Why Hiring Feels Broken

Part 1 of this Convergence followed institutional knowledge out the door as experienced workers retire and leave faster than organizations can replace them. The obvious next question is why replacement itself has become so hard. Talk to employers and you'll hear about a persistent skills shortage — roles sitting open for months, a talent pipeline that can't keep pace. Talk to job seekers and you'll hear the opposite complaint: hundreds of applications, most met with silence, and a growing sense that no human ever sees a resume before it's filtered out. Both descriptions are backed by real data. That is the puzzle this investigation follows — not "AI broke hiring," which a companion investigation in our first Convergence already examined, but something broader and stranger: a hiring market whose own sorting mechanisms — job postings, applicant tracking software, degree filters — appear to be failing both sides of the transaction at once.

The clearest finding to emerge from the evidence is not a technology story. It's a rhetoric-versus-practice gap: employers overwhelmingly say they've modernized hiring — skills over degrees, faster processes, wider nets — while the measurable data on job postings, degree requirements, and actual hiring outcomes shows the opposite happening in practice. That gap, more than any single villain, is what this investigation traces.

Hiring used to be a bottleneck of human attention: a hiring manager could only physically read so many resumes, so postings drew dozens or low hundreds of applicants and a person looked at most of them. Applicant tracking systems emerged in the 1990s and 2000s to solve exactly that bottleneck — index resumes, rank them by keyword match, route the best matches to a recruiter's queue. By 2019, roughly 99 percent of Fortune 500 companies used some form of this software [5]. The technology did what it was built to do: it let one recruiter process far more applications than a human ever could unassisted.

What it didn't do is scale the human judgment on the other end. The number of applications a single posting can attract has grown by orders of magnitude — online job boards, one-click applications, and aggregator sites mean a single opening can draw hundreds of submissions within days. The infrastructure for filtering grew to match that volume. The infrastructure for actually looking at a filtered-out candidate a second time did not.

Start with postings that don't lead anywhere. Federal JOLTS data shows the gap between job openings and actual hires running at 28–38 percent a month since 2021, versus under 10 percent through the 2010s [1]. A March 2025 survey of 918 HR professionals found 93 percent admit to posting listings they have no intention of filling at least occasionally — most commonly to keep a pipeline of pre-screened resumes on file for a role that might open later [2]. Not all of that gap is deception; some reflects ordinary recruiting lag. But the scale and persistence of the gap is new, and documented.

Second, application volume has exploded on the applicant side. LinkedIn recorded a 45.5 percent jump in applications in a single quarter even as postings fell 10.6 percent, and cites an average of roughly 57 applicants per hire on its platform [3]. Recruiters report piles of 300 or more resumes for a single role.

Third — and this is the sharpest finding — a joint Harvard Business School and Burning Glass Institute research program found that while about 85 percent of employers publicly claim to prioritize skills-based hiring, only roughly 0.14 percent of actual hires — fewer than 1 in 700 — trace to a company actually removing a degree requirement and hiring someone without one. Forty-five percent of self-described "skills-based" employers had changed their job postings without changing who they actually hired [4]. Consistent with that, after bottoming at 18.5 percent in March 2024, the share of US postings requiring a bachelor's degree climbed back to 19.3 percent by November 2025 — reversing the trend the "skills-first" movement was supposed to represent [5].

Fourth, on the filtering mechanics themselves: a two-year Harvard Business School and Accenture study of "hidden workers" found 88 percent of surveyed employers admitted that qualified, high-skilled candidates were rejected outright for not exactly matching stated criteria — including automated filters, used by roughly half of employers, that exclude anyone with an employment gap over six months. The study estimates 27 million such workers in the US are willing and able to work but consistently filtered out before a human ever sees them [6].

Two things the numbers above don't settle are worth naming directly. One widely repeated figure is unsupported: the claim that "75 percent of resumes are rejected by ATS software before a human sees them" traces to an unverified, long-defunct startup source from around 2013 and does not hold up under scrutiny. Established evidence A 2025 audit of 25 recruiters across more than 10 ATS platforms found most systems rank and route candidates by keyword match rather than silently auto-rejecting them — though the same audit found the median resume scores only 48 out of 100 on keyword match, and roughly half never clear a typical passing threshold [7]. The mechanism is real; the specific viral statistic describing it is not, and this piece treats that distinction as itself part of the evidence.

Competing hypothesis Whether a "skills shortage" is real in the aggregate is a genuinely contested claim. SHRM's 2025 research found that about a third of US job openings can't be filled by unemployed workers whose most recent job was in the same occupational category — a genuine, measurable mismatch [8]. But the Economic Policy Institute has long argued that a true broad-based shortage should show up as accelerated wage growth in the affected occupations, and argues that this hasn't materialized at the scale employer surveys claim [8]. Both organizations are working from credible data; they disagree on what it means. This investigation does not resolve that dispute — it notes that "shortage" and "mismatch" may be two different claims being talked about as if they were one.

This hiring dysfunction doesn't sit in isolation. Part 1 of this Convergence documented experienced workers leaving faster than organizations can replace their institutional knowledge — the same broken sorting mechanisms traced here are part of why replacement has become so hard, filtering out exactly the nontraditional and returning candidates who might otherwise fill that gap [6]. Part 3 of this Convergence then finds many of these same organizations, in the same quarter they describe hiring as frozen or unworkable, committing enormous capital to AI infrastructure instead — caution on people, conviction on compute. The gap between what employers say about skills-based hiring and what their own hiring data shows [4][5] is the same kind of institutional gap between stated intention and measured practice that shows up across this entire Convergence, whether the subject is retirement, hiring, or capital allocation.

The evidence above supports more than one honest read of why hiring feels broken. These aren't forecasts of which one wins — they're the distinct, evidence-grounded explanations that survive scrutiny, laid out so you can weigh them against what you've actually experienced.

No single villain explains this: postings, filters, and credential requirements were each optimized locally for volume and liability management, not for actually connecting a specific available person to a specific open role — and the resulting gap between what employers say about hiring and what their own data shows is the pattern worth tracking next.

Public, trackable indicators worth watching over the coming months and years:

• Labor data releases — Whether the JOLTS openings-to-hires gap narrows or widens in the next several BLS releases [1]. Whether SHRM's occupational-mismatch percentage narrows as training and reskilling programs mature, or holds steady, suggesting a structural rather than cyclical mismatch [8]. Whether wage growth in occupations with the loudest "shortage" claims accelerates in the way a genuine shortage would predict — the test EPI's argument rests on [8].
• Employer disclosures — Any employer disclosure tying skills-based hiring claims to actual hiring-outcome data rather than posting language alone. Whether Indeed Hiring Lab's bachelor's-degree-requirement share continues climbing past 19.3 percent or reverses again [5].
• Vendor/platform data — Whether major ATS vendors publish audited (not vendor-claimed) data on rejection and routing rates.

If you've been on either side of this — hiring, or applying — what part of this matches your experience, and what doesn't? Tell us. We read every message, and it shapes what we investigate next: Part 3 turns to the enormous capital bets on AI infrastructure being made by many of the same organizations struggling to hire well.