One-page map

    Where high-risk hiring decisions slip past review

    Most talent teams can say where a human sits in an AI-assisted hiring decision. Far fewer can say whether that person knows, in the moment, that the decision in front of them carries more risk than the ten before it.

    The problem

    When every CV score, assessment result and shortlist arrives looking the same, review becomes a reflex. Research on automation bias finds that people defer to a confident system, and defer more under high workload. The recruiter is present, but not engaged.

    The result is oversight that holds on paper and fails in practice, on exactly the decisions a candidate is most likely to challenge: who was screened out, who was rejected, and why.

    The map

    A typical AI-assisted hiring funnel, stage by stage

    1. 01Sourcing

      Human: Recruiter reviews AI-suggested candidate lists and outreach.

      Missed: Who the search never surfaced, and why.

    2. 02CV screening

      Human: Recruiter spot-checks a sample of AI match scores.

      High risk: Borderline scores and non-standard CVs filtered out unseen.

    3. 03Assessment scoring

      Human: Hiring team reads test and video-interview scores.

      Missed: Whether the score is reliable for this candidate.

    4. 04Shortlist

      Human: Hiring manager approves the ranked shortlist.

      Missed: That the ranking order shaped who was seen at all.

    5. 05Rejection

      Human: Often no one: rejections can go out automatically in bulk.

      Highest risk: Solely automated decisions with legal exposure.

    6. 06Offer

      Human: Hiring manager and talent team sign off together.

      Lower risk: Usually well reviewed; the risk sits earlier in the funnel.

    The decisions that are hardest to undo for a candidate, being screened out or rejected, are often the ones that get the lightest human review. UK data law lets organisations automate more decisions as long as people can reach meaningful human intervention, and review only counts if the reviewer can genuinely change the outcome. In the EU, AI used for recruitment or selection can be high-risk under the AI Act.

    Five signals that should change how a reviewer engages

    A checkpoint is only useful if the reviewer can tell a routine decision from a high-risk one. These are the five inputs the Checkpoint Scale uses to decide whether a decision runs automatically, gets logged, or is held for a human.

    SignalThe question it answersWhere it shows up in hiring
    ReversibilityCan this be undone cheaply if it is wrong?A rejection email or a lost candidate is hard to reverse; a provisional ranking is not
    ConsequenceHow much harm if it is wrong, to the candidate or the business?Automated rejections, final shortlists, offer decisions
    ConfidenceHow sure is the system, and is that confidence earned?Borderline CV match scores, low-quality video or assessment data
    PrecedentHas the system seen cases like this before?New roles, career changers, non-standard CVs, new markets
    Regulatory exposureWould a regulator, auditor or tribunal care how this was decided?Solely automated rejections, decisions touching protected characteristics, hiring in the EU

    When two or more signals run high, the reviewer should be told so before they decide, not after.

    Three questions to ask your team this week

    1. Pick one AI-assisted screening or rejection decision from last week. Can anyone say why a human was involved at that point?
    2. Did the recruiter know that decision carried more risk than the ones around it? What told them?
    3. If a candidate, auditor or regulator asked how that decision was overseen, what record would you show?

    Founding audits

    First audit 50% off

    Priced by company size.

    3 founding spots. 3 remaining.

    Two weeks, fixed scope. You leave with an oversight risk map of every agent-to-human handoff.

    Claim a founding spot

    Find out where your checkpoints sit

    If the honest answer to any of these is "not really", that is the gap a Behaviorist audit closes. We map every agent-to-human handoff in your hiring funnel, score each one on the five signals above, and show where reviewers need to be alerted.

    Book a founding audit
    Sources and method
    1. Parasuraman and Manzey (2010), Complacency and bias in human use of automation
    2. Data (Use and Access) Act 2025, section 80
    3. ICO: guidance on individual rights in AI systems
    4. EU AI Act Service Desk: employment

    This map is an illustrative pattern drawn from common hiring workflows, not findings from a study.