What happened
Foodstuffs North Island trialed live facial recognition in 25 New World and PAK'nSAVE supermarkets for six months from 8 February 2024. Cameras compared shoppers' faces with each store's watchlist of people who had committed serious harmful behavior there, and two trained staff were to verify each alert. New Zealand's Privacy Commissioner ran a parallel inquiry and released its report on 4 June 2025. His office visited 10 stores, interviewed staff and reviewed the trial data. The Commissioner concluded that the system was an effective way to reduce serious repeat offending during the trial. Scarlatti, an evaluator that Foodstuffs commissioned, compared the trial stores with 128 stores without the technology. It estimated a 16% reduction in serious harmful behavior and a 21% reduction in shoplifting. The Commissioner noted that the stores were not chosen at random and that reporting differed between the groups. The report counts nine misidentifications among 1,735 alerts, two of which harmed the people involved. The report states that the algorithm was trained on groups in Australia that include Māori and Pacific people.
Sources
-
Regulatory document, by authors with no tie to the system.
Office of the Privacy Commissioner, New Zealand. Inquiry into Foodstuffs North Island trial use of facial recognition technology. Report dated May 2025, released 4 June 2025. Executive summary, para 8(b), p. 6. The Commissioner combined the evaluation that Foodstuffs commissioned with his office's own store visits, staff interviews and data requests.
The independent evaluation, combined with our own interviews with staff in trial stores, provides sufficiently strong evidence to conclude that, on balance, using a live facial recognition system was an effective way to reduce serious repeat offending during the trial period.
Overall, the evaluation found an estimated reduction in the incidence of serious harmful behaviours of 16% and an estimated reduction in shoplifting of 21%.
Details
AI system
- Workflow
-
Cameras at store entrances and inside the stores matched faces with that store's own watchlist. Staff acted on an alert only after two trained staff verified it and more than one camera had matched. Only people who had committed serious harmful behavior in the store were enrolled, under set criteria and retention limits.
Follow-up
- Conditions for replication
-
The match score needed for action must be high enough to avoid misidentification, at least 92.5% after the trial's changes. Watchlist images must be of good quality and reviewed regularly.
Entry details
- Sector
-
Retail security.
- Purpose
-
Imagus built the FRT System to match faces seen by cameras with a database of enrolled images in real time. Foodstuffs North Island used it to spot people who had committed serious harmful behavior in its stores.
Not yet checked: model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). Live Imagus face matching cut serious repeat offending in 25 stores (AIBD-0860). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0860.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)