Every TA leader knows the feeling. The queue won't clear. Resumes pile up. And somewhere in that pile, you're sure, sits a great candidate you'll never call.
We believed that story too. Then we went and read the data. It changed how we think about the problem, and what we decided to build.
The volume is real. This part is not a feeling.
Start with what's measurable.
Applications per open role more than doubled since 2021, from 46 to 95 (HrPanda, 2026). Applications per hire have tripled (Ashby, 2025). Recruiters are handling about 93% more applications than in 2021, while recruiting headcount is down 14% (Gem's 2026 report, built on 165 million applicants and 1.2 million hires).
The math isn't subtle. Arrival rate doubled. Capacity went flat or down. A backlog isn't a sign of a lazy team. It's arithmetic.
Generative AI widened the gap. LinkedIn now sees roughly 11,000 applications a minute, up 45% even as job postings fell. Recruiters report spotting AI-written applications in under 20 seconds. The resume as a signal of genuine effort is collapsing in real time.
The part we get wrong
Here's the story most teams tell themselves. "Our screen is missing great people."
We tried to prove it with data. We couldn't. Not because the data is missing, but because the data structurally cannot show it.
Look at the funnel. Roughly 1 hire per 180 to 200 applicants. Only about 8% of applicants clear the first screen. Industry sources estimate around 15% of a high-volume pile genuinely match the role, and the other 85% are spray-and-pray.
Now the trap. If 15% are qualified but only 3% reach an interview, it looks obvious that we're rejecting qualified people in bulk. That feels like proof we're missing great candidates.
It isn't. A narrow funnel rejects qualified people for two completely different reasons, and they are not the same.
One is scarcity. With 200 applicants, 30 qualified, and one opening, you reject 29 qualified people no matter how good your screen is. That's not a miss. That's selection.
The other is misranking. Advancing a weaker candidate over a stronger one. That, and only that, is a screening failure.
Yield numbers can't tell these two apart. So "we're missing great people" stays unproven, every single time, by construction. It might be true. The funnel data will never show it.
What the data does prove
Three things, cleanly.
The funnel is brutally narrow, so throughput is the binding constraint. Measured, not felt.
Manual screening is inconsistent. Two recruiters score the same resume differently, and so does the same recruiter on a different day. Well documented across decades of selection research.
And consistency, the obvious fix, is double-edged. A consistent screen running a flawed rule rejects ten thousand people the same wrong way. That stopped being a fairness footnote. Workday is facing a class action over its screening AI. The EEOC already settled with iTutorGroup over software that auto-rejected older applicants. The EU AI Act classifies hiring as high risk. And in India, audit studies show low-caste applicants must send about 20% more resumes for the same callback, with the gap widening when the recruiter is male or Hindu. Consistency without a valid, defensible rule just industrializes the bias.
What actually moves the needle
Three shifts.
Fix the pool before the screen. The strongest lever in the data is channel mix. Job-board applicants apply to everything with little regard for fit. Referral and career-page applicants are pre-qualified by their own intent. Same role, very different prevalence of strong candidates. The cheapest quality gain usually sits upstream of the resume review, not inside it.
Make the screen structured, not just faster. The validity research was rewritten in 2022. Sackett and colleagues showed the old estimates were inflated, and that structured interviews, not general cognitive ability, now sit at the top of the validity hierarchy. A structured, consistent, documented screen beats a fast gut call. A fast gut call at scale just produces more mistakes per hour.
Measure the one thing you've been assuming. You cannot get your misranking rate from a benchmark. You can get it from your own rejects. Pull a random sample of rejected resumes. Strip the names. Have a calibrated panel re-score them. Measure how often they disagree with the original decision. That number, the rate at which you reject people your own best reviewers would have advanced, is the only honest measure of whether you're missing talent. Almost no one runs it.
The reframe
The problem was never "our screen misses great people." We can't prove that, and chasing it leads somewhere expensive: building a faster version of a screen we never validated.
We run an inconsistent, undocumented, low-validity screen, at a volume that guarantees backlog, on a pool whose quality we set upstream and never optimized, and we have never once measured whether the screen ranks correctly.
That's a harder problem statement. It's also a true one. And every piece of it is something a team can own and fix.
If you do one thing this quarter, run the reject backtest. Stop assuming the answer. Go measure it.
See your first round on a real role
tringHR scans and screens every applicant to the same standard, with the evidence behind each decision, and your recruiters make the final call. The best way to judge it is on one of your own open roles.
Book a demo →Data sources referenced: Gem 2026 Recruiting Benchmarks Report; CareerPlug 2025 Recruiting Metrics Report; Ashby 2025; HrPanda 2026; LinkedIn application-volume data (2024 to 2026); Sackett, Zhang, Berry and Lievens (2022, Journal of Applied Psychology); Siddique (2010) and Banerjee, Bertrand, Datta and Mullainathan (2009) on caste audit studies; EEOC iTutorGroup settlement; Workday litigation; EU AI Act high-risk classification. Figures from ATS vendors converge on funnel shape; their interpretation and the 15% prevalence estimate are directional.