Cutting admin on a recruitment desk without losing the judgment

Where does a recruiter's week actually go?

A recruiter's week splits into judgment and typing, and on a calendar the two look identical.

Two hours reading applications against a brief and two hours copying the same candidate details into a third system both appear in the diary as a four-hour block called admin. They are not the same work. One is the job. The other is the tax you pay to do the job, and it is the reason the first is done at eleven at night on the roles that matter most.

The typing is measurable and it is bigger than most desks think. Totaljobs surveyed 748 HR leaders and found recruiters spend an average of 17.7 hours of administrative work per vacancy: 3.6 hours reviewing applications, 2.5 hours scheduling interviews and another three hours processing post-interview notes. In the same research, 61% named manual data entry as a common issue delaying recruitment decisions. That is British data, published on 19 August 2025, and we could not find an Irish equivalent, so take it as an order of magnitude rather than a fact about your desk.

The judgment does not appear in that list at all, because nobody measures it. It is the call where a candidate tells you the real reason they are leaving, which is never the reason on the form. It is knowing that a client says they want ten years and will hire the person with six who interviews like they want the job. It is the conversation at offer stage that saves a placement that was about to fall over on a counter-offer nobody saw coming. None of that automates, and all of it is what a client is actually paying for.

Here is the test worth running before you talk to anyone about software. Take last week's calendar and mark every hour J or T. Judgment or typing. Most recruiters are surprised twice: by how much of the week is T, and by how much of the J happened outside working hours because the T ran long.

What can be automated without touching candidate judgment?

The T. All of it, in principle, and five parts of it in practice.

What they have in common is that each takes a job with a right answer and stops a person doing it by hand. Reading every application against the brief, so the hundredth gets the same attention as the first. Re-reading the whole database against a new role, which is where the placements hiding in your own file live. Turning a job spec into competencies, questions and a scoring sheet in the agency's own format. Capturing activity into the record instead of a Friday afternoon of catching up. Pre-filling the contract, chasing the signature and tracking what expires.

One thing we do not build, and it is worth saying on a public page: cold scraping and volume outreach to strangers. It is the workflow most likely to put a small agency on the wrong side of a data protection question, and the honest version of sourcing starts with the people who already consented to be on your file.

What stays with the recruiter?

Every hiring decision, and this is not a hedge, it is the design.

The system reads, ranks and explains. You decide who gets called, who gets rejected, and who gets pushed at the client despite where they sit on a list. We do not build automatic rejection, and nobody disappears from your view: a candidate who fails a criterion is flagged as failing it and stays on the list, because the most common way these systems go wrong is by quietly deleting people.

The brief stays yours too. You write the criteria and the weighting before the role goes live. A system that infers its own criteria from your past placements will reproduce your past placements, including the parts you would not defend in a room.

The law points the same way. Annex III of the EU AI Act classes as high risk "AI systems intended to be used for the recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates". That does not prohibit them. It attaches obligations, and the one that shapes how we build is Article 26(2): deployers "shall assign human oversight to natural persons who have the necessary competence, training and authority, as well as the necessary support". Oversight has to be real. A person who has neither the time nor the authority to overturn a ranking is not oversight, they are a signature. The CV screening page sets out what that means on a live desk, alongside GDPR and the Employment Equality Acts.

How do you keep the desk consistent when the process is automated?

By writing the standard down first, which is the part most desks skip.

Consistency is not a by-product of software. It is what you get when the brief has a structure, the knockout criteria are in plain words, the scorecard is the same one every interviewer uses, and the reason a candidate was rejected is recorded next to the decision rather than remembered. Automating an inconsistent process gets you the same inconsistency, faster and in writing. So the first session is not about tools. It is about what good looks like on your desk, in your words, and it usually surfaces two or three disagreements between people who thought they ran the same process.

The record matters for a second reason, and agencies feel this one directly. The Workplace Relations Commission sets out that an employer must notify each new employee, in writing, within five days of commencement of employment, of eleven core terms, with a fuller written statement inside a month. For agency workers, the WRC states that "the party who pays the wages is the employer for the purposes of this Act and is responsible for providing the written statement". If you pay the worker, that clock is yours. A five-day statutory clock is not something to leave until somebody gets to it on Friday, which is exactly the argument for automating the paperwork and not the judgment.

What does a pilot look like on one desk?

One desk, one workflow, two to four weeks.

We start with a session on how the desk actually runs, not how the process document says it runs, and we write down where the judgment sits. Then the build is tested against roles you have already filled, so you can see whether it would have surfaced the person you actually hired. That is the only honest test available before it goes live. Then a parallel run on two or three real vacancies: you work as normal, the build works alongside, and you compare.

Then handover. The people are trained, ownership is documented, exception handling is written down, and the result is compared against the baseline agreed before anything was built. On a screening pilot that baseline is usually four numbers: days from advert close to shortlist, how many applications somebody read to the end, the proportion of your shortlist the client takes to interview, and placements from candidates who ranked outside the first twenty. Take three months of history on those before we start. If they do not move, the build did not work, and we would rather find that out on your desk than argue about it.

Then we leave. No retainer, you own what was built, and you can change the criteria without calling us.

What does it cost to find out?

The first conversation is a conversation, not an engagement.

It runs about an hour, and it starts with your last five vacancies rather than with software: how long each took from advert close to shortlist, how many applications were read to the end, and how many hours of the week went into typing rather than judgment. At the end of it you get a straight answer on which of the five is worth doing first on your desk, what it would take, and whether it is worth doing at all. Sometimes it is not.

Build cost is quoted per workflow after that conversation, against a scope you have read. There is no retainer, no licence and no per-seat fee, because you own what is built. If your desk is accounting rather than recruitment, the same argument runs over on the accounting and finance side, and the people behind both are on the about page.