This walkthrough uses entirely fictional recruitment data and a fictional organisation. It demonstrates the BiasLens evidence discipline; it does not represent a real client finding or legal conclusion.
Use the sequence below as the narration and visual storyboard for the production demo video.
A fictional organisation uses an AI-assisted recruitment tool to support candidate shortlisting. The organisation is concerned about whether disabled applicants could experience different outcomes.
What BiasLens preserves: System purpose and decision context are defined before any bias conclusion is attempted.
The vendor says the model is fair, but the organisation has no accessible test report, no clear affected-group breakdown and incomplete documentation about the population used for validation.
What BiasLens preserves: Vendor assurance is recorded as an assertion. Missing validation evidence remains an explicit unknown.
BiasLens examines preexisting, technical and emergent pathways. In this fictional case, representation gaps may indicate preexisting risk, interface barriers may create technical risk, and population or use changes may create emergent risk.
What BiasLens preserves: Possible pathways are documented without presenting them as proven causes.
A fictional outcome comparison suggests a difference between groups, but the available sample is limited. BiasLens treats the result as a signal requiring investigation rather than proof of discrimination.
What BiasLens preserves: Evidence strength, sample limitation and the need for further investigation remain visible together.
The output records what was observed, the evidence status, the rationale, limitations and a recommended next action — for example, obtaining better denominator data, testing accessibility and requesting stronger vendor evidence.
What BiasLens preserves: The organisation gains a documented evidence trail rather than a binary pass/fail label.
The question changes from ‘Is this AI biased?’ to ‘What does our evidence support, what remains unverified and what must we investigate next?’
What BiasLens preserves: That is the operating discipline BiasLens is built to support.
“This fictional organisation uses AI to support recruitment shortlisting. The concern is simple: could disabled applicants experience different outcomes? BiasLens does not begin by declaring the system biased. It begins by defining the system, the decision it influences and the evidence available.”
“The vendor says the model is fair. But an assurance is not the same as verified evidence. BiasLens records what is documented, what is missing and what remains unverified. It then examines possible preexisting, technical and emergent bias pathways without presenting a possible pathway as a proven cause.”
“Where outcome data is available, BiasLens can surface a fairness signal. In this fictional example the sample is limited, so the result is not treated as proof of discrimination. The limitation stays visible. The finding records the evidence status, rationale and next action.”
“The result is not a false green light or a dramatic red flag. It is a traceable evidence question: what do we know, what do we not know, and what must we investigate next? That is BiasLens — evidence-led algorithmic accountability by BeAccessible.”
Read the public methodology note for the principles behind the workflow, or bring one real AI system into the qualification process.
