Accountable and transparent
Aspect 2 of 10. When an AI system flags a shopper, matches a face, or closes a case, two questions decide whether your organization can stand behind it: who owns this, and can we show our work?
What you will be able to do
- Name a single, real owner for every AI tool in your AP program.
- Explain the difference between accountability and transparency.
- Build a decision record that survives staff turnover and litigation holds.
- Tell customers and associates what a system does in language they can act on.
Accountability means a named person
Accountability is a named human who answers for the outcome of an AI system: the decision to adopt facial recognition or video analytics, the decision to keep it running, and the decision to shut it off.
Shared ownership sounds collaborative and usually means nobody. If the answer to "who owns this system" is the AP department, a loss prevention steering group, or the video analytics vendor, you do not have an owner yet.
A test you can run today
Pick any AI tool your team uses, such as your self-checkout loss detection platform. Ask three people who owns it. If you get three different answers, that is your finding.
Transparency means the work is visible
Transparency is whether someone outside the room can reconstruct what happened and why. It has two audiences.
- Internal. AP leadership, auditors, and the next investigator in your seat need the paper trail: why this tool, what it was tested on, what limits were set for person-of-interest matching or ORC group detection.
- External. Customers and associates need to know an AI system was involved in flagging their transaction or their image, and what to do if they disagree with the result.
Openness about limits, such as known false-match rates in facial recognition, builds more trust than confident claims do.
The decision record
One page per AI system. Written when you adopt it, updated when anything material changes.
- What the system does, in one sentence, for example: matches images against a person-of-interest watchlist.
- Who owns it, and who to contact when it misbehaves.
- What decisions it influences, for example a detention, a trespass notice, or a civil recovery demand, and how much weight it carries.
- What evidence you reviewed before saying yes.
- What limits you set, and what would make you turn it off.
- Date, author, and next review date.
This page is what you hand a regulator, a plaintiff's attorney, corporate counsel, or a new investigator. Write it once and it does all four jobs.
Notice and recourse
Most laws and most trust problems come down to the same two duties.
- Notice. The customer or associate is told that AI, such as video analytics or facial recognition, was used, before or at the time of the decision, in words they understand.
- Recourse. There is a real path to question the result, for example a disputed shoplifting flag or a facial recognition match, and reach a human with authority to change it.
A posted sign with no recourse is a disclaimer. Recourse with no notice is a maze. You need both.
A scenario
A grocery chain uses an AI facial recognition tool to alert store teams when a person on the internal person-of-interest list enters the building.
A regular customer is stopped by a loss prevention associate and accused of a theft from another store. She is not the person on the list. Store staff explain that "the system flagged her," nobody in AP can say what threshold triggered the alert, and the vendor's account manager has since left the company.
What went wrong
The tool may have been working close to its designed accuracy. The failure was governance: no owner, no decision record, and no way for a wrongly flagged customer to be heard quickly. That is an accountability problem, and it lands on the retailer.
Questions to bring to the table
- Who is the named owner of this AP system, and do they know it?
- Where is the decision record, and when was it last updated?
- Are customers and associates told that video analytics or facial recognition is in use?
- If someone disputes a match, a detention, or a civil recovery letter, what exactly happens next, and who has authority to override?
- If the AP analytics owner left tomorrow, could the next person pick this up?
Recap
- Accountability belongs to one named person.
- Transparency serves two audiences: the AP team who must answer for the system, and the customers and associates affected by it.
- A one-page decision record is the smallest artifact that satisfies both.
- People need both notice and a recourse path they know about.

