POS Monitoring & Loss Prevention
A cashier monitoring system running in production. It combines live CCTV with POS transaction data, with AI and computer-vision extensions planned.
Capabilities live in production today
Live multi-camera POS grid
Simultaneous view of all active checkout lanes with real-time transaction overlay.
Transaction–video sync
Click any scanned line item to jump the recording to that exact moment.
Rule-based alarm engine
Configurable alerts for cancellations, voids, employee discounts, loyalty card use and flagged products.
Advanced receipt search
Filter by date, POS, value, discount, card digits, employee ID and alarm status.
Alarm review queue
Automatically lists receipts with a fraud risk for a supervisor to review.
Employee statistics
Sales rankings and individual performance reports, exportable to Excel.
Multimedia evidence export
Clip and export H264/MP4 segments linked to specific receipts for evidence packs.
Bilingual UI · light & dark
Polish and English interface, light and dark themes. Runs on standard Windows workstations.
Works with your existing infrastructure.
Real-time integration with your existing CCTV recording system and the store’s transaction stream. No changes to camera hardware are needed.
- HikVision DVR / NVR
- UTC CCTV
- iPOS / POSLog (transaction log)
From reactive review to proactive detection.
The system already provides the data that AI models need: synchronised video and transaction streams. The planned next phase adds AI features on top of this data.
- CV
Skip-scan detection
Detects items that pass the scanner with no matching POSLog event. The model would be trained on existing CCTV footage.
- CV
Product misidentification
Detects the “banana trick”, where an expensive product is entered at the till as a cheaper one: compares the product on the scanner with the code entered at the till.
- CV
Cashier behaviour analysis
Flags items held below the scanner, items bagged before scanning, and unusually fast scans of high-value items.
- ML
Transaction anomaly scoring
Replaces yes/no rules with a probability score for each transaction, from a model trained on past alarms.
- ML
Predictive risk profiling
Models risk by time of day, shift and checkout lane, based on past patterns.
- NLP
Automated evidence packaging
Clips the video, attaches the receipt, marks the anomaly and produces a formatted report.
- NLP
Narrative report generation
Automatically turns raw statistics into written shift summaries and weekly loss-prevention reports.
Two ways to run it.
On-premise
Runs on a server next to your existing DVR infrastructure. No changes to camera hardware are needed. Suitable for environments with strict data requirements.
Cloud
Video and transaction data from each store are streamed to the cloud and shown in a web dashboard. Hosted in the EU by default.