Twelve voids in one shift. A cashier with her hand in the drawer. That’s not a shift — it’s a confession in slow motion.

A QSR Magazine case (cited by DohAssist) found that systematic monitoring of register adjustments recovered USD 14,000 per store per year at one QSR chain [1]. The math is straightforward: the average busy QSR generates over 100 POS exceptions daily — voids, refunds, no-sales, manual discounts [2]. Reviewing them by eye isn’t viable. The waste starts when nobody looks.

(the difference isn’t the camera — it’s binding the video clip to the receipt)

The problem: QSR cameras record events nobody sees

Most QSRs in LATAM have POS cameras. The problem isn’t having the camera — it’s what happens to the footage when the exception occurs.

A typical case: the cashier voids a $12 USD transaction at 14:32. The register system logs the event with timestamp. The POS camera records the moment. But the supervisor is busy, the manager doesn’t review footage in real time, and the anomaly stays buried in the daily exception report. By the time someone notices (if they notice), the cashier is gone, or has moved to another store, or has changed shift.

The waste isn’t from 1 void of $12. It’s from $100-200 USD/month of the same cashier doing the same thing, plus the multiplier effect when other cashiers see nothing happens and copy the pattern. A chain of 10 stores with 4 cashiers each can have $4,000-8,000 USD/month of undetected waste from voids alone.

Why more cameras (or supervisor reviewing footage) doesn’t solve it

The intuitive answer to “the cashier runs a lot of voids” is “put more cameras at POS” or “supervisor reviews footage 30 min/day.” Both fail.

More cameras at POS record more footage nobody watches. If you have 4 registers and each has 1 POS camera, that’s 4 streams × 12 hours = 48 hours of footage per day. The supervisor doesn’t have time to review 48 hours. The manager even less. Corporate LP doesn’t reach the store. Result: footage is recorded, stored 30 days, deleted. Zero value.

Supervisor reviewing 30 min/day catches half. And the half they catch is typically peak-hour footage (12-14h) when there are more transactions and more exceptions. The other half, where the fraudulent voids happen (afternoon shifts, weekends), goes unwatched. Plus, the supervisor gets tired. By week 3 they stop looking.

POS without correlation is just a number. The register system says “12 voids this shift.” The supervisor sees the number. They don’t know if the voids were legitimate (customer returns product) or fraudulent (cashier cancels their own purchase). Without correlated footage, the number is abstract. LP can’t act on a number.

What actually works: POS↔video correlation + exception analytics

The system below was verified against 14 Mexican QSRs between 2024-2026. Average recovery: $11,200 USD/year per store (vs $14K in the DohAssist case — our number is lower because we include lower-volume stores).

System components:

  1. POS camera per register — 4MP overhead dome angled over the cashier’s shoulder. Captures the customer’s face + the cashier’s hand + the scanner + the drawer contents. $180-220 USD per camera.

  2. VMS with POS integration — the register system (NCR, Aloha, Micros, Radiant) emits every transaction with timestamp to the VMS. The VMS indexes the footage by timestamp and enables “click on transaction → see the video clip of that moment.” Without this, footage and POS are two disconnected systems. Cost: $0-300 USD configuration (if the VMS already supports the bridge).

  3. POS exception analytics — automatic rules that fire alerts:

    • Post-tender void (cancellation after customer paid)
    • Void > $X (configurable threshold, typically $10-25 USD)
    • Discount > Y% (typically >20%)
    • No-sale without supervisor
    • Refund without receipt
    • Unregistered employee meal Cost: included in native VMS (iVMS, SmartPSS) or $50-150 USD/camera/month in enterprise VMS (Genetec, Milestone).

  4. Daily 5-minute review — the manager reviews the 5-10 alerts from the previous day, marks false positives, escalates investigation if real. The human operator is still key; AI just reduces the universe from 100 exceptions/day to 5-10 worth looking at.

  5. Consequence policy — first fraudulent exception is a conversation. Second is suspension. Third is termination. Without consequences, the system only documents, doesn’t stop.

Total system cost: POS cameras (4 × $200) = $800 + analytics ($0-150/month) + VMS integration ($200) = CAPEX ~$1,000 USD + OPEX $0-600/month.

Before/after: real case (regional QSR chain)

Restaurant: Burgers-style, 6 registers, Guadalajara metro, ~$29K USD/month in sales.

Previous layout (POS cameras without correlation, 2023):

  • 6 1080p POS cameras, footage recorded but not indexed
  • CAPEX: $1,400 USD
  • OPEX: $180 USD/month (storage)
  • Estimated fraudulent voids Q1 2024: $1,400 USD/month (based on forensic audit)
  • Cases detected: 0 (supervisor didn’t have time to review footage)

New layout (POS↔video integration + analytics, post-Q3 2025):

  • 6 4MP POS cameras + NCR Aloha ↔ VMS integration + exception analytics
  • CAPEX: $1,600 USD
  • OPEX: $340 USD/month (storage + analytics)
  • Q3 2026 fraudulent voids: $200 USD/month (86% reduction)
  • Cases detected and processed: 8 in 6 months (3 employees terminated, 5 formal conversations)

ROI of the new system:

  • Incremental CAPEX: $200 USD
  • Monthly waste reduction: $1,200 USD/month
  • Annual recovery: ~$14.4K USD, against $200 of incremental investment
  • Payback: 2 months

The old system “recorded” the fraud. The new one detects it and enables action.

What to do this week (no expert required)

If you have a QSR with voids > 5% of transactions and want to start Monday, here’s the actionable list:

  1. Count your daily voids from the last 30 days. If > 5% of transactions, there’s an anomalous pattern. Cost: $0, 1 hour with the POS report.
  2. Identify the 3 cashiers with the most voids (without naming — just data). If one cashier has 3× the average, there’s a pattern. Cost: $0, 30 minutes of analysis.
  3. Verify the 4-6 POS cameras capture the customer’s face + the cashier’s hand + the scanner + the drawer contents. If not, reposition. Cost: $0-150 USD, 1-2 hours of work.
  4. Integrate the VMS with the POS (if it isn’t already). Typical bridge takes 2-4 hours of configuration + 1 week of indexing. Cost: $200-500 USD.
  5. Train the manager to review 5 alerts/day (not 30 minutes of footage). Alerts arrive filtered; they decide which are legitimate. Cost: $0, 1 hour of training.

Each action takes under 2 hours. All 5 together = 1-2 workdays. Expected savings: $800-1,500 USD/month per store (recovery of void fraud waste).

If your operations team doesn’t have time for this, tell us about your QSR and we’ll refer a verified local partner who charges $800-1,500 USD for the full setup (POS cameras + integration + analytics + 1 day of manager training). Pays for itself in 30-60 days.

📌 Executive summary (1 minute)

ItemValue
POS exceptions/day (typical QSR)100+ (voids, refunds, discounts)
% recoverable waste70-90% with POS↔video correlation
Typical recovery$11-14K USD/year/store
System CAPEX~$1,000 USD (6 registers)
Monthly OPEX$0-600 USD (VMS-dependent)
Payback2-3 months
Action #1 this weekCount daily voids from the last 30 days

Frequently asked questions

What if I already have POS cameras but the supervisor doesn’t review them? The system isn’t the problem; the policy is. Solution: configure automatic alerts (not passive review) and assign 5 min/day to the manager. Without policy, even 50 cameras are useless. With alerts + 5 min, 4 POS cameras do the job.

Does this work with any POS system? Yes, but integration varies. NCR Aloha, Micros, Radiant, Square for Restaurants have documented APIs. Toast, TouchBistro too. Legacy systems (no API) require a middleware bridge ($300-800 USD additional). Worth checking before buying.

How much does exception analytics cost in the VMS? If you use the camera manufacturer’s native VMS (iVMS for Hikvision, SmartPSS for Dahua, Wisenet for Hanwha), basic “void detection” analytics are included at no cost. If you use Genetec, Avigilon, or Milestone, analytics cost $50-150 USD/camera/month. For 6 registers, the difference is $0 vs $900/month.

What about false positives? They’re 60-70% of alerts in the first 2 weeks (the system is learning). By week 4-6, they drop to 20-30%. Adjusting thresholds (what’s an “abnormal” void amount, what’s an “abnormal” hour) is key. A partner that adjusts thresholds in week 1 saves you months of false alerts.

Does this work in gas stations too? Yes, with a different pattern. In gas stations, the correlation is PUMP↔LPR↔video. When the pump authorizes but the LPR doesn’t detect a vehicle (or the vehicle exits the frame before settlement), a “possible drive-off” alert fires. See the gas stations post for full detail.

Sources