The Most Consequential Person in Hiring May Be the Least Qualified to Assess the Work
Recruiters decide who experts get to assess. In an AI-mediated market, that first gate may matter more than the sophistication of everything that follows.
Recruiters perform one of the most consequential functions in the labour market. They decide who gets seen. Before the technical interview, investment case or partner meeting, somebody determines whether a candidate deserves to reach the person qualified to judge the work.
Specialist recruitment expertise and specialist occupational expertise are different. Yet the former frequently determines who reaches the latter.
The first screen is already a capability decision.
Most sophisticated hiring processes eventually introduce genuine subject-matter expertise. Engineers interview engineers. Lawyers interview lawyers. Investors interview investors. That is intuitive because complex professional capability is difficult to evaluate without understanding the work.
Formal job-analysis practice starts from the same premise. The U.S. Office of Personnel Management defines a subject-matter expert as somebody with bona fide knowledge of what it takes to do a particular job, often a supervisor or strong incumbent with current knowledge of its requirements.1
Recruitment often reverses that logic at the beginning of the funnel. The expert assesses candidates later. The recruiter decides who the expert gets to assess.
If 300 people apply and twelve progress, the hiring manager is not choosing the best candidate from 300. They are choosing from the twelve people an earlier filter judged legible enough to continue. A strong final process cannot recover somebody rejected before reaching it.
The accuracy of the first screen places an upper bound on the quality of everything that follows.
Knowing the market for the work is not the same as knowing the work.
Sandery interpretationThere are exceptional specialist recruiters. Years of conversations, searches and relationships can produce deep knowledge of employers, compensation, team movements, reputations, candidate motivations and the practical mechanics of a transaction.
That is substantive expertise. But it is expertise in the market for professionals—not necessarily first-hand expertise in the profession.
A recruiter who has spent ten years recruiting investors may know which funds develop talent, how titles translate and which profiles receive interviews. Unless they have invested capital, they may have no direct basis for determining which candidate has better investment judgment.
An investor may instead ask: what did you believe that others did not? Which assumption drove the downside? What information changed your view? Where were you wrong? Those questions interrogate the work. Prestige, title and tenure become context rather than the assessment.
why and at what price?
work actually require?
Recruitment experience teaches patterns of acceptance.
Empirical evidenceRepeated exposure teaches a recruiter which employers, titles, qualifications and trajectories their clients recognise. That knowledge is commercially valuable. It can also be mistaken for a validated model of performance.
SHRM surveyed more than 1,030 recruiters and ran an experiment using simulated applications. At the initial screen, 84% said years of experience mattered and 83% cited type of experience. Skills credentials and education carried substantially less weight.3 The study describes what recruiters considered, not whether those factors produced the best future employee.
A recruiter's internal evidence base also has a missing-data problem. They observe who clients interview, reject and hire. They may have much less systematic evidence about people excluded early who could have performed exceptionally.
Decision science calls a related problem selective labels: outcomes are observed for cases allowed through an earlier decision, but not for those the decision excluded.5 That research examines judicial decisions rather than hiring. The application here is our interpretation.
A recruiter can become highly specialised in predicting who will be selected without becoming equally good at predicting who will perform.
The first gate has more to process and less reliable material to process it with.
Greenhouse analysed more than 17 million applications across over 800 European companies. Between 2022 and 2025, applications per vacancy rose from 99 to 183. Annual applications per recruiter rose from 98 to 474, while recruiters per organisation fell from 14.41 to 6.14.8
At the same time, candidates are using generative AI throughout the job search.9 A CV can be rewritten around stated criteria, terminology aligned, experience reordered and an awkward transition made persuasive in seconds.
Presentation has always influenced evaluation. A Journal of Management study of 72 applicants found self-promotion was strongly associated with interviewer perceptions of person–job fit.4 It predates generative AI and does not test AI-assisted applications. It does show that the representation of capability can shape the perception of it.
Our interpretation: the initial screen increasingly evaluates the candidate plus the quality of their optimisation system. Those variables are becoming difficult to separate.
Increase from 99 applications in 2022 to 183 in 2025.
Greenhouse, EuropeIncrease from 98 annually in 2022 to 474 in 2025.
Greenhouse, EuropeDecline from 14.41 in 2022 to 6.14 in 2025.
Greenhouse, EuropeAI can make a weak selection model faster—and more consistent.
A job description represents the role. An AI-optimised CV represents the candidate. A recruiter compares the two, increasingly with assistance from another system. The entire first stage can become sophisticated without anybody examining the work at meaningful depth.
This matters because consistency is not automatically accuracy. Stanford researchers followed 3.4 million people submitting four million applications across 1,700 postings, 150 employers and eleven sectors, all assessed by one vendor's tool. They found evidence that the same candidates could be rejected repeatedly across employers, alongside position-level adverse impact.7
The study is observational, concerns one vendor and does not establish that every automated hiring system behaves similarly. It does demonstrate a more general risk: when organisations adopt common filters, the quirks of a model can travel across the market.
Automating historical selection patterns may make recruitment more efficient. It may also make yesterday's definition of an acceptable candidate harder to escape.
Ease of explanation can masquerade as quality.
Sandery interpretationA conventional candidate is easy to communicate. Recognisable employer. Expected title. Familiar progression. Relevant industry label. The client immediately understands why the person is on the shortlist.
An unconventional candidate requires translation. Someone from a smaller organisation may have exercised greater responsibility without the brand signal. A cross-functional career may have produced rare capability without a familiar title. A move between industries may demonstrate transferability while making the profile harder to categorise.
Contingency recruitment makes the tension particularly visible because payment generally occurs only when a placement is made.10 The following is a simplified model, not a claim about every recruiter:
÷ resources consumed
of the best available hire
Those objectives overlap substantially, but not perfectly. A brilliant candidate who requires extensive explanation may carry more transaction risk than a very good candidate the client recognises immediately.
The risk is structural, not moral. Rational people respond to the information, incentives and time available to them.
The first screen should not try to identify the winner.
Application volumes make filtration unavoidable. Senior practitioners cannot assess everybody. The question is what the first filter should attempt to determine.
A false positive sends a weaker candidate to further assessment. A false negative removes a potentially exceptional candidate permanently. At the first gate, those errors are not equally expensive.
The conventional question is often: does this person look like somebody the client would plausibly hire?
A better question is: is there sufficient evidence that this person could perform the required work to justify expert assessment?
The first optimises for conventional acceptance. The second protects capability recall.
The first screen does not need to identify the winner. It needs to avoid eliminating them.
Moving expertise forward changes the gate.
Institutional evidenceA U.S. Digital Service and OPM pilot tested a qualification process in which subject-matter experts defined requirements alongside HR specialists, then reviewed CVs and structured phone-interview transcripts before candidates were deemed qualified.2
Compared with earlier agency hiring events, the experts judged fewer applicants qualified—11% at the Department of the Interior and 22% at Health and Human Services—yet the pilots ultimately produced thirteen and seven hires respectively. At Interior, average time to candidate selection fell from 45 days to 16.
This was an operational pilot, not a randomised trial, and cannot prove the model will generalise. It is nevertheless a useful demonstration: recruiter and practitioner expertise can be combined earlier without asking either group to pretend to be the other.
Updated meta-analytic work in personnel selection also found structured interviews remained the highest-ranked common selection procedure after historical validity estimates were revised downward.6 Expertise matters most when it is given structured, job-relevant evidence to interrogate.
Hiring contains three different questions.
Recruitment becomes clearer when screening, assessment and intermediation are treated as distinct problems.
- Who might be capable?
- Protect plausible candidates
- Prioritise evidence
- Escalate uncertainty
- Who demonstrates capability?
- Interrogate the work
- Use structured methods
- Test contextual fit
- Can a transaction occur?
- Build trust
- Resolve concerns
- Align expectations
Strong recruiters can be exceptionally valuable across the first and third questions. They understand people, markets, mandates, motivations and transactions. The mistake is allowing that expertise—and repeated exposure to historical selections—to imply definitive expertise in the second.
A specialist recruiter should not need to claim to be a better judge of engineering than engineers, investing than investors or law than lawyers. Their role becomes more credible when its real value is stated precisely.
Give human judgment better evidence before the gate closes.
The Sandery viewAI could reinforce the existing system: train on historical placements, rank candidates by resemblance and optimise for likely acceptance.
Or it could move below the profile. Decompose the opportunity into the work required. Preserve deeper evidence about responsibilities, decisions, achievements, operating environments and outcomes. Recognise transferable experience without requiring an exact career-pattern match.
The recruiter then becomes the person ensuring potentially relevant talent reaches the right assessment—not the person asked to make a definitive judgment about specialist capability from a thin representation.
Sandery does not need to replace the person qualified to assess the work. It needs to improve the evidence available before that assessment—and help ensure the right people reach it.
- Persistent career evidence
- Context and provenance
- Longitudinal understanding
- Individual control
- Informed screening
- Specialist assessment
- Targeted validation
- Trusted intermediation
The Sandery view
The gatekeeper should not be the final epistemic authority.
The first screen cannot disappear. There will always be more plausible candidates than organisations can assess deeply.
But the gate should recognise the limits of what it knows. Its purpose is not to declare who is the superior investor, engineer, lawyer or operator. It is to determine whose evidence deserves examination by somebody qualified to make that judgment.
As applications become more polished and screening more automated, that distinction becomes more important—not less.
The implication
The great risk is not that the wrong candidate performs badly in the final interview.
It is that the right candidate never reaches it.
The future of recruitment should not ask recruiters to be something they often are not. It should make them exceptionally good at ensuring non-obvious capability is not filtered out before somebody qualified to judge it gets the chance.
Build the evidence before the gate
Make more of the career visible.
References
Evidence used in this note.
The distinction between predicting selection and predicting performance, the placeability analysis and the proposed hiring architecture are Sandery interpretation. The evidence supports the importance of role expertise, the pressure on initial screening and the value of structured, job-relevant assessment; it does not establish that every recruiter, employer or automated system behaves uniformly.
Research & institutional evidence
- What is a subject matter expert?U.S. Office of Personnel ManagementJob-analysis guidance on the role of people with current, thorough knowledge of the work ↗
- Changing how the government hires technical talentU.S. Digital Service and OPMInstitutional pilot introducing subject-matter expertise earlier in qualification assessment ↗
- Recruiters Say Experience Top Factor in Applicant EvaluationSociety for Human Resource ManagementSurvey and simulated-application experiment involving more than 1,030 recruiters, 2023 ↗
- Applicant Impression ManagementAmy Kristof-Brown, Murray Barrick and Melinda FrankeJournal of Management · Study of presentation effects on recruiter perceptions of fit, 2002 ↗
- Human Decisions and Machine PredictionsJon Kleinberg et al.Quarterly Journal of Economics · Analysis of selective labels in human decision systems, 2018 ↗
- Revisiting Meta-Analytic Estimates of Validity in Personnel SelectionPaul R. Sackett et al.Journal of Applied Psychology · Revised validity estimates across selection methods, 2022 ↗
- AI Hiring Tools Can Yield Racial Bias and Systemic RejectionStanford Institute for Human-Centered Artificial IntelligenceObservational study of 3.4 million people and 4 million applications assessed by one vendor, 2026 ↗
Industry research
- Hiring benchmarks 2026GreenhouseEuropean platform data from more than 800 companies and 17 million applications, 2022–2025 ↗
- The 2026 AI in Hiring ReportGreenhouseSurvey research with candidates, recruiters and hiring managers across the UK, Ireland and Germany ↗
- RecruitmentSmartRecruitersIndustry definition of contingency recruitment and pay-on-placement economics ↗