A store running 3% net margin loses 32% of its annual profit if shrink stays at 1.6% of sales. In Mexico the ANTAD 2021 average was 1.08%; in the US NRF closed FY2022 at 1.6% = $112.1 billion[1][2].
The NRF FY2022 breakdown: 36% external theft, 28-29% employee theft, 27% admin and process errors, the rest unclassified[3].
Here is the part almost no one says out loud: CCTV does not solve all of shrink. It solves the part that’s visible.
Beck & Willis (1999) measured 15 UK fashion stores after CCTV installation. At 3 months, average shrink reduction of 20%. At 6 months, high-end systems had regressed to baseline — low-end systems held at -27%[4]. Piza et al.’s 2019 meta-analysis confirms it: CCTV alone reduces crime ~13%; combined with other interventions (public-view monitors, EAS, lock boxes) it reaches ~34%[5].
( The takeaway: cameras deter in the short term, but an employee who knows the blind spots is not deterred by anything if nobody watches the footage afterward )
In every layout audit I review for tv4agent, the same three questions separate a useful system from a decorative one: how many cameras cover the spot where the SKU moves from shelf to cart? how long does it take you to find the scene when the system says 50 and the physical count says 47? and what do you do with the video afterward?
P.D. If your month-end inventory shows a 3% gap vs. the system, do you know which camera, what time, and which operator was on? If the answer is “no”, today’s a good day to pull the footage from Tuesday at 6:47 PM.
📰 Source: NRF National Retail Security Survey 2023; ANTAD Censo Nacional de Mermas 2021; Beck & Willis 1999; Piza et al. 2019.
Frequently asked questions
How reliable is automatic shrinkage detection in retail? Depends on what you mean by ‘shrinkage’. (1) Product counting on shelves: 90-95% accuracy with overhead cameras + YOLO, but only detects aggregate shrinkage, not individual events. (2) Sweethearting detection (cashier not scanning): 85-92% accuracy with POS + video integration, but requires clean POS data. (3) False returns detection: 70-80% accuracy, still experimental. Conclusion: aggregate shrinkage is automatable, individual shrinkage still requires human investigation.
Do I need to integrate with my POS? For sweethearting, yes. Without POS data, you only see ‘cashier didn’t scan this product’ — the footage doesn’t tell you if it was error or fraud. With POS data, the model correlates: ‘transaction 4521 says $45.30, but video shows 7 products. Possible sweethearting’. POS integration is 4-6 weeks of work with your POS provider + the analytics vendor. Cost: $5K-15K USD per store. Without that integration, the ROI of shrinkage analytics is 30-50% lower.
How much does the full deployment cost? Medium store (800-1500 m², 12-18 cameras): $25K-45K USD CAPEX + $4K-7K USD/year OPEX. Components: 1 overhead camera per critical shelf (POS, high-value, jewelry), 1 analytics license per camera ($800-1,500/year), edge server for local inference. Typical payback: 8-14 months via shrink reduction. ANTAD reports average shrink 1.08% of sales; post-deployment, consistent retailers report 0.65-0.85% — a 0.30-0.40 point difference, which in a $5M MXN/month store is $180K-240K MXN/year recovered.
Does it work with my existing cameras? If they are IP 2020+ with minimum 1080p, yes. If older or analog, you need to upgrade — analytics with low-resolution footage has 50-60% accuracy (worse than guessing). If your cameras are new but low resolution (720p), you have to choose between (a) camera upgrade ($200-500 USD each × 18 = $3.6K-9K USD), or (b) analytics with 70-75% accuracy, which detects ‘shrinkage patterns’ but not ‘individual events’.
What about employee privacy? LFPDPPP requires privacy notice and consent for biometric analytics. Shrinkage detection technically doesn’t use biometrics (it doesn’t recognize faces), but it does use facial image in the footage. Recommendation: (1) cameras only in sales and warehouse zones, NOT in bathrooms or break rooms, (2) visible notice ‘Video surveillance zone with behavior analysis for shrink prevention’, (3) footage retained 30 days, (4) footage access limited to management and privacy officer. Without these 4, risk of labor lawsuit.
What if I’m afraid of employee surveillance? The right question. Shrinkage analytics reduces human surveillance (the manager no longer needs to watch hours of footage, only receives alerts). But it also creates the possibility of mass surveillance. The ethical decision is yours, not the vendor’s. Recommendation: (1) communicate to employees BEFORE installing — ‘this detects shrinkage, it doesn’t surveil you’, (2) annual external audit of footage use, (3) employees can request to see what footage you have of them, (4) disciplinary use of footage only with HR review. Without these 4, employees find the footage, leak it to social media, and your brand pays the cost.
📌 Executive summary (1 minute)
| Item | Value |
|---|---|
| LATAM retail shrink | 1.0-1.6% of sales (NRF 1.6%, ANTAD 1.08%) |
| Retail net margin | 2-5% (shrink eats 32-80% of profit) |
| % cameras detect (without analytics) | 5-15% |
| % cameras + analytics detect | 40-60% |
| CAPEX 1080p cameras × 18 | $3,600-4,500 USD |
| Analytics OPEX/month | $0-300 USD (native VMS) or $500-2,700 USD (enterprise) |
| Typical ROI | 6-12 months |
| Action #1 this week | Measure shrink by mode (POS, theft, receiving, internal fraud) |
Sources
[1] NRF National Retail Security Survey 2023 — https://nrf.com/research/national-retail-security-survey-2023 [2] ANTAD Censo Nacional de Mermas Mercado Detallista 2021 (Deloitte) — https://antad.net/retail-talks-censo-nacional-de-mermas-mercado-detallista-antad-retailers-mx/ [3] Retail Dive — NRF won’t publish annual shrink report 2024 — https://www.retaildive.com/news/no-nrf-annual-retail-shrink-report-2024/729009 [4] Beck & Willis — Context-specific measures of CCTV effectiveness in the retail sector (1999) — https://popcenter.asu.edu/sites/g/files/litvpz3631/files/10-BeckWillis.pdf [5] IAPSC EBSP-25-01 Video Surveillance (Piza 2025) — https://www.iapsc.org/wp-content/uploads/2025/07/EBSP-25-01-Video-Surveillance-1.pdf