Eminence Growth Solutions, home
Menu

CoE | AI Talent & People

Eminence Growth Solutions

Practice note

A Missing Keyword Is Not Proof That a Candidate Lacks the Skill

Derek Mobley alleges that employers using Workday’s screening tools rejected him for more than 100 jobs, one email arriving at 1:50 a.m. for an application made at 12:55 a.m.; a federal court has let the case proceed and preliminarily certified an age collective. When AI screens applications, the rule that rejects people automatically is a selection procedure, and it needs a definition of the evidence each requirement takes.

Download PDF
Watercolor of substantial work-evidence panels connected to one small empty label field.

Key takeaways

  • Translate each essential requirement into observable work evidence and distinguish entry needs from trainable knowledge.
  • Use consistent follow-up where an application leaves relevant capability unresolved.
  • Test whether changing a product name changes rejection decisions while the underlying work stays the same.

Derek Mobley alleges that since at least 2017 he applied for more than 100 positions with employers using Workday’s platform and was rejected every time. One application, his complaint says, went in at about 12:55 a.m. and drew a rejection at about 1:50 a.m.; he argues the timing shows the applications were screened out automatically. Workday denies the allegations and denies that it is a covered entity under the statutes the complaint cites. The court held in July 2024 that the complaint adequately alleged Workday acted as its customers’ agent, and in May 2025 it preliminarily certified a collective on the age claim. The discovery order filed October 7, 2026 recorded November 10, 2026 as Workday’s deadline to oppose class certification. That is a procedural deadline, not a determination of the discrimination allegations.

The allegations and procedural rulings above do not establish that a keyword rule caused those rejections. This article examines a screening-design question: what happens when a product name is treated as proof of capability, or its absence as proof that capability is missing?

Define the work evidence behind the requirement before turning it into a rule that rejects an application.

The tool name hides the real requirement

The federal government has moved its own hiring the same way. OPM’s Merit Hiring Plan of May 2025 says federal hiring relied heavily on credentials, job titles and years of experience, which can overlook people who gained skills by other routes, and calls skills-based hiring a shift from what applicants say on a resume to what they can do. Its February 2026 guidance tells agencies to judge the complexity and relevance of the work a person has done, “not just duration or level”, and, where crediting plans are used, to avoid evaluative words such as “expert” or “extensive” in the questions and response options.

Separate day-one requirements from trainable ones

Define which requirements are essential, why, and what evidence can meet them. OPM calls job analysis the foundation for assessment and selection decisions: it establishes the competencies a job requires and their link to its tasks. OPM’s job-analysis method rates each competency on whether it is needed at entry, from “Needed the first day” to “Must be acquired after the first 6 months”. Decide whether platform-specific knowledge is essential on entry or whether the requirement is the underlying work. Write the reason into the assessment standard.

Make “insufficient evidence” an allowed answer

  • Ask the AI for the exact passage that supports each proposed match, and for “insufficient evidence” where the application does not answer the question.
  • Route incomplete but relevant applications to a consistent follow-up, such as a short work sample or a structured screening question, rather than letting the model create its own exceptions.
  • Review a sample of rejections, especially those near the threshold and those written in unfamiliar terms, against a job-related assessment.
  • Do not write inferred skills into a person’s record as established fact. That is a different error from rejecting them too early.

Unknown is a useful decision state. Forcing every application into yes or no hides the most important review question.

An automatic rejection is a regulated selection procedure

The law increasingly treats automated screening as a selection decision in its own right.

  • New York City. Local Law 144 covers an automated employment decision tool that substantially assists or replaces discretionary decisions, and bars its use to screen candidates without a bias audit within the prior year and a published summary. Candidates must get at least ten business days’ notice, with instructions for requesting an alternative process or accommodation, though the rules do not require an employer to provide an alternative process.
  • California. Civil Rights Council regulations in force since October 1, 2025 cover automated-decision systems used in employment decisions, treat an employer’s agent as including anyone screening or hiring on its behalf, and require four years’ retention of automated-decision data. The privacy agency’s rules give software that screens resumes to decide whom to hire as an example of automated decisionmaking, and businesses using it for hiring decisions must comply with the notice, opt-out and access requirements from January 1, 2027.
  • European Union. The AI Act lists AI used to filter applications and evaluate candidates as high-risk, with those obligations applying from December 2, 2027.
  • Federal. The Uniform Guidelines treat a selection rate below four-fifths of the highest group’s rate as generally regarded as evidence of adverse impact. Since April 2025 an executive order and a June 2026 Justice Department opinion have turned federal enforcement away from disparate impact; Mobley is a private suit and continues.

The EEOC’s 2023 settlement with iTutorGroup shows how plain the theory can be: the agency alleged the company’s software automatically rejected women aged 55 or older and men aged 60 or older, and the company agreed to pay $365,000 under a consent decree entered in September 2023. “Even when technology automates the discrimination, the employer is still responsible,” the EEOC’s then-chair, Charlotte Burrows, said when the suit was filed in May 2022.

Are you at risk?

  • Do any knock-out rules in your applicant tracking system reject applications without a person seeing them?
  • For each essential requirement, is there a written definition of the evidence that meets it?
  • Can the screen answer “insufficient evidence”, and what happens to those applications?
  • When did someone last review a sample of automated rejections?
  • Does the tool fall under Local Law 144, California’s rules or the EU AI Act for any of your locations?

Screen for the work, not the word

Let AI widen the search for evidence, then apply one consistent standard to what it finds, and keep a person on every rejection the evidence does not settle. The aim is a better-supported shortlist, not an objective-looking score with an invisible definition.

Test it in your next cycle: take a few strong past applications, replace the product names while keeping the work they describe, and check whether the screen’s decisions change.

Sources and further reading

Review status

Orientation, not legal advice.

Published by Eminence Growth Solutions
https://eminencegrowthsolutions.com/coe/talent/missing-keyword-not-missing-skill/

Help improve this page

Better intelligence comes from better information. You don’t need to have a fully researched recommendation or a formal correction. Sometimes a question, observation, or firsthand experience is enough to identify something worth investigating.

Questions, corrections and responses from organizations we cover: egsconnect@eminencegrowthsolutions.com

Stay Connected

Follow what matters
to you and your work.

AI Cost & Governance | AI Talent & People | Vendor Intelligence
Eminence Scan | EmTrace | Eminence Edge

Choose updates by center and tool.

Choose your alerts

Research and practical tools, chosen by you.

Sign in with your email address. There is no password.