# Enterprise face recognition buyer’s checklist

Twelve items in the order they should be answered. The first four regularly
eliminate entire categories of system, which is why doing them first saves the
most time.

The five categories are: cloud face API, device-only terminal, identity
verification service, enterprise on-premise platform, and edge appliance. They
are different products that share underlying technology, and comparing one
against another on a feature grid produces a grid that says nothing.

Full guidance: https://facerec.ayonix.com/compare/enterprise-face-recognition-software

## The checklist

- [ ] State the operational problem in one sentence, without naming a technology. "Reduce the queue at the north entrance at shift change" is a requirement; "implement face recognition" is a solution looking for one.
- [ ] Establish whether biometric data may leave the premises, in writing, with whoever owns the legal position. A no eliminates two of the five categories immediately.
- [ ] Decide 1:1 verification or 1:N identification, and state the gallery size at three years rather than today.
- [ ] Set the latency budget per decision point. A door, a gate and a control-room alert tolerate different delays.
- [ ] Survey the cameras before shortlisting anyone. Measure pixels across the face where people pass, in the worst lighting of the day.
- [ ] List every system that must receive a result, and for each confirm the documented integration mechanism exists before contract.
- [ ] Ask all six evidence questions of every vendor, in writing, with the same deadline.
- [ ] Specify liveness against named attack instruments or not at all. Which instruments, at what level, tested by whom, on what date.
- [ ] Design the fallback path and cost its staffing. This is usually the largest recurring cost in the business case.
- [ ] Write the pilot acceptance criteria before the pilot: population, window, both error types, threshold, and reporting format.
- [ ] Agree the governance controls in the contract: retention per data category, access logging, human review before consequential action, and demonstrable deletion.
- [ ] Ask what leaving costs — template export or destruction, data migration, and what you are left holding.

## Cost components to model

- [ ] Licence model — per camera, per door, per transaction, per server or per site. These produce very different totals at the same list price.
- [ ] Hardware, including the cameras you will have to move, replace or add once the survey is done. The camera line is the one most often missing from a first estimate.
- [ ] Integration effort for VMS, access control and business systems, plus the enrolment process itself.
- [ ] Operations: who runs it, patches it, monitors it and restores it — and for an air-gapped site, who travels to it.
- [ ] The fallback path staffing. Recurring, often exceeds the licence, and most often left out entirely.
- [ ] Re-enrolment, if an engine version change makes existing templates incompatible.
- [ ] Measurement and tuning, including re-measurement after any change to cameras, lighting or enrolment.
- [ ] Exit costs.

## Warning signs in a vendor response

- [ ] An accuracy percentage offered immediately, with no threshold, dataset or gallery size attached.
- [ ] A "NIST certified" badge. NIST publishes comparative evaluations and does not certify or endorse vendors; no such certification exists.
- [ ] "ISO 30107-3 certified". That standard defines a test methodology, not a pass mark, and issues no certificate.
- [ ] Unwillingness to agree pilot failure criteria in advance.
- [ ] A proposal to demonstrate on the vendor’s own cameras rather than yours.
- [ ] A claim of VMS certification or marketplace status that the VMS vendor cannot confirm.
- [ ] A platform or language compatibility matrix that has not been tested against your specific version.

---

Published by Ayonix, facerec.ayonix.com. Ayonix sells face recognition and
therefore has an interest in how these checklists are used — they are written
so they can be applied to Ayonix as readily as to any other supplier, and if
following one leads you to a different supplier or a different category of
system, that is a legitimate outcome.

No accuracy percentage appears in any Ayonix material, for Ayonix or any other
vendor, because a figure without its threshold, dataset, gallery size and
demographic breakdown cannot be reproduced.

Corrections: infojp@ayonix.com. Last reviewed 11 September 2026.