FCA authorised · FRN 724952 0117 325 0027Quote & buy →
Aipedia: Recruitment agencies
An agency that ranks candidates with a matching algorithm should be able to show it has tested the tool for bias and accuracy. Matching decides who consultants call first, so a skewed model can quietly exclude protected groups.
Part of: AI in recruitment agencies
In short
Matching engines compare candidate records with vacancies and rank who fits best. They learn from past placements, which can carry past bias into future shortlists. DSIT's Responsible AI in Recruitment guide recommends bias audits, performance testing, model cards and a feedback route at procurement, pilot and live use. Agencies are employment service-providers under the Equality Act 2010, so a model that disadvantages a protected group can lead to an indirect discrimination claim. Documenting your audits helps defend claims and answer insurers' questions at renewal.
Matching tools sit inside the agency's database or applicant tracking system. When a vacancy arrives, the engine scores every candidate and presents a ranked list. Some learn from which candidates consultants contacted or placed. Others use language models to compare CV text with job descriptions. Consultants may not scroll far down the list, so the ranking effectively decides who is considered.
DSIT's Responsible AI in Recruitment guide (25 March 2024), written with contributions from the REC and APSCo, describes assurance at three stages: procurement, pilot and live use. Its tools include a DPIA, a bias audit, performance testing, model cards, transparency and a feedback route. The REC AI guidance advises auditing for bias and checking that criteria do not disadvantage protected groups, and APSCo Global recommends risk assessment of AI tools and a board-level AI lead.
The ICO's 2024 audit found tools that filtered by protected characteristic or inferred ethnicity or gender from names. If a ranking leads directly to a rejection without meaningful human involvement, the Article 22C safeguards apply.
A matching model does not need to see a protected characteristic to discriminate. Proxies such as postcode, career gaps, graduation year, names of schools or hobbies can correlate with race, age, sex or disability. Under section 55 of the Equality Act 2010, an employment service-provider must not discriminate in its arrangements for selecting who receives its services. REC Code principle 4 requires respect for diversity. The Manjang v Uber Eats case shows automated systems can be challenged under equality law, even where the claim settles.
Illustrative: an agency's matching engine learns from five years of placements in engineering roles. Because past placements were mostly men under 40, the model ranks similar candidates higher. A woman returning after a career break notices she is never put forward despite strong qualifications. A subject access request reveals her consistently low match scores. She brings an indirect discrimination claim and the agency cannot show any bias testing.
Discrimination claims by candidates may be excluded from PI or limited, and the agency's employment practices cover usually relates to its own staff. A client's claim that a poorly performing matching service caused it loss may fall within PI, subject to the wording. Some insurers may ask about AI use at renewal; a documented audit trail shows the risk is managed. See AI bias and discrimination claims and AI exclusions and clauses.
An audit only helps if it is written down and kept. A short AI register for each matching tool might record:
This is the evidence a claimant's lawyer, the ICO or an underwriter would ask for, and it also shows the board where the agency's AI risk sits.
Apex arranges PI for recruitment agencies; a broker can help you present your AI governance to insurers. See the AI in recruitment agencies hub, AI CV screening, AI candidate sourcing and does recruiters' PI cover AI mistakes?.
No law names a bias audit as mandatory, but DSIT's guide recommends one, the REC advises auditing for bias, and the Equality Act prohibits indirect discrimination. An audit is the clearest evidence that the agency took reasonable steps.
Yes. Proxies such as postcode, graduation year or career gaps can correlate with protected characteristics, so a model can disadvantage a group without seeing the characteristic itself. That is why outcome testing matters.
Some insurers may ask how firms use AI. Evidence of procurement checks, bias audits and performance testing shows the risk is managed and can help at renewal. It also helps defend a claim.
The agency uses the tool to provide its service, so it faces candidates' claims and client complaints. It may have contractual rights against the supplier, depending on the terms. DSIT and APSCo both encourage holding suppliers to account through procurement checks and contracts.
Tell us about your firm and the work you do. A broker will review how your current PI treats AI-assisted work, or arrange new cover. No fee for a quote. Or call 0117 325 0027.
Start a PI quote Request a callbackApex Insurance Brokers Limited is authorised and regulated by the Financial Conduct Authority. Registered in England and Wales, company number 07014570. This page is general information about how AI is used in the profession and how insurance may respond. It is not legal, regulatory or insurance advice on your circumstances, and it does not guarantee that cover will be available or on what terms.
