Independent AI acceptance before you pay
I check whether the system works on your real tasks, where it makes mistakes, what data leaves the company, and whether the contractor’s work can be accepted.
- Can we pay the contractor?
- Can we trust the system’s answers?
- What breaks if the volume or the data change?
Result: a test protocol and a verdict: accept, fix first, or reject.
Scope depends on the number of scenarios and integrations. Acceptance is for systems built by a third-party contractor. My own implementation is a separate format.
Free, 30 minutes, no presentation and no sales pitch. I’ll tell you whether you need an independent check and what exactly is worth checking.
An outside check, not a second sales pitch
My job is not to sell you AI but to protect your budget and your decision.
- I don’t sell a platform. I’m not tied to one vendor or one technology, so I can say “you don’t need AI here, ordinary automation is enough”.
- Criteria are fixed before the work starts. The result is checked against what was written down at the start, not against the impression from a demo.
- Every check is repeated. An AI system can answer the same question differently. One good answer proves nothing, so each scenario is run several times.
Acceptance: checking an AI platform
An excerpt from an independent audit of an AI platform, anonymised. Each check is run several times, because an AI system can answer differently to the same question.
| # | Check | Runs | Result |
|---|---|---|---|
| T-01 | Control case with a known correct answer | 4 | defect reproduced |
Method: how to test an AI system before you rely on it.
What you get in hand
- A test protocol: scenarios, runs, defects with steps to reproduce
- A clear recommendation: accept, fix first, or reject
How is this different from a security audit?
A security audit looks for vulnerabilities and leaks. Acceptance checks whether the system does what you are paying for: whether answers are correct, whether cited sources match the originals, whether results repeat. I also check what data leaves the company, but deep penetration testing is a job for a separate specialist.
Do you need access to the source code?
No. I test the system the way your staff will use it, through its interface or API, on your real tasks.
Can you check a system that is already running?
Yes. Both a system already in use and one a contractor is about to hand over.
What happens to our data?
It stays yours. I work on anonymised examples or inside your environment, and sign an NDA if needed.
Tell me your industry and the task. I reply personally.
Other ways to reach me: LinkedIn