The Resume Screener With a 0.20 Impact Ratio and a Refused Accommodation
Corvus Bank is filling 60 customer-service representative openings from 1,200 applicants. To cut screening time, the talent team deployed a vendor's AI resume-scoring tool in January, configured to weight career continuity and prestige of prior employer. Results: of 800 applicants who did not self-identify as a member of a protected racial group, 120 were advanced to interview; of 400 who did, 12 were advanced. The bank holds $4 million in federal contracts annually and maintains an affirmative action plan with the Section 503 seven percent utilization goal for individuals with disabilities. During the pilot, a candidate who uses a wheelchair asked to skip the timed typing test and submit a work sample instead; the recruiter told him the test is a business necessity and cannot be modified.
1. Compute and interpret the adverse impact. What legal exposure does the bank face, and what does the vendor's validation claim actually buy the bank?
Read the worked answer
Selection rates: 120 of 800 equals 15.0 percent for the non-minority group; 12 of 400 equals 3.0 percent for the other group. The impact ratio is 3.0 divided by 15.0, or 0.20, far below the 0.80 threshold in the Uniform Guidelines. With those counts the gap is also statistically significant, so the four-fifths rule is not the only problem.
Exposure: disparate impact under Title VII, and because Corvus is a federal contractor, OFCCP obligations under Executive Order 11246 and Section 503. An automated scorer is a selection procedure under the Uniform Guidelines even though it is vendor-built and resume-based. The vendor's validation material is not a defense unless the study matches Corvus's jobs, criteria, and applicant pool; a validity study on another employer's warehouse roles transfers nothing.
The suspicious weights are the reasoning core: career continuity and prestige of prior employer are not demonstrably job-related for a customer-service representative, and both proxy for socioeconomic background and access to elite employers. Practical guidance: suspend the tool for screening decisions now, preserve its configuration and the disposition data, and do not fix the problem by moving the cutoff until the rates look acceptable. Adjusting the threshold to hit a demographic target is race-norming and is worse than the original defect.
2. Was the recruiter correct about the typing test? Explain the legal basis and the steps the bank should take now.
Read the worked answer
The recruiter is wrong on both the facts and the law. Under the ADA, and under Section 503 for a federal contractor, an application procedure must be reasonably accommodated unless doing so imposes undue hardship, or the test is job-related and consistent with business necessity and no alternative exists. A work sample is a plausible alternative that measures the same construct, and wheelchair use has no bearing on typing. The refusal is also procedurally defective: the employer must engage in an interactive process rather than issue a blanket business-necessity statement, and it must document the analysis.
Steps now: approve the work sample or an alternate untimed format, score it with the same rubric applied to all candidates, and log the request, the response, and the rationale. Train recruiters that accommodation requests at the application stage are routine and do not require a diagnosis. Check the disposition codes for this candidate, because inaccurate coding in a federal contractor's applicant flow data creates a separate compliance problem. Finally, if the typing test itself has never been validated against performance for this role, the accommodation question is masking a larger selection problem that the redesign must address.
3. Recommend the redesign path and the metrics you would monitor. Why is validation an ongoing obligation rather than a one-time project?
Read the worked answer
Redesign: (1) derive KSAOs from a job analysis of the customer-service role, covering typing speed, written communication, problem-solving, and de-escalation; (2) replace resume scoring with a scored application screen plus a work sample and a structured behavioral interview with anchored rating scales; (3) build the validity evidence, content validity for the work sample and criterion-related or transportability evidence for any test; (4) pilot on a live requisition and compute impact ratios at every stage before full deployment; (5) monitor applicant flow and selection rates by race, sex, disability, and veteran status at screen, interview, offer, and hire, along with progress toward the seven percent utilization goal under Section 503; (6) set a quarterly review cadence tied to the affirmative action plan.
Dashboard metrics: stage-by-stage conversion by group, impact ratios, time-to-fill, offer acceptance, quality-of-hire, and disposition-code completeness.
The reason this cannot be a one-time validation: adverse impact is a property of the procedure as applied to your applicant pool and your job, and it shifts as the pool, the job, and the tool's configuration change. A vendor white paper cannot satisfy the obligation for a selection rate that is 0.20 in your own data.