Screening Bias and Personalization Audit (QA)

Screening Bias and Personalization Audit (QA)

1. 01

   Pull the last 90 days of screening decisions from the ATS and tag each by source channel

   Channel-level pass rates expose whether a job-board integration is quietly filtering out qualified applicants before a human ever opens the file.

2. 02

   Re-score a random sample of 30 rejected candidates against the client's written must-have criteria

   If more than 3 of the 30 clear the bar on manual review, the automated screen is over-rejecting and the client's time-to-hire numbers are being propped up by volume, not fit.

3. 03

   Compare rejection reasons across demographic proxies the platform exposes, such as name, location, and education field

   Document the comparison even when the spread looks even; the written record is what protects the agency if a client's legal team asks how screening works.

4. 04

   Read 20 automated outreach messages end to end and flag any that misstate the role, the client, or the pay range

   Template drift is common after a client changes a job spec mid-search, and a candidate who receives a stale pitch tells the hiring manager before the agency hears about it.

5. 05

   Verify that every candidate who reached a human interview received at least one non-automated touchpoint

   A recruiter reply, a scheduling call, or a personalized note counts; three sequential bot messages does not, and candidates notice the difference.

6. 06

   Log findings in a one-page audit note with the sample size, the rejection-rate delta, and the specific workflows changed

   Attach the note to the client's retainer file so the next quarterly review starts from evidence rather than a fresh argument about whether the tool works.

7. 07

   Re-run the sample check 30 days after any threshold or prompt change to confirm the fix held

   Screening models and scoring prompts drift as job boards change their applicant pools, so a single clean audit is a snapshot, not a clearance.