About AI Test Group
AI Test Group tests what your product does when its AI is wrong. We specialise in health tech, where a wrong answer costs the most, and every result has a named person behind it.
Who we are
Every engagement is run by people who understand your product and what a wrong answer would mean for the people using it. Not a queue of tickets handed to whoever is free, and not a template filled in after the fact.
We test the product you actually built, on your staging environment, by feeding it the wrong answers your AI could give and recording what happens next. The pack you end up with is a record of work that happened, not a document written to look like it did.
20+
Years in regulated systems
8
Runs of every test scenario
Our leads
A small team that does the work itself, with a named person behind every result.
Founder and CEO
Helping health tech teams find out what their product does when its AI is wrong.
I’m the founder of AI Test Group and co-founder of ClinVerify. AI Test Group exists for one problem: teams check whether their AI’s answers look right, but rarely test what the product does when an answer is wrong. We test that, and a named person signs every result.
Head of Engineering
Head of Testing
Clinical Safety Lead
Head of Infrastructure
Business Relationship Manager
How we work
Benchmarks tell you how a model does in general. We test what your product does when the model is wrong, on your staging environment, with your code untouched.
AI rarely answers the same way twice, so every scenario runs eight times. A safeguard that holds six times out of eight is not a safeguard.
Correctness is measured only against reference answers your own experts agree. We never decide for you that your product’s answers are right.
Every result, and every difference after a model update, is reviewed by a person you can name. Every pack also states what it did not test.
What clients get from us
Good testing is not only a final pack. It is the right questions asked at the right point, and the answer kept current as your model, provider and prompts change.
Before any test runs, we agree the AI feature, the ways it could be wrong, the safeguard meant to catch each, and what your experts count as a wrong answer.
The pack is written so your engineers, your buyers, your board and your clinical safety lead can follow what happened without us in the room.
When your model, provider or prompt changes, the same set runs again, and a named person reviews every difference before you release.
A free 45-minute session. One AI feature, five ways it could be wrong, and the safeguard meant to catch each. You keep the map.
