picked.
Log inPost a role
Product · Decision engine
Every score explainable. Every decision auditable.
The bit you cannot copy. Validated psychometric science, adaptive AI, explainable per-trait scoring, and bias monitoring on every role, every stage, every protected group. Not a black box. A glass box with the working shown.
Post a role →Read the model cards
What a score is made of
Four ingredients. Every one visible.
01
Traits.
The competencies we are measuring, mapped from the role's must-haves. For an SDR, that is something like resilience, commercial curiosity, structured communication. The trait set is visible before the role goes live.
02
Evidence.
For every trait, the items, transcripts, and scenarios that contributed. You can click any score and see the bits the model relied on.
03
Weight.
How much each trait counted toward the final score for this role. Set by the role's profile and adjustable by you.
04
Calibration.
Your own preferences, learned over time. After three roles, the model knows that for you, "shipped end-to-end" matters more than years of experience. It tells you when it applies that.
Fairness, per role, per stage, per protected group
Monitored continuously. Reported transparently.

Every assessment, every interview, every shortlist is monitored for adverse impact against protected groups, per role, per stage, per audit cycle. The 4/5ths rule is the floor; we report against tighter thresholds and publish the breach rate annually.

  • Per-role drift detection: if a role's pass rates shift on a protected attribute, we know and so do you.
  • Adverse-impact dashboard in product, shareable with HR or legal.
  • Annual independent bias audit, summary published. See /trust/fairness once live.
  • A breach is a stop-the-line event. We pause the role; we investigate; we explain.
When you over-rule the model
Thumbs down. The list re-ranks with reasoning.
Your calibration is the most important signal we get on a role. Disagreement is not a failure case; it is a feature.
See the shortlist page →
The post-hire loop
Every hire closes the loop on the recommendation.

At 90 days and again at 12 months, the hiring manager confirms how the hire is performing. That signal feeds back into the model. Recommendations that turned out to be right reinforce the weight; recommendations that turned out to be wrong correct it. Validity is not declared at launch; it is measured continuously.

V3 closes this loop with an outcome-based pricing tier: pay only if the hire stays six or twelve months. Possible because we own the outcome data.
Anti-fraud, end to end
The same four layers, across the funnel.
01
AI-assistance detection.
Cross-stage; an AI-written essay in the assessment plus an unusually fluent screen answer plus a hesitant interview triangulates better than any single layer.
02
Proxy detection.
Voice biometrics across screen, assessment, and interview, with consent. One voice across three stages or none.
03
Behavioural consistency.
Persona drift across the funnel is itself a signal. Honest candidates are consistent; coached or proxied candidates rarely are.
04
Liveness (V2).
When video lands, face-consistency and liveness checks. V1 is voice-only so this is deferred.
Candidate rights
Any candidate can ask for a human review of a decision.
GDPR Article 22 compliance is not optional and not lip service. A candidate who asks for a human review gets one, within five working days, conducted by a Picked-trained reviewer with full access to the file. The original decision is paused while review is open.
Read the candidate rights →
Other product surfaces
PRODUCT · SHORTLIST
Three finalists. Forty-eight hours.
The dashboard moment, every score explained.
Read →
PRODUCT · ASSESSMENT
The Neuroworx item bank, ported to AI.
Cognitive, behavioural, situational, role-specific.
Read →
PRODUCT · CANDIDATE EXPERIENCE
The other side of the marketplace.
Real conversation. Real decision. Profile they own.
Read →
See the working
Glass box. Not a black box.
Post a role →Read the trust posture
picked.
The AI-native replacement for inbound hiring. For the hiring manager, not HR.
© 2026 Picked · Made in London · Three finalists in forty-eight hours.
Decision engine · Picked.ai