A ranked shortlist with a named human behind it

It would be straightforward to let the tool auto-reject everyone below the line. We deliberately do not build that, and the reason is not squeamishness. A system that orders a list and a system that removes people from consideration sit in very different places under automated employment decision rules.

Where an automated employment decision tool is in scope, obligations attach to the tool that substantially assists or replaces a hiring decision. A ranking that a named person reviews and can overturn in one click, with no candidate removed and every candidate still visible, has a far narrower surface than a screening filter that quietly disposes of two thirds of the applicants before anyone reads them. Keeping the boundary at ranking is a design decision that makes the product easier to defend and easier to audit, and it happens also to produce better hires, because the eleventh candidate is frequently the one worth calling.

Every shortlist run is written to the requisition: when it ran, which candidate set it considered, the configuration and weightings in force at the time, the order it produced, which names a recruiter moved, and the user who approved the final list.

That matters twice. Internally it answers the question of why these five were presented to the client and not those five, months after everyone has forgotten. Externally, where an independent bias audit is commissioned, the product records which candidates were scored and exports the data such an audit needs. Commissioning the audit and issuing any candidate notice remain your responsibility as the employer or the agency; the record you would need for it is kept for you.

Shortlist settings are attached to the requisition rather than hidden in an administrative area, because the person who should be tuning them is the person working the job. You set the list length, the weighting profile, whether database candidates are considered alongside direct applicants, and which requirements act as hard gates.

Weighting profiles differ honestly by desk. A contract desk usually needs availability, notice period and rate at the top. A retained search needs sector history and seniority band and cares far less about immediate availability. Defaults come from the vertical template chosen at signup, and the profile in force is shown on every run so a strange result can be traced to its configuration rather than blamed on the model.

Once approved, the shortlist is something you act on rather than a report you read. Names can be moved to a submittal in bulk, shared with a client contact through the portal without giving them a seat, or sent an interview invitation directly from the shortlist screen.

Interview invitations from a shortlist carry the requisition context with them, so the scheduling request already knows the panel, the stage and the interview kit that stage uses. Client feedback comes back against the shortlist rather than into somebody's inbox, which is how the submittal-to-interview ratio for that client stays a real number.

A shortlist is only as good as the requisition behind it. The most common failure is not a bad ranking; it is a brief that never distinguished a hard requirement from a preference, so everything scored close together and the ordering became arbitrary.

Dismissal reasons are the feedback loop. When a recruiter removes a proposed name, they choose from the same coded disposition list the pipeline uses, and those reasons accumulate against the requisition. Nine removals for salary above budget is a signal about the brief and about the client conversation, not about the shortlisting. The requisition health view surfaces that pattern rather than waiting for someone to notice it in a monthly report.

AutoShortlist can be disabled for a tenant, for a client or for a single requisition. Some clients contractually require that no automated tool assists candidate selection on their work, and some jurisdictions make that the cautious default.

When it is off, it is off, and the requisition records that it was off for the whole time it was open. That record is worth as much as the runs themselves, because being able to show that a tool was not used on a given search is frequently the question that gets asked.