How AI is used in hiring

No candidate is rejected or advanced by software alone. Scores support a person’s decision; they do not make it.

Sourcing agent: Finds people who match an open role and adds them to the role's longlist.

Screening agent: Reads each CV and records evidence against your stated requirements.

A score is calculated against the rubric for one role. The rubric lists the criteria, a weight for each, and whether each is a must-have.

Each criterion is checked against evidence: the passage of the CV it came from is stored with the score. A score run is tied to one candidate and one role, and it is append-only, so a later change creates a new run instead of editing the old one.

A run is produced by rule and term matching, by model-assisted evaluation, or by a reviewer’s adjustments on top of a machine run. The run records which.

A candidate score is not shown to a client until it has been approved. The platform checks that an approved score run exists, that the integrity checks pass and that every required criterion has valid evidence before a candidate becomes visible to a client.

A reviewer can mark a requirement as met, not met or not applicable, and the score is recalculated and labelled as human adjusted. Every hiring decision is made by you.

We do not publish bias test results, and we hold no independent audit report. The security page is our own account of how the system works, not a third-party verification.

Candidates have the right to request a human explanation of any automated assessment, or to contest the result. This is stated in section 10 of the Privacy Notice.

To ask, email privacy@taasflow.com. The Privacy Notice says we respond to requests within 30 days.

You can export your candidate records at any time. Workspace access for a one-off package lasts three months; records are deleted after that unless you renew.

The Privacy Notice also sets retention periods for candidate data, including CVs and profiles. See section 8 of the Privacy Notice.

These periods are published policy and are honoured on request — a deletion or anonymisation request is recorded and actioned by the team. They are not yet enforced by a scheduled job, so a record may be held past its stated period until someone asks. Automating this is planned; until it ships, this section says so rather than implying the timers run themselves.

The Privacy Notice lists Google LLC as a sub-processor for the AI models used for CV parsing and role-fit scoring. The models are Google Gemini models, called through TaaSFlow’s managed AI gateway.

The Privacy Notice also states that sub-processors are contractually prohibited from using your personal data for any purpose other than providing services to us.