Adverse impact analysis and bias auditing

The Uniform Guidelines on Employee Selection Procedures (1978) provide a rule of thumb: compute the selection rate for each group, identify the group with the highest rate, and divide every other group's rate by it. A ratio below 0.8 is treated as evidence of adverse impact warranting investigation.

It is a screening heuristic, not a legal standard, and it is unstable in small samples — a single hiring decision can swing the ratio across the threshold in a pipeline of forty. TalentSpark reports the ratio alongside the sample size and a statistical significance test, because a ratio of 0.72 on eleven candidates means nothing and a ratio of 0.79 on nine hundred means a great deal.

By the time an annual audit surfaces a disparity, you have made a year of decisions on the process that produced it. The disparity is now a pattern, the pattern is documented, and every individual decision inside it is harder to defend than it would have been in isolation.

Continuous monitoring changes the posture from discovery to prevention. Selection rates recompute as candidates move through the pipeline, and drift raises a flag while the sample is still small enough to interrogate.

Three regimes matter to most customers, and they are converging on the same requirement: produce records showing what your system did and why.

Content validity evidence linking each assessed construct to a job analysis. Selection rates by protected group, computed on the live pipeline with sample sizes attached. A per-decision log capturing the scores, the flags, the generated recommendation and the human who decided.

TalentSpark produces all of it as a by-product of ordinary use, because a compliance artefact assembled retrospectively is an artefact you are asking someone to trust. One assembled continuously is evidence.