Why AI Recruiting Pilots Speed Up the Step That Wasn't Slow

A lot of recruiting teams have now run an AI pilot, and the results tend to come back in an odd shape. The tools work and recruiters save time. Yet roles take about as long to fill as before, and the share of qualified candidates on the shortlist hardly moves. How does a team get faster at its work without hiring any faster?

I think the answer is that they sped up a step that wasn't slow. Picture a production line where one machine handles ten parts an hour and every other machine handles fifty. You can make the fast machines twice as fast and the line will still turn out ten parts an hour. The only improvement that shows up in output is one made at the slow machine.

Hiring has a slow machine too, and most pilots start somewhere else. The first things teams automate are usually the most visible ones: writing job posts, sourcing candidates on LinkedIn, filtering resumes by keyword. Suppose a recruiter used to spend 45 minutes on a job description and now spends 8. That's a real saving.

But the job description was never what held hiring up. What held it up was everything after the applications came in. By one estimate, a typical corporate role gets 200 to 400 applications in the first 72 hours. Saving 37 minutes on the description does nothing to help you talk to those people. It just gets their applications to the same bottleneck a little sooner.

So where is the slow machine? In most funnels I'd put it in the stretch between an application arriving and the first real conversation with the person who sent it. It matters more than its length suggests, because it's when the candidate is paying the most attention. Every day nobody calls, some of that attention drifts elsewhere.

One 2025 study put the fall in candidate engagement at roughly 40% for each extra week before first contact. I'd hold the exact figure loosely, but the direction matches common sense. If a recruiter's queue finally clears on day eight, some of the strongest applicants will have taken another offer by then, or withdrawn without saying so, or simply stopped answering. Your role didn't get worse in those eight days. Nobody gave them a reason to keep looking at it.

It's tempting to blame recruiters for this, but the numbers don't support that. A skilled recruiter with a normal load can do something like 8 to 12 screening calls a day. At that rate, a campaign that brings in 300 applications in its first week is five or six weeks of nothing but screening calls for one person.

Keyword filtering doesn't rescue you either, and it's getting weaker on its own. Plenty of candidates now polish their resumes with AI, and those resumes pass keyword filters easily. A filter that almost everyone passes isn't doing much sorting. A conversation is a lot harder to polish.

That's why, as far as I can tell, the teams that did see time-to-hire and quality of hire improve mostly got there by automating first contact rather than job descriptions.

It's also why first contact is the step we built Asendia for. It's the one place in the funnel where adding people can't keep pace with applications. When someone applies, Asendia phones them, often that same evening or over the weekend, and talks with them. We chose a spoken conversation over an emailed form on purpose. A form collects whatever the candidate decided to type, and typed answers are as easy to polish with AI as a resume is. On a call, the software works through the screening criteria set for the role, such as relevant experience, location, availability and the questions specific to the job, and when an answer is vague it asks a follow-up, the way a good recruiter would. That back-and-forth is much harder to script in advance, and it tells you more than a text box can.

What goes back into the ATS afterward is a qualification summary, a ranked shortlist, and quotes from the conversation, rather than a resume score. A recruiter who would have started Monday with a few hundred unread applications starts with a short list of people who have already been talked to. Their time goes to the parts that need a person, like assessing candidates more deeply and negotiating offers. Hiring managers see candidates of more even quality, because the shortlist came out of conversations instead of keyword matches.

If you're planning a pilot, I'd start from a different question. Instead of asking where you could add AI, ask where candidates go quiet. In most teams it will be the silence right after they apply.

Then judge the pilot by the numbers that stage controls: time to first contact, and how many of the candidates on your shortlist turn out to be qualified. Those are what move offer acceptance and reduce the cost of hires that don't work out. Minutes saved on job descriptions won't show up in either. There's a longer argument for AI that carries out pipeline steps itself, rather than just assisting the people who do, in our post on agentic recruiting.

Most teams haven't moved their automation to first contact yet. That's a little strange, given where the slow machine is, but it means there's still time to be early.

Ready to transform your hiring strategy? Schedule a Demo with our founders today!