From "the vendor says it's fine" to signed evidence in eight working days
Every HumanLens engagement runs the same five steps. The only thing that changes is who signs at the end: our expert for a single assessment, your own experts under ongoing governance.
Know what you are governing before you test it
We inventory the system, its owners and the people it affects, then tier its risk on a handful of dimensions reviewers can actually defend: domain, internal or external use, how many people are affected, whether it was built and tested properly, and whether it is deterministic or generative.
Tested with the HumanLens test library
Classical models get accuracy and bias testing against your data. Generative and RAG systems get an evaluation set scored for accuracy and bias, plus a manual expert read of a subset so automated scores are checked against a human eye. Every run is logged, so the evidence exists the moment the test does.
"Answer cites the right pathway but omits the licensure prerequisite. Flag for remediation."
Scarce human attention, routed where it matters
Reviewers do not read everything. They read what the tier and the test results say they should: the high-risk calls, the failed checks, the edge cases. Findings are written up with the evidence attached, so the person who signs can see exactly why.
One document your auditor, your board and your engineers can all read
The system card records what the system is, who owns it, how it was tested, what was found and who signed. Single assessments are signed by HumanLens's expert. Under ongoing governance, your own experts sign, because accountability belongs with the organization that owns the AI. The audit trail is generated, not assembled afterwards.
[SYSTEM NAME]
Owner · [NAME, ROLE]
Governance that keeps up with the model
Models drift, vendors ship updates and new systems appear. Ongoing governance keeps every covered system on a review cadence, re-runs the tests when something changes and keeps the system cards current, so the next audit is a print job rather than a project.
One assessment first. Then keep it current.
Single assessment
One system, eight working days, fixed scope and fixed price. HumanLens's expert signs.
Ongoing governance
Every covered system on a review cadence, priced per system. HumanLens supplies reviews and evidence; your own experts sign.
Have your own reviewers and want to run the process yourselves? The governance tool our team uses is available to design partners. Ask us about access.
Security testing is not currently in scope. We assess accuracy, bias and fairness, documentation and accountability; we will tell you plainly when a question belongs with a security partner.