WHO ANSWERS FOR A HIRING DECISION AN ALGORITHM MADE

Automated screening moves the work to a vendor and leaves the accountability with the employer. Four named roles and one decision record close that gap.

Emmanuel Duncan · 7 September 2026 · 6 min read

A candidate writes to ask why they were not shortlisted. The recruiter checks the system and finds the application was filtered before any person opened it. The recruiter asks the vendor, who explains that the score is produced by a model and points at a support article. The reply that eventually goes back to the candidate says the application was not successful on this occasion.

That exchange is where most organisations discover what they bought. The screening work moved to the software. The obligation to account for the outcome stayed exactly where it was.

This article is about closing that gap before it opens, and it is a governance question rather than a technical one.

The Gap Has a Shape

Three things separate when screening is automated, and they used to sit together.

Who performs the assessment. Who can explain it. Who carries the consequence of it being wrong.

In a manual process all three sit with the recruiter and the hiring manager. They read the application, they can say what they thought, and they answer for the shortlist. Automation moves the first to a vendor, leaves the second nowhere in particular, and leaves the third with the employer, because the employer is the one making the hiring decision in law and in the eyes of the applicant.

Nothing in the vendor contract changes the third. It is worth reading the contract to confirm how firmly it says so.

Four Roles to Name Before Deployment

Governance here is mostly a matter of writing four names down and telling people what they are for.

The decision owner. One named person, senior enough to stop the system, who is accountable for outcomes produced with the tool. Not a committee. This person signs off deployment and signs off continued use after each review.

The explainer. The person who can answer, in plain language and within a working week, why a specific application was scored as it was. If nobody can hold this role, the tool is not deployable for screening, and that conclusion is worth reaching before purchase.

The reviewer. Someone outside the recruitment team who examines outcomes on a schedule and reports to the decision owner. Recruitment reviewing its own selection results is a weak control, because the people who chose the tool are the people assessing it.

The route for challenge. A stated way for an applicant to ask for a human review, and a person who performs it. This is the part organisations leave out most often, and it is the cheapest of the four.

What a Human in the Loop Has to Mean

Almost every vendor will tell you a human remains in the loop. The phrase covers arrangements that differ enormously.

The weak version is a recruiter approving a shortlist the system produced, working from a ranked list, under time pressure, with no visibility of who was filtered out. That is a person confirming a decision rather than making one, and it will not survive being examined.

The stronger version has three features. The reviewer can see the applications that were screened out, at least as a sample. The reviewer has time and authority to disagree, and disagreement is recorded rather than requiring a justification. And the override rate is monitored, because a rate near zero over months means the loop is not doing anything.

That last measure is the most useful single indicator of whether your governance is real.

What to Record, From Day One

If you cannot reconstruct a decision, you cannot defend it. The record needs four things per applicant: the score or outcome, the version of the model that produced it, the date, and whether a person reviewed it and what they concluded.

Model version matters more than it sounds. Vendors update models, sometimes without announcement. An organisation that cannot say which version assessed a given candidate cannot investigate a complaint about a decision made in March.

Ask for this in the contract. Ask specifically for notice of model changes and for the ability to export the decision log.

The Monitoring That Actually Matters

Two things, reviewed quarterly.

Selection rates by group at each stage. Applications received, passed automated screening, interviewed, offered. Where the pattern narrows sharply at the automated stage relative to the others, that is where to look. Whether you can examine this at all depends on collecting the relevant fields at application, which is a decision to take deliberately and to explain to applicants.

Outcomes of the people hired through the tool, compared with those hired before it. Six and twelve month retention, probation outcomes, manager ratings. A screening tool that improves speed and lowers the quality of hires is a cost that will not appear in any recruitment report.

Both of these require the historical data to still exist, which is another argument for the export clause.

The Obligations That Already Apply

In Ghana, the Data Protection Act, 2012 places the responsibilities of a data controller on the organisation collecting and using personal data, and applicant data sits squarely inside that. Processing carried out on your behalf by a vendor does not transfer those responsibilities to the vendor. That includes the duty to process fairly, to be able to say what data is held and why, and to correct or delete it on request.

Organisations recruiting internationally may also be reached by rules in the candidate's jurisdiction, some of which give a person the right to ask for human review of an automated decision and to be told the logic behind it. If you advertise across borders, get advice on which of those apply to you before the first complaint arrives rather than after.

I would treat all of this as a floor. The reputational cost of a screening decision you cannot explain arrives long before any regulator does.

Where to Start

If you already have a tool running, ask for one thing this week. A written answer explaining why one specific rejected applicant was rejected, produced by your own team from your own records, without contacting the vendor.

If it takes more than a working week, or if it cannot be produced at all, you have found the gap, and you have found it in a case where nobody is waiting for the answer. That is the best circumstance in which to find it.

Emmanuel Duncan is the founder of Aletheia Lunara Consult, a professional education and consulting practice working with organisations across Ghana and beyond on people analytics, workforce research and organisational capability.