Why are my candidates still dropping out after adding AI

This is the question I get about six weeks after an agency switches on AI sourcing or screening. The front of the pipeline is fuller than it has ever been, the consultants are submitting faster, and the dropout rate has gone up. Owners assume the new candidates are lower quality, or the market has changed. When I pull up the pipeline with them, it is almost never either of those. The candidates are leaving from exactly the stage they used to leave from, and there are more of them sitting in it.

I have watched enough desks through this to be confident about the pattern, and the fix is smaller than people expect. It is four pieces of plumbing, none of them clever, all of them aimed at the one stretch of the process where a candidate is waiting on you.

Where the dropouts actually happen

Draw the stages a candidate passes through and mark where they leave. Almost nobody drops out during the screening call, because they are talking to you. Almost nobody drops out on interview day. They drop out in the gaps between stages, and the longest gap on every desk is the one after the CV goes to the client: waiting for interview times, waiting for feedback, waiting for a decision. That gap was always where the leak was, long before anyone bought an AI tool.

What AI did was put more people into that gap. More sourced means more submitted, and a hiring manager who saw two people a week before the tool still sees two a week after it. So each candidate waits longer for their turn, and every extra day in that gap is another day for the other agency's offer to arrive. The dropout rate went up because the queue got longer, and the queue got longer because the front got wider.

The second thing AI changed, which is worse

There is a subtler effect I see on desks that added screening. Because every application now carries a score and a summary, consultants trust the top of the list and submit it faster, with less of a conversation first. The candidate goes into the client gap earlier, having spoken to nobody, and a candidate who has never had a real conversation with the agency has no reason to wait for it. They were a CV to you, and you are a job ad to them, and either side can walk away from that without noticing.

So the desks with the worst dropout after AI are usually the ones where the screening call got shorter or disappeared. That is worth saying plainly, because the fix for it is a habit: the conversation before submission is the thing that makes a candidate wait for you, and no automation replaces it.

The four leaks I see, and what closed each one

When I sit with an agency and go through the pipeline stage by stage, the leaks fall into four kinds, and each has a specific fix that we built into Recruitly because we kept seeing the same four.

Silence after they apply. The candidate hears nothing for a week and assumes the answer is no. The fix is that the pipeline talks for you: an auto response when they apply, and an email to the candidate every time they move stage, sent by the automation rule and not by a consultant remembering. The candidate knows where they are, which is most of what they wanted.

The interview that takes a week to book. Three emails to the client for times, two to the candidate, one reschedule, and the good candidate has been to someone else's interview in the meantime. The fix is the booking page: the client's availability comes from their calendar, the candidate picks a slot, both diaries update, and the reminders go out without anyone touching it. On the desks that put this in first, the client gap shrinks by the days that used to be spent on email.

The reply nobody saw. The candidate replied to a sequence email or a CV-share email saying they were interested, or asking a question, and it landed in a mailbox with sixty other replies. The fix is reply intent: replies are read and sorted as they arrive into genuine, not interested, out of office and bounce, so the genuine one is in front of a consultant the same day.

The one who went quiet. A candidate you liked has not replied in ten days. Left to a person, this is found out when the client asks for them. The fix is a no-activity trigger that emails the candidate asking for an update, and a follow-up sequence that starts when a CV goes to a client, so the chasing happens on a clock.

Let the candidate see where they are

The single change that removes the most "any news" emails is giving candidates a place to look. The candidate portal on Recruitly shows a person the roles they are in and the stage each one is at, so the question they would have emailed you is answered before they ask it. I watch the inbound "just checking in" messages fall away on desks that switch it on, and every one of those was a candidate wondering whether to keep waiting.

Put together, the portal and the four fixes mean a candidate in the client gap hears from you on the day they apply, on the day they move stage, when their interview is booked, when they have gone quiet, and any time they look. That is a candidate who knows the agency is still there. The client's diary is the same length, and a candidate who feels held during the wait stays in it.

The check to run this week

None of this needs a project. It needs one look at the pipeline with a stopwatch, and then one automation, and then the next.

1. Measure the days between each stage for the last twenty candidates who dropped out. The longest gap is your leak, and on most desks it is the one after submission.
2. Put the first automation on that gap. If it is the client gap, that is the booking page and the stage-change emails. If it is the post-application gap, that is the auto response.
3. Put the screening call back if it got shorter. The conversation before submission is what makes a candidate wait for you, and the score does not do that job.
4. Call your top three candidates on every live role once a week yourself, whatever the automations are doing. The automations keep the seventy warm, and the three you want most get a human.

The candidate experience on most desks is decided in the gap after the CV goes out, and AI made that gap busier without making it shorter. The agencies whose dropout fell after adding AI are the ones that noticed the leak was where it always was, and pointed the automation at the waiting. A candidate who calls you back is one who was never left wondering whether to.

Written for agency owners whose pipeline got fuller and whose dropout got worse, and who assumed the two were unrelated.