Auror reports 85,000 retail stores and 3,500 law enforcement agencies on its network, with more than two million repeat-offender links created to date. Facewatch logged 297,433 repeat-offender alerts in the first half of this year across 125-plus retailers — roughly 1,643 a day.
Cross-retailer offender intelligence is no longer a pilot. It is infrastructure.
Now the part that has not kept pace: I could not find a single peer-reviewed study, government evaluation or independent audit anywhere in the world establishing that pooling offender data between retailers reduces loss.
This edition is about what the evidence actually shows, and about a legal asymmetry sitting underneath all of it that almost nobody in this industry has been shown.
The network scaled. The evidence base did not.
Three findings, none of them from a vendor.
The College of Policing classifies Auror at the "Untested" practice stage — no formal evidence rating assigned. The outcome data it does cite comes from Devon and Cornwall Police internal monitoring: crimes reported through the platform carried an 8.9% higher charge rate and an average recording time of 6 hours against 53. Real numbers, and they measure police process efficiency — not loss, not recidivism.
The most-cited benefit study was paid for by the vendor. Estimating the Benefits of Auror to the New Zealand Police was authored by Sense Partners and commissioned by Auror. It calculates $92 million in police workforce efficiency, equivalent to 451 constables. Read the method: case volumes multiplied by average time saved multiplied by police hourly cost. It is a study of police hours, not of retail loss, and it says so.
The only regulator-run evaluation lost its control group. New Zealand's Privacy Commissioner examined Foodstuffs North Island's facial recognition trial across 25 supermarkets: 225,972,004 face scans, 1,742 alerts, 1,208 confirmed matches, a 92.5% match threshold and two-person verification. The Commissioner found the trial compliant with the Privacy Act. But the trial's comparison between FRT and non-FRT stores broke down mid-way for lack of reporting, collapsing into a before-and-after with no control. It validated compliance. It never established crime reduction.
The strongest genuine outcome data in this field is UK policing's Project Pegasus: 257 arrests, 605 offenders identified and a 73% reduction in offending by monitored organised crime groups over two years. Note what that is — a police operation drawing on retailer referrals, not a measurement of any platform.
The asymmetry nobody points out
Here is the fact that should reframe this for anyone who has signed one of these contracts.
The Retail Equation — the cross-retailer database that scores your customers' return behaviour — is listed by the Consumer Financial Protection Bureau as a consumer reporting company. Because it produces a report that affects how a merchant treats you, consumers get statutory rights: a free copy on request, and a free, reasonable investigation of any dispute.
A database that records how often you return a sweater is a regulated credit-reporting instrument.
A database that labels you an offender and circulates that label to competing retailers is not.
Same industry, same data-sharing architecture, same commercial consequence for the individual — and one carries dispute rights while the other carries none. That is not a privacy argument. It is a consumer-protection gap, and the moment a US regulator notices it, the compliance burden lands on retailers, not vendors.
A perspective from the field
I have sat on both sides of this. Shared offender data genuinely works as an investigative tool — it is how you connect a face in one city to a case in another, and I would not give that up.
But we adopted the network at the speed of a product launch and the audit trail at the speed of nothing. Ask yourself three questions about your own list: who can add a name, what evidence standard applies, and how does a person get off it. If you cannot answer all three from a written policy today, you are not running an intelligence programme. You are running an accusation database with a search bar.
Two vendors, opposite bets
Watch where the market has split, because it tells you the operators disagree about the risk.
Auror's own policy states that retailers cannot share facial persons-of-interest lists with other retailers or law enforcement. The biggest sharing network in retail deliberately refuses to share at the biometric layer.
Facewatch is built the other way. Its model centralises subject lists uploaded by subscribers and distributes them to surrounding subscribing businesses. Inclusion rests on people reasonably suspected of unlawful acts, evidenced by witness account or CCTV.
The consequences of that second model are documented and named. A 19-year-old in Manchester was misidentified in a Home Bargains, searched, removed from the store and told she was banned nationwide through the system; the vendor later admitted the error. Ian Clayton, 67, was asked to leave a store in February and only obtained confirmation that the identification was wrong after filing a subject access request.
Note the pattern in the responses: the failure is attributed to staff, not to the model. That places the liability on the retailer while preserving the vendor's accuracy claim.
And note where regulation is going. The UK Home Office consultation that closed in February and the ICO's March strategy update both cover police use only. Connecticut's proposed ban, Syracuse's unanimous ordinance in May and the New York City proposals all regulate the camera. Not one of them regulates the list. A retailer could comply with every 2026 biometric bill on the table and still maintain an unaudited offender database.
Three practical moves for the next 90 days
Write down your inclusion standard. What evidence puts a name on your list, who signs off, and what happens at 12 months. If the answer today is "a store manager decides," you have a defamation and false-imprisonment exposure with no paper trail. One page fixes it.
Build the removal path before someone builds it for you. The only functioning appeal route in the documented misidentification cases was a statutory data request filed by the accused. That is not a process — it is what happens when there is no process. Give people a named contact and a decision deadline.
Decide your sharing posture deliberately, in writing. The two largest vendors have made opposite bets on whether sharing faces is survivable. Whichever way you go, the decision should sit with your legal and risk leadership on the record, not be inherited from a default setting in a platform configuration.
Closing note
The investigative case for shared offender data is real, and this edition is not an argument against it.
The argument is narrower: the industry built the network before it built the audit trail, and the only jurisdiction that looked closely found a system that was compliant and unproven at the same time. When the first serious US challenge lands, the retailer whose list has a written standard, an owner and an appeal path will be in a very different position from the one whose list is a spreadsheet with a search function.
If you have a written inclusion-and-removal standard for your offender data, I want to see how it is worded. Reply with anything you can share, anonymized always.
Forward this to one LP or AP leader who should be reading it.
— Gabriel
The LP Brief is a weekly intelligence read for senior loss prevention and asset protection leaders. Free. No vendor noise.
Not yet subscribed? thelpbrief.com