The five metrics that actually predict hiring outcomes
Pick five and run them every month. These are the five we ask every CurriculoATS customer to track because they predict cost, quality, and speed in combination. The order matters: the metrics later in the list are usually symptoms of problems further up. SHRM's 2025 Benchmarking Report sets the national baselines for two of them. Lever and other ATS vendors publish benchmarks for the others. The five:
- Time-to-hire – application to accepted offer.
- Cost-per-hire – fully loaded, including tools and time.
- Screening efficiency – applications per hire and reviewer-hours per shortlist.
- Candidate drop-off rate – by funnel stage, not aggregate.
- Offer acceptance rate – and the time the offer sat unsigned.
Together they form a closed system. A team with strong time-to-hire and weak offer acceptance has a compensation problem, not a sourcing one. A team with strong cost-per-hire but a high drop-off at the technical interview stage has built a fast machine that is hostile to the people inside it. The metrics in isolation are noise. As a set, they are diagnosis.
Time-to-hire and cost-per-hire: the two SHRM benchmarks every founder should know
SHRM's 2025 Benchmarking Report puts the US average cost-per-hire at $5,475 for non-executive roles, with executive hires running roughly 7x that. Average time-to-fill sits at about 41 days, with senior and technical roles routinely past 60. Compare your numbers to those baselines. If you are 30% above either, you are not unusual – you are funding the inefficiency of your toolchain.
Cost-per-hire is where most founders under-count. The honest formula adds:
- External spend: job board fees, sourcing tool subscriptions, agency fees if any.
- Internal time: hours spent by recruiter, hiring manager, panel, founder, prorated to fully loaded comp.
- ATS and assessment tool costs, allocated per hire.
- Onboarding cost in the first 30 days.
For a 50-person startup, fully loaded cost-per-hire often lands between $4,500 and $8,000 – and a chunk of that is hiring-manager time spent re-reading resumes the ATS already "ranked." That is a fixable cost. Two changes typically pull the number down by 30-40% inside a single quarter: switching to an ATS that produces written reasoning per candidate (so reviewers trust the rank instead of redoing it), and consolidating sourcing into one paid channel instead of three. The first removes hidden time cost. The second removes hidden tool cost.
Screening efficiency, drop-off, and offer acceptance: the three founders ignore
The other three metrics are where startups bleed quality. They are also the three that legacy ATS dashboards underweight, because they expose the platform's own limitations.
Screening efficiency is the ratio of applications to interviews and the reviewer-hours required to produce that ratio. A team reviewing 200 applications to surface 8 interviews has a 25:1 ratio. If a recruiter spent 12 hours producing that shortlist, the unit cost per interview is 1.5 hours. With outcome-based AI scoring and written reasoning, the same shortlist takes 30 to 45 minutes of focused review. The ratio drops, the unit cost drops, and quality usually rises – because the human is reading the borderline cases instead of skimming everyone.
Candidate drop-off rate needs to be measured by stage, not in aggregate. A 40% drop-off between application and first recruiter contact is a screening problem. A 30% drop-off between technical interview and onsite is a process or interviewer problem. The aggregate number hides both. Track it stage-by-stage every month.
Offer acceptance rate is the lagging indicator that exposes everything else. An 85% rate is healthy for a startup with strong brand and competitive comp. Below 70%, something upstream is broken – usually compensation alignment or a candidate experience that felt extractive. Pair this with "days from offer extended to signed." An offer that sits seven days unsigned and then declines was a no the whole time.
The two metrics most founders track wrong (and what to do instead)
Two of the five metrics get measured in ways that hide the actual problem. The first is time-to-hire, which most teams measure from "role posted" to "offer accepted." That definition skips the part of the funnel where the worst delays happen – the gap between application and first human response. A more honest measurement starts the clock when the first qualified candidate enters the system. If you posted a role on March 1, the first strong candidate applied on March 3, and you sent the first email on March 12, your real time-to-decision starts on March 3, not March 1. Teams that re-baseline against the candidate-side clock typically discover their numbers are 10-14 days worse than what they were reporting to the board.
The second commonly miscounted metric is cost-per-hire, which many startups report excluding internal time. SHRM's full formula assumes fully loaded comp on every hour the team spends reading resumes, debriefing panels, and writing rejection notes. A founder who spends six hours reading 200 resumes per role and bills their own time at $200/hour has just added $1,200 of true cost-per-hire that the dashboard never showed. Re-running the math with internal time included is the moment most founders decide to fix screening throughput, because the leak is suddenly visible. The fix is not a better dashboard. It is honest definitions applied consistently, and the willingness to look at the number even when it is uncomfortable. McKinsey's research on operational excellence calls this "counting the cost of what you actually do, not the cost of what you wish you did."
How to instrument these metrics without a data team
You do not need a BI stack. You need a spreadsheet with five tabs and a discipline of filling it monthly. Practical setup:
- Choose your source of truth. Your ATS holds application timestamps, stage transitions, and final outcomes. If your current ATS exports those cleanly, use it. If not, that is itself a signal.
- Define each metric the same way every month. "Time-to-hire" is application date to offer-accepted date. Not posting date. Not first-interview date. Pick one and never change it.
- Pull data on the same day each month. First Monday works. Track 90-day rolling averages, not single-month numbers – startup volume is too lumpy for monthly precision.
- Compare to benchmarks, not to your past. Improving from 60 to 50 days when the benchmark is 41 days means you are still slow, just less slow.
- Pick one metric per quarter to fix. Founders who try to fix five at once fix none. Pick the worst-performing one against benchmark and run a single experiment.