Everseen Evercheck

par Everseen

AI Application

Faible confiance

Evercheck is Everseen's computer vision loss prevention system for grocery and general retail checkouts. Overhead cameras at self-checkout and staffed lanes feed edge-deployed vision models that track items and hand movements and reconcile them in real time with POS scan data, flagging more than 30 loss patterns such as missed scans, items left in carts or baskets, product switching and abandoned transactions, plus produce recognition. When a discrepancy is detected the shopper sees a 'Soft Nudge' video replay on the kiosk screen and can correct it; unresolved events alert store staff, and the Everdash dashboard reports loss, interventions and ROI. Vendor-reported scale: 150,000+ live checkouts, 10,000+ stores, 6+ petabytes of video a day, customers including Kroger, Coles, Woolworths, Meijer and Morrisons. The vendor says models continually learn from live store activity. Rated high. The system continuously films identifiable shoppers and staff in a publicly accessible space and its purpose is to detect suspected theft, so its outputs can lead to people being challenged, embarrassed or reported, and alerts amount to data relating to suspected offences (GDPR Art. 10 handling and national rules apply). A DPIA is mandatory for the retailer under GDPR Art. 35(3)(c) (systematic monitoring of a publicly accessible area at large scale). On the EU AI Act: it does not appear to fall under the Art. 5 prohibitions, since the vendor says it analyses checkout behaviour rather than identifying individuals (so no remote biometric identification) and it flags objective scan discrepancies rather than predicting criminality from profiling or personality traits; however this rests on vendor statements, not published technical documentation. When used on staffed lanes to detect 'scan avoidance' or to measure cashier performance (and with the Evereagle labour efficiency module), it can fall under Annex III point 4(b), AI used to monitor and evaluate the performance and behaviour of workers, which is high-risk and also triggers works council and employee information duties. Assurance is weak: no public SOC 2 or ISO 27001, no public DPA, sub-processor list, retention periods, hosting locations, or accuracy and false-positive metrics, and the privacy policy's claim of a 'zero requirement to collect, use or store any PII' is not credible for a video system. Historic concerns include Walmart employee allegations of frequent false positives (2020) and a Walmart litigation settled in 2021. The rating would be reconsidered toward medium only if the vendor shows on-device processing with no identifiable video leaving the store, short documented retention, published accuracy and bias testing, a clear Annex III analysis for staffed-lane use, and a DPA with breach-notice terms. It would move to prohibited territory if any facial recognition, re-identification across visits, or emotion or intent inference were added.