How AI changes a recruitment desk

Almost everything written about AI and recruitment is about whether the job survives. It is the wrong question and it has produced two years of useless content, because the answer was never in doubt to anybody who has actually sat on a desk.

A recruiter's value has never been in collecting CVs, typing them into a system, chasing a hiring manager for feedback, or remembering that a candidate on a three-month notice period needs a call in week eleven. That is the part everybody spent their time on, and it is not the part anybody was paying for. It is overhead that grew around the work, and it is exactly the part a machine handles well.

So the honest description of what has happened is not replacement. It is that the ceiling moved. The number of live jobs one consultant could hold was never set by their ability to recruit. It was set by how much admin each job dragged behind it. Take the admin away and the same person holds ten times as many jobs, which is the argument we made in full in a 10x recruiter runs a hundred jobs.

This piece is about what that actually looks like on a Tuesday.

What AI actually took off the desk

Four kinds of work, and it is worth being precise about them, because the vendors are not.

Collecting. Finding people who could do the job, assembling them into a list, pulling their details together from four places. This was never skilled work; it was time-consuming work that only a person could do because only a person could read a CV. That stopped being true.

Typing. Getting what you already know into the system. Parsing the CV, filling the job fields, writing up the call you just had, logging the activity so the pipeline is accurate. Every hour of this was an hour where an experienced consultant was doing data entry, and every consultant under pressure skipped it, which is why most agency databases are a partial record of what actually happened.

Chasing. The client who has not come back on three CVs. The candidate who has gone quiet. The reference that has not landed. The timesheet that is not in. This is not thinking work, it is remembering work, and a machine remembers perfectly and without resentment.

Sorting. Opening ninety replies to find the four that said yes. Reading a stack of applications to find the six worth a call. Working out which of today's forty notifications matter.

Notice what is not on that list. Deciding whether someone is right for a business rather than merely right for the job spec. Telling a client something they do not want to hear. Talking a good candidate out of a counter-offer they will regret. Knowing that the role as written is not the role they actually need. None of that has moved an inch, and none of it is close to moving.

Monday morning, before and after

This is the clearest way to see it, because the change is not abstract. It is in the first ninety minutes of the week.

The old Monday. Open the inbox to somewhere between sixty and two hundred messages accumulated since Friday. Spend the first hour triaging: which of these are candidates, which are clients, which are job boards, which are noise. Find the four replies that matter buried among the automated ones. Then open the CRM and start reconstructing last week, because Friday afternoon's calls never got written up. Then look at the pipeline and try to remember which of these candidates have gone cold. By eleven o'clock you have not spoken to anybody.

The new Monday. The replies were sorted as they arrived, so the interested ones are already at the top and the out-of-offices and bounces are already filed. Friday's calls are already written up on the records, because the call notes were produced from the call itself rather than from memory. The candidates who have gone quiet are flagged, with how long it has been. The jobs that came in as emails already have their fields filled from your own lists. Your first act on a Monday is a decision about who to phone, not an hour of reconstruction.

The activity map

Laid out properly, because the shape of the change is easier to argue with when it is vague.

Activity	Then	Now
Building a longlist	Hours per job	Minutes, and you edit rather than assemble
Parsing a CV into the system	Manual, so often skipped	Done on arrival, checked by you
Writing up a call	From memory, at six o'clock	Written from the call, on the record
Sorting replies	An hour of triage a day	Sorted as they land
Chasing feedback and paperwork	Whoever remembered	Prompted, every time, without resentment
Deciding who to put forward	Yours	Yours, and you have more time for it
The client conversation	Yours	Yours, and it is now most of the job
Holding a candidate through an offer	Yours	Yours, and it matters more than ever

Read the right-hand column downwards and the pattern is unmistakable. Everything a machine took was something you did to keep the system fed. Everything left is something a person does to another person. That is not a diminished job. That is the job, finally without the overhead.

What is on the desk that did not exist two years ago

This is the part nobody writes about, and it is where the good consultants are pulling away from the rest.

Supervising the machine. A handful of cases a day come back because the software was not sure. A reply that could be interest or could be a polite no. A name that matches two people. These are the genuinely ambiguous cases, and handling them well is a real skill that did not exist as a distinct activity before. It is also the reason software that quietly guesses is so damaging: it removes the decision instead of routing it to you, which we went into in why the AI in your ATS gets things wrong.

Curating what the machine knows. Your seniority levels, your pipeline stages, your reasons for rejection, the way your agency describes a good fit. The software fills fields from those lists, so the lists are now a piece of infrastructure. An hour spent getting them right pays out on every job afterwards.

Running a portfolio rather than a queue. At ten live jobs you could hold all of them in your head. At a hundred you cannot, and the skill becomes deciding where your attention goes this morning. Which three jobs are at a moment where a phone call changes the outcome. That is a genuinely different discipline, closer to how a good account director thinks, and it is what the consultants placing three or four times the person at the next desk are actually doing.

Designing the questions. When a machine screens, what it screens for is something somebody chose. Getting that right for your market is high-value work that lands on the senior person in the room, and getting it wrong quietly wastes a quarter.

Why ten job orders becomes a hundred

The number I hear at onboarding is ten to fifteen live jobs. The number I watch good desks reach a few months later is far beyond that, and the reason is arithmetic rather than heroics.

A job does not take a fixed amount of recruiting. It takes a small amount of recruiting and a large amount of overhead, and the overhead is per job. Ten jobs meant ten sets of longlists, ten sets of write-ups, ten chase cycles, ten mental threads to keep warm. The recruiting was never the constraint. The overhead was, and it scaled linearly with every job you took on.

This is also why a recruiter is worth more now, not less. A consultant whose judgement used to be applied to ten situations a week now applies the same judgement to a hundred. The scarce thing was never the hours. It was the person, and you have just been given far more of them to spend on the part only they can do.

The part that never changes

I want to be straight about the limit, because a piece that pretends there isn't one is useless to you.

Your client still interviews two or three people a week. That number is set by a hiring manager's diary, not by anybody's software, and it has not moved in twenty years. Everything upstream of it can get faster and the number of placements will not move, because the placements come out of that meeting. We made this case at length in the pipe doesn't care where you widened it, and it is the single most important thing to understand before you spend a pound on AI.

The other permanent part is everything that happens in an offer. The counter-offer conversation. The candidate whose partner does not want to move. The client who needs telling that their salary band is two years out of date. Judgement under relationship pressure, where the right answer depends on knowing a person. No model is close, and more importantly, no client wants it to be. They are paying for someone to have been in the room.

The trap, and how to avoid it

Here is where agencies get this wrong, and it is worth more to you than anything else on this page.

You get the hours back and you spend them on more sourcing. It is the obvious move, it feels productive, every tool you bought encourages it, and it makes things worse. More candidates into a pipe whose exit is a hiring manager with two slots a week produces a longer queue, and a longer queue means more good people waiting, getting colder, and taking another offer while they sit in it. That is the mechanism behind candidates dropping out after you added AI, and it is a self-inflicted wound.

Spend them at the squeeze instead. More time on the client, so two interview slots become four. More time qualifying the job properly at the start, so the ones you run are winnable. More time on the candidates already in process, so nobody is lost to silence. More time on business development, because a hundred live jobs across twelve clients is a better business than a hundred across three. The order to attack it in is the subject of which recruitment tasks to automate first.

The agencies who got this right are visibly pulling ahead this year. The ones who pointed all their new capacity at the front of the pipe are busier than ever and placing the same number.

What this makes a recruiter worth

More, and I would argue considerably more.

Every part of the job that a machine can do is a part nobody was ever going to pay a premium for. What is left is a person who can read a situation, tell a client the truth, hold a candidate through a wobble, and decide which of a hundred live jobs deserves the next hour. Those have always been the valuable things, and they were being diluted by six hours a day of admin that nobody chose.

The recruiters I would back over the next decade are not the ones fighting the tooling or waiting for it to go away. They are the ones who let it take the overhead, took on five times the portfolio, and spent the difference on the part of the job that was always theirs. If I did not have a conflict of interest, I would be investing in agencies right now rather than writing about them.

What to do on Monday

Three things, in order.

Find out where the admin actually is. Sit with your best consultant for two hours and count the times they type something the system already knows, or go looking for something it should have surfaced. That list is your roadmap and it will be specific to your desk.

Fix the list before the tool. Your seniority levels, your stages, your rejection reasons. Everything downstream fills from those, and an hour on them pays out on every job for a year.

Decide where the reclaimed hours go before you have them. Write it down, because otherwise they will silently go into more sourcing. The answer is almost always the client relationship and the candidates already in process.

We built our AI, Simi and the agents around exactly this idea: the machine takes the admin, the recruiter keeps the judgement, and anything the software is not certain about comes back to a person rather than being quietly guessed. That is why I say Recruitly is the best recruiting CRM in the world, and it is why the desks running on it are holding portfolios their owners would not have believed three years ago.

AI is not coming for your job. It has come for the six hours a day that were stopping you doing it.

---

Gowri is the CEO of Recruitly.