The pipe doesn't care where you widened it

Picture a hosepipe running from a tap to a bucket. Squeeze it anywhere along its length and the bucket fills slower. Now widen it anywhere except the squeeze. The bucket fills at exactly the rate it did before, and the only thing that has changed is the pressure building up behind the narrow bit.

Recruitment is that hose. It runs from a pool of people to a signed contract, and for the past two years the entire industry, candidates and agencies alike, has been busy widening the hose everywhere except the place it is squeezed. This post is about what that pressure does, where it goes, and what it looks like to plumb the thing properly instead.

Where the tap is set

An agency pipe has five sections: source, screen, qualify, client interview, offer. They have never been the same width. A consultant can source a hundred people in a morning, screen fifty in a day and qualify ten by phone in an afternoon. A hiring manager will see two, maybe three, in a week, and that number has not moved in twenty years because it is set by her diary, not by her software.

Keep that picture in your head, because every decision in the rest of this post is a decision about where on this pipe to spend money.

The first thing candidates did

Candidates got there first. A person with a laptop and a subscription can now apply for a thousand jobs in an afternoon, each with a CV written to the job ad, each with a cover letter that reads like it was drafted by someone who cares. The front of their pipe went from a garden hose to a fire hose overnight.

Then the replies start coming back. Screening calls, forms, assessment links, recruiters asking for a chat. A thousand applications produces a few hundred of those, and this is where it falls apart, because answering a recruiter's call is the one part of the job a candidate cannot hand to a machine. The calls that matter, the three companies they would actually take, arrive in the same stream as the hundreds they would not, and they cannot tell which is which until they have picked up.

The candidate did not get more offers. They got a fuller inbox and a worse hit rate on the calls they answered, and the good ones went to whoever picked up first.

Then recruiters did the same thing

Agencies looked at that fire hose pointed at their inbox and responded with a fire hose of their own. AI sourcing, AI outreach, AI screening. The front of the recruiter's pipe is now as wide as the candidate's, and the two are pointed at each other.

Something quieter happened at the same time, and it matters more than the volume. Every CV now looks the same. When every application is immaculate, the CV stops telling you anything about the person, because the thing that used to separate a strong candidate from a weak one on paper was effort and clarity, and both are now free. The document that recruiters spent decades learning to read in six seconds has been wiped clean.

Widening the front of the pipe did not just add volume. It moved the filtering work from a document a recruiter could skim to a conversation a recruiter has to have. The narrowest human section in the agency just got more to do, not less.

So they put AI on the phone

The obvious next move, and plenty of agencies have made it, is to automate the call as well. Voice agents ring the candidate, run through the questions, score the answers, mark them qualified. On paper the recruiter's narrowing just disappeared.

Three things happen instead. The candidate's AI picks up, or the candidate reads from one, so scripted questions meet scripted answers and the call stops carrying signal the same way the CV did. The agent qualifies on what it can check, notice period, salary, right to work, and not on whether the person is any good, so a large share of the noise gets stamped qualified. And none of it touches the client, who still has two slots a week. The pile that used to sit in front of the recruiter now sits in front of the hiring manager, and she does the real sorting in the interview room.

Put the three pictures side by side and the pattern is the same each time. Every place AI is added, the unknown turns into a confident colour, and the work of finding out lands on whoever is next. The client is the last human in the pipe, so that is where it ends up. That is the decision fatigue everyone is complaining about, and it was manufactured upstream.

What pressure actually does

A hose under pressure does not just back up. It leaks. Here is a week that will be familiar to anyone who has run a desk. A consultant sources three hundred people for a finance role, phone screens sixty over five days, and sends eight to the client. The client is travelling and sees four of them in the second week. By then the two strongest have accepted elsewhere, one has stopped replying, and one has told his mates the agency is slow.

There is a piece of queueing maths behind this that is worth stating plainly, because it is the whole argument in one line. The number of people sitting in your pipe equals the rate they come out multiplied by the time each one spends inside. The rate they come out is fixed by the client. So every extra person you push in at the front does not raise the rate out, it raises the time each one waits, and waiting time is the single best predictor of a candidate going somewhere else.

The screen that doesn't screen

A lot of what is sold as AI screening is not screening at all. A filter is only a filter if things stop at it. Most tools take everything in purple, attach a score, and pass everything through in amber. The count going in and the count coming out are the same, which means the section after it is carrying exactly as much as before, now with a number next to each item that somebody downstream feels obliged to read.

The second version is harder to build and harder to sell, because it involves the AI telling people no, and telling them the same day, and being right often enough that the agency is comfortable with it doing so. That is the version that reduces load. The first version only redistributes it.

The section that shrinks when you push

This is where rushing fires back, and it is the part that should worry agency owners most. Push unreviewed volume at the client and the client section does not stay the same width. It gets narrower.

An agency turns on auto-submit. The hiring manager gets six CVs on Monday morning, three of them clearly wrong for the role. On Tuesday she reads the next batch more slowly. By Thursday she has stopped opening the emails, and the agency's submissions now go into a folder that gets looked at on Friday afternoon, if at all. The narrowest section in the whole pipe is made of trust, and the agency just made it tighter with its own hands, in order to save a recruiter twenty minutes.

That is the whole mechanism in one picture. Speed bought in one section is paid for in the next, and the section that pays is the one you can least afford to damage.

Fix one: a valve, not a bigger pump

The first fix is not at the front of the pipe at all. It is a valve at the intake, governed by what the far end can absorb this week. If the hiring manager has three slots and two decisions still pending, the intake opens to match that and no wider. Source to what can be decided, not to what can be found.

This feels wrong to anyone who grew up on activity targets, because it means deliberately sourcing fewer people. It produces the same number of placements with shorter waits, fewer dropouts, and a recruiter whose week is spent on the ten people who matter instead of triaging the hundred who do not. The AI is doing plenty of work here. It is just doing it in the right direction.

Fix two: branches for the traffic

The second fix is where AI earns its keep. Most of what clogs a recruiter's day is not judgement, it is traffic. Scheduling, chasing, reminders, what's the parking like, can you resend the job spec, right to work checks, reference calls. That traffic currently shares a pipe with the one thing only a human can do, which is decide whether this person is right for this client and hold the relationship on both sides.

Split it. AI carries the volume on its own branches, where speed is the whole point and nobody's judgement is being replaced. The human branch carries only what needs a human. And the client branch gets widened the only way it can be, with pre-booked interview slots, structured feedback that comes back the same day, and decisions that get made instead of drifting, rather than with more CVs.

What happens if the AI really does match better

The strongest objection to all of this is the one every vendor makes: if AI matches candidates to roles better than a consultant does, two interviews a week convert to offers more often, and placements rise without the client changing a thing. It is worth taking seriously, because it is the only version of widening the front that could work.

Two things have to be true for it to hold. The first is that the AI can actually see quality, and the earlier sections explain why that is harder than it was, since the CV has stopped carrying signal and the scripted call has gone the same way, so a score is mostly measuring how well the application was written. The second is that the improvement shows up as fewer wasted interviews, and if it does, the right response is to send the client fewer, better people, which is the valve in fix one. Better matching, where it exists, is an argument for sourcing to what the client can decide, and it lands on the same side as everything above.

The number that settles it is interview-to-offer ratio, per client, before and after the tool went in. If matching improved, that ratio rises. I have not seen a vendor publish it.

---

What to do before you buy anything

None of this is an argument against AI in recruitment. It is an argument against putting it in one place and calling the job done. Before the next demo, do four things.

1. Draw your own pipe, with real numbers. How many sourced, screened, submitted, interviewed and offered per role, and how many days each one sat waiting between stages. The narrowest section will be obvious, and it will almost certainly be a client's diary.
2. For any tool you are considering, ask one question: does it remove work from the pipe, or move it? A screen that says no and sends the rejection the same day removes work. A screen that scores everything and passes it on moves it, and it moves it towards the person with the least time.
3. Wire the signal backwards. Let the client's real capacity this week set how much you source, and stop measuring consultants on activity that only raises pressure.
4. Put AI on the branches first. Scheduling, chasing, questions, admin, references. That is where the volume is, that is where speed is pure upside, and it takes those jobs out of the pipe that your judgement has to travel through.

The hose does not care where you widened it. It only cares where it is squeezed, and in recruitment that has always been a human being making a decision. Every AI decision you make is a decision about how much pressure to put on that person. Make it deliberately.

Written for agency owners who have been sold a bigger pump and are wondering why the bucket is not filling any faster.