carries a typical investment between USD $25,000 and $80,000 depending on existing infrastructure. Documented ROI between 3 and 8 months from reduced shrinkage.

Scaling to the full chain (30-100 stores) lowers the per-camera unit cost by 30% to 50% thanks to economies of scale in platform and services.

How to get started

Three common mistakes we see in mid-market retail:

Buying cameras before understanding the use case. A camera for shrinkage detection is different from one for a heat map — resolution, angle, AI model. Diagnose first.
Not integrating with the POS. Without correlation to the transaction, the alert is generic. The real power is knowing “this person walked past the register without paying for product X.”
Deploying without involving the loss-prevention team. AI is a tool. If your team doesn’t know how to respond to the event in real time, the investment is wasted.

If you want a concrete plan for what to do in your operation, book 15 minutes or ask us for a 48-hour assessment. The first step is always understanding your current setup.

Soluciones Migura is a Hikvision Premium Partner. More about the unit at /en/seguridad/.

Frequently asked questions

What's the average shrinkage rate in Mexican retail?

Between 1.5% and 4% of annual sales depending on the vertical. Large supermarkets sit at 1.5-2.5%, fashion and electronics retail at 2.5-4%. Shrinkage breaks down into external theft (~35%), internal theft (~30%), administrative errors (~25%) and product damage (~10%).

Does AI applied to video replace the human guard?

No. It amplifies them. The human guard is reserved for the 5% of real events that require physical action. AI filters out the 95% of false positives that used to eat up hours of footage review. The ROI comes from eliminating wasted time, not from cutting headcount.

Does it work with the IP cameras I already have installed?

It depends. Modern professional IP cameras (Hikvision, Dahua, AXIS from the last 5 years) are usually compatible with AI analytics via a central platform like HikCentral. Analog or very old cameras need replacing. The initial assessment tells you what you can reuse and what you can't.

Does it comply with Mexico's LFPDPPP law?

Yes, when implemented correctly. Cameras in common and sales areas are legitimate with signage notice. Facial recognition requires informed consent under the privacy notice. When the use case allows it (e.g. counting, heat maps), we anonymize to avoid PII.

And in your operation?

Did this article resonate with you?

A free 90-minute assessment with a senior consultant. Executive report in 7 business days. No commitment.

Request assessment

Computer VisionRetailShrinkageHikvisionSecurityAI