Specialist service

AI oversight testing

When people check an AI system's outputs before they take effect, and you need to show the check works.

Buy the readiness reviewRequest the audit

The readiness review costs €1,650 + VAT, paid up front, with the scoping form the same day. Need an invoice? The audit, and the audit and retest, start with a scoping call and are invoiced.

At a glance

Format and price
three fixed packages, from €1,650 + VAT.
You receive
evidence about your review interface: whether it can be tested, and how many planted errors your reviewers catch against a threshold you set in advance.
For
any organisation where a person reviews an AI system's outputs before they take effect.

Prices exclude VAT. EU businesses with a valid VAT number are invoiced under reverse charge.

The problem

Our documentation says a person reviews every output. But we cannot show that they would catch a wrong one. Approving takes one click, and rejecting opens a form.

What you get

Three fixed packages. Prices exclude VAT; EU businesses with a valid VAT number are invoiced under reverse charge.

Oversight readiness review

€1,650+ VAT · 5 working days
  • One oversight interface, reviewed against the 20-point checklist and the five capabilities of Room to Disagree.
  • Confirmation that a wrong output can still be stopped after the reviewer's decision and before it takes effect. Without that, the audit cannot run.
  • A short report: whether the interface can be tested yet, and what to fix first.
  • The full fee is credited against a seeded-error audit.
Buy the readiness reviewPaid up front. Scoping form the same day. Need an invoice or purchase order? Ask us.
Recommended

Seeded-error audit

€8,250+ VAT · about 4 weeks
  • Written authorisation and ethics terms agreed before any test error goes in, including workforce-level notice and no disciplinary use.
  • Test errors built from the errors your system actually makes, at severities that matter.
  • Two weeks of unannounced test errors in the live queue, at realistic frequency.
  • Your threshold recorded in writing before the first test error.
  • A report on the interface and its conditions, never on individuals, with the changes we recommend.
Request the auditStarts with a scoping call. Invoiced.

Audit and retest

€12,100+ VAT
  • The complete seeded-error audit, with your threshold recorded in writing first.
  • After your fixes, a retest within three months, with new test errors your reviewers have not seen.
  • Before-and-after results against the same threshold.
Request the audit and retestStarts with a scoping call. Invoiced.

Is it right for you?

For

  • Organisations where people check an AI system's outputs before they take effect.
  • Teams that need evidence the review works, not only a record that it ran.
  • Teams that want an independent check of an oversight interface.

Not for

  • Workflows where approval executes immediately, such as a control action or a live transaction. The method does not apply, and we will tell you so at the scoping call.
  • Assessing individual reviewers. Results are reported at the level of the interface and its conditions, never as individual performance data.
  • The Greek public-sector AI Registry declaration on its own. Registry declaration with a tested oversight check (in Greek) includes a shorter, one-week test as evidence for item (θ). The two are priced separately.

How it works

Readiness review

  1. Buy online. Payment is up front.
  2. You get a scoping form the same day.
  3. Five working days on one oversight interface.
  4. A short report: can it be tested yet, and what to fix first.

Seeded-error audit

Every audit starts with a scoping call. Then the protocol runs in five steps:

  1. Ground rules. Written management authorisation and ethics terms, fixed before any test error goes in.
  2. Test-error design. Test errors drawn from your system's real error analysis, at severities that matter. Trivial ones produce flattering numbers.
  3. In-situ insertion. Known-wrong outputs go into your reviewers' real queue, at realistic frequency, without announcement.
  4. Measurement. We measure whether reviewers, working through this interface under this load, detect and reject them.
  5. Reading the result. The detection rate is read against the threshold you committed to before the audit began.

Proof and further reading

Matthaios Mantzios, founder of UXellence.

Who does the work

Matthaios Mantzios leads every engagement himself, and we take on a limited number at a time. Matthaios is Greece's delegated expert in CEN-CENELEC JTC 21 Working Group 4, "Foundational and societal aspects", appointed in September 2026 on the nomination of ELOT, the national standards body.

Questions

Do our reviewers know they are being tested?

The practice is disclosed, the timing is not. Your workforce is told that audits with planted test errors occur, on aviation's precedent, and no result is used for discipline.

What detection rate counts as a pass?

We set no universal threshold. Before the audit runs, you record the rate below which you will change or withdraw the review step, and justify it against your risk assessment. The audit reports against that number.

What if approval takes effect immediately?

Then the method does not apply, because a test error must be stopped before it takes effect. The readiness review checks this first.

What if we need more afterwards?

The readiness review fee is credited in full against a seeded-error audit. Audit and retest adds a second round, within three months of your fixes, with new test errors.

Ready to get started?

Tell us where you stand. We'll suggest the right first step.

Talk to us