Which Recruitment KPIs Still Mean Something After AI Screening

A metric is only useful as long as it moves when the thing you care about moves. Most of the numbers on a recruiting dashboard were never measuring hiring quality directly. They were proxies, and they were good proxies for a particular kind of process: one where a person read every resume and every screen was a 30-minute phone call. If you've put AI into the early part of your funnel, that process is gone, and I think it's worth asking which of the old proxies still work.

Take the three most common: time-to-hire, cost-per-hire and applicants-per-opening.

Time-to-hire became the main recruiting KPI when the constraint was recruiter time. A role sat open because someone had to read, say, 200 applications, schedule a dozen phone screens, coordinate a few panel interviews and chase feedback from busy hiring managers. Shorten any of those steps and the role closed sooner, so the number was a fair stand-in for how efficiently the team worked.

Cost-per-hire followed the same logic. The big costs were advertising, agency fees and recruiter hours. If you paid a search firm $25,000 to fill a senior role, you knew where the money went and what it bought. And applicants-per-opening was a rough check on pipeline health. If only nine people applied, something was probably wrong with your sourcing or your job post, and a thin top of the funnel meant a thin bottom.

All of that holds together when people do most of the work. Once AI is doing most of the early screening, each of these numbers starts measuring something else.

What happens to the old numbers

Suppose an AI voice interviewer screens 400 candidates over a weekend and your time-to-hire drops from 32 days to 19. That's worth something, but look at why it dropped. You removed a scheduling bottleneck. You didn't get any better at judging people, and the new number says nothing about whether the five people you hired will still be there in six months.

Cost-per-hire gets noisy in a different way. Many AI recruiting tools charge a flat license whether they screen 50 candidates a month or 500, so your per-hire cost swings with volume. Comparing this quarter to last quarter means comparing two different cost structures, and the number can look bad in a busy month when the tool is doing exactly what you bought it for.

Applicants-per-opening is the strangest case. A well-calibrated screener should pass fewer people forward, only the ones who really fit. If you judge pipeline health by how many candidates are in it, a screener that's working will look like it's shrinking your pipeline. You'd conclude the tool is failing at the moment it's succeeding.

What these three have in common is that they tell you how fast the funnel moves and how much it costs to run. None of them tells you whether the funnel is making good decisions. That was always true, but when people did the screening there wasn't much you could measure about decision quality at scale anyway. With AI doing the first conversation, you can.

What to track instead

I'd start with four numbers.

The first is screen-to-shortlist rate. Of every hundred candidates who go through the AI screen, how many reach a human? If it's 40, either the posting is attracting a lot of poor fits or the screening criteria are too loose. If it's 4, you may be screening out people who would have been good hires. The right range depends on the role and the volume. What matters is that you now have a number you can tune against.

Then look at what happens after that. Of the candidates who reach a human interviewer, how many get offers? I think this is the clearest check on whether the AI screen is selecting for what hiring managers actually care about. If only one in twenty gets an offer, the screen and the people making the final decision are working from different ideas of what qualified means. That's a calibration problem, and you can't fix it until you can see it.

Quality of hire at 90 days is the most honest number in recruiting, and the most work to get, since it means connecting your ATS to your HR system. Did the person clear their 90-day milestone? Were they still there at six months? This is the number AI recruiting tools should be judged on over time, more than how much they compress the top of the funnel. A faster process that hires people who leave just gets you to the next vacancy sooner.

Finally, track speed separately by role type. AI doesn't perform the same across every position. For a high-volume customer service role, a screener might move someone from application to shortlist in under two hours. For a senior engineering role, the same tool needs more calibration. If you average the two together you get a number that hides both what's working and what needs tuning.

We built Asendia with this kind of measurement in mind. From the candidate's side it's a phone call soon after they apply: an AI interviewer asks about the role, listens, and follows up on what they said, and the fortieth applicant of the day gets the same core questions as the first. Every one of those conversations ends in structured output: qualification notes, red flags, a ranking and a short summary of the candidate. That output goes into the ATS, which is where screen-to-shortlist and interview-to-offer already live. If the screening data lands there in a form you can query, you don't need a new analytics tool; the system your team works in just starts having the data that makes those numbers meaningful. My guess is that the real bottleneck in most agencies is less the speed of screening than knowing quickly enough whether to move a candidate forward, and good structured data from every conversation is what answers that.

If you're working out where AI fits in your recruiting more broadly, we've written about what AI hiring automation changes in high-volume recruiting and about why every team needs an AI recruiter agent for sourcing. Both cover parts of the funnel that feed into these numbers.

You don't need all four to start. Pick two, write down where they are today, and look again in 90 days. That will tell you more about whether your AI hiring setup is working than any drop in time-to-hire.

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