Skills Over Degrees: The Hiring Shift, Measured Against Reality

A slogan and a statistic

"Skills over degrees" has become the closest thing corporate hiring has to a consensus slogan. It shows up in earnings calls, LinkedIn posts, and state-government press releases with equal confidence, and it is repeated so often that it has started to sound like a fact rather than a claim. Some of it is fact. Employers really have pulled degree language out of job postings at scale, and state governments really have rewritten civil-service rules to drop the four-year mandate. That part of the story is measured, dated, and not seriously disputed.

The part that gets asserted far more casually is whether any of this changed who actually got the job. That is a different question, and it has a different, much less flattering answer. The most careful attempt to measure it, a joint study from Harvard Business School and the Burning Glass Institute tracking more than eleven thousand job postings across nearly a decade, found that dropping the degree requirement changed hiring outcomes for a number small enough to state precisely: fewer than one in seven hundred new hires. Nearly half of the firms that made the public announcement changed nothing about who they actually hired.

This study holds both halves of that finding at once, because both are true and neither cancels the other. The paper policy shift is real, structural, and pre-dates the current AI moment by years. The hiring-outcome shift is real in specific, mostly public-sector pockets, and largely absent everywhere else. What follows is an attempt to measure the gap between the announcement and the outcome, sector by sector, without pretending the gap does not exist because the slogan is popular.

Dropping the degree requirement changed hiring outcomes for fewer than one in seven hundred new hires. Nearly half the firms that announced the change hired exactly the same people as before.

The reset that is real, on paper

Start with what genuinely moved. The share of US job postings requiring a bachelor's degree fell from 51 percent in 2017 to 44 percent in 2024, a seven-point drop across the whole labor market, according to Burning Glass Institute and Lightcast analysis in "The Emerging Degree Reset." The shift was concentrated where it mattered most for scale: 46 percent of middle-skill occupations and 37 percent of high-skill occupations reset their degree requirements between 2017 and 2019 alone. The dates matter, because 63 percent of those resets happened before the pandemic. This was not a temporary, COVID-era labor shortage improvisation. It was already a structural trend before anyone had heard of a talent shortage caused by a lockdown.

The corporate names attached to the shift are large enough to move the aggregate numbers on their own. IBM removed degree requirements from more than half of its US job postings in 2021, and Dell and Bank of America followed with similar moves. LinkedIn's own platform data shows the same direction at smaller scale: the share of paid job posts on the platform that required no degree rose from 22 percent in 2020 to 26 percent in 2023. None of these are small or symbolic employers. Between them they represent a meaningful share of the postings any degree-reset statistic is built from.

Behind the posting-language shift sits a demographic fact that makes the reset make sense as a labor-supply strategy rather than a purely values-driven one. More than 70 million US workers, over half the national workforce, are what the nonprofit Opportunity@Work calls "Skilled Through Alternative Routes," or STARs: workers who built real capability through work experience, military service, or informal training rather than a four-year degree. About 64 percent of working-age Americans do not hold a bachelor's degree at all. A hiring strategy that screens on the credential rather than the skill is, by construction, screening out most of the available labor pool. Employers noticing that fact and rewriting job postings accordingly is not surprising. Whether the rewrite changed anything downstream is the separate question this study turns to next.

Where the policy shift also moved outcomes

Government hiring is the clearest counter-example to the "announcement without action" pattern, and it is worth stating plainly before the more skeptical material that follows, because it shows the reset can work when it is backed by mandate rather than press release. At least sixteen to twenty US states had dropped four-year-degree mandates for most state jobs by mid-2025, starting with Maryland in 2022. This is not a symbolic gesture buried in an HR policy document. It rewired the formal eligibility rules for large public-sector workforces.

Pennsylvania is the sharpest case in the public record. The state removed the degree requirement from roughly 92 percent of its state positions, and roughly 60 percent of the new hires who filled those reset roles arrived without a college credential. That is a policy change that produced a measured behavioral change, at a scale that dwarfs anything found in the private-sector data below. California removed degree requirements for roughly 30,000 public-sector jobs in 2025, extending the same policy to the country's largest state government.

The mechanism worth naming is that government hiring removes a variable that quietly undermines private-sector "commitments": discretion. A private employer that drops a degree requirement from a posting still has a hiring manager free to prefer the degree-holding candidate anyway, informally, without anyone auditing the choice. A state civil-service system that removes the mandate removes the legal basis for that discretion. The outcome data reflects the difference. Where the requirement was removed by rule, hiring changed. Where it was removed by announcement, it mostly did not, and that distinction is the hinge the rest of this study turns on.

The number that settles the argument

The single most important dataset in this debate comes from Harvard Business School and the Burning Glass Institute, published as "Skills-Based Hiring: The Long Road from Pronouncements to Practice." The study tracked more than 11,000 job postings from firms that had publicly announced they were dropping degree requirements, across 2014 to 2023, and then checked what those firms actually did afterward. The finding is precise enough to repeat exactly: non-degree hiring at these firms rose by only about 3.5 percentage points on average. Translated into people rather than percentages, that increase represented real new opportunity for fewer than one in seven hundred new hires, roughly 0.14 percent.

The study also breaks the aggregate apart in a way that explains why the number is so small. About 45 percent of the firms that made the public announcement showed no measurable change in hiring behavior at all: what the researchers describe as "in name only" adopters, firms that updated the posting language without changing who screened resumes, how those resumes were scored, or who ultimately got the interview. Only about 37 percent of the firms, described as genuine "skills-based hiring leaders," changed behavior in a way that showed up in the data, and those firms did make a real difference: they hired roughly 18 to 20 percent more non-degreed workers than they had before.

That last figure is worth sitting with, because it is the encouraging half of an otherwise sobering result. Skills-based hiring is not fictional. It works, measurably, for the minority of employers who actually rebuild their screening process around it. The problem the aggregate number exposes is one of adoption depth, not of concept validity: most employers who announced the change did not do the harder, less visible work of rebuilding the screen, and the announcement did the public-relations job on its own.

The paradox in one line

Put the survey data and the outcome data side by side and the disconnect becomes almost comic. In a 2025 employer survey, 85 percent of respondents say they practice skills-based hiring, up sharply from 57 percent just three years earlier in 2022. In the same period, roughly 88 percent of job postings still list a bachelor's degree as a requirement. Eighty-five percent claim the practice; eighty-eight percent of the postings contradict the claim; and the outcome data above shows the measured hiring effect sits near zero for most firms that announced the change. All three numbers can be true simultaneously, and they are.

The caveat belongs here and should not be softened. The 85 percent figure comes from TestGorilla's State of Skills-Based Hiring 2025 survey, a self-reported instrument published by a company that sells skills-assessment software and has a direct commercial interest in employers believing skills-based hiring is both popular and necessary. That does not make the number false, but it does mean it measures what employers say about themselves to a vendor, not what an independent researcher found them doing. The Harvard Business School and Burning Glass Institute figures are the reality check against that self-report, built from an outcome audit rather than a survey question, and this study treats them as the more reliable of the two.

The added weight behind the skepticism is the STARs recovery data. Skilled Through Alternative Routes workers, the population the whole skills-based hiring movement is nominally meant to open doors for, have regained only about 10 percent of the middle- and high-wage jobs they had previously been displaced from by credential inflation. If the posting-language shift were translating cleanly into hiring outcomes, that recovery figure should be far higher than 10 percent nine years into a documented reset. It is not, and that gap is the paradox stated as a single trend line rather than a snapshot.

Eighty-five percent of employers claim skills-based hiring. Eighty-eight percent of postings still require the degree anyway. Both numbers are true, and the gap between them is the entire debate.

Where it is real, and where it is not

The pattern is not uniform across the economy, and treating it as one national verdict obscures more than it reveals. In several domains skills-based hiring is not a slogan but an operating reality. Technology and IT were the earliest movers, the sector where IBM, Dell, and Bank of America made their public commitments and where technical assessments and portfolios have long substituted for credential screens in practice. Skilled trades never really ran on the four-year-degree signal to begin with, so the reset there reflects formalizing a practice that already existed rather than inventing a new one. Middle-skill administrative and support roles were among the fastest to reset formally, the 46 percent figure cited above. And public-sector hiring, discussed above, is the strongest outcome case precisely because rule replaced discretion.

Set against that is a large, largely untouched category: licensure-gated professions. Law, medicine, nursing, accounting, engineering, and architecture all require the credential by statute, not by employer preference. A hiring manager at a hospital cannot decide a nurse's clinical experience substitutes for a nursing license, because the state licensing board, not the employer, controls the gate. The same logic holds for the bar exam, the CPA credential, and the professional engineering license. No amount of posting-language reform touches these fields, because the requirement is not a corporate policy choice to begin with.

Finance sits in an interesting middle position. It has no formal licensure requirement analogous to a bar exam for most roles, yet the industry keeps the degree as what amounts to a de facto screen, an informal but durable filter that persists even where the formal barrier does not exist. That distinction, formal licensure versus informal but entrenched preference, matters for anyone trying to predict where the skills-based shift goes next. Formal licensure gates will not move without legislative or regulatory action. Informal preferences, like finance's, are exactly the kind of soft barrier the Harvard Business School data suggests most employers are unwilling to actually dismantle, even when they say they have.

AI as an unproven accelerant

Generative AI has entered the hiring process at exactly the moment this debate was already underway, and it is tempting to assume it will settle the question one way or the other. The early data does not yet support a confident answer in either direction. Among recruiters who have adopted generative AI tools, about 35 percent report redirecting the time they save toward candidate screening, and about 26 percent redirect it toward building or running skills assessments. Both figures point toward AI making skills-based evaluation cheaper to run at scale, which is the optimistic reading.

The pessimistic reading deserves equal space, because it is just as plausible on the current evidence. AI screening tools are trained on historical hiring data, and historical hiring data is saturated with degree-based outcomes built up over decades of credential-first screening. A model trained to predict "successful hire" from that history can quietly relearn the degree as a proxy signal, buried inside dozens of other features, even when the explicit degree checkbox has been removed from the application form. This is not a hypothetical risk; it is the standard failure mode documented across algorithmic hiring tools more broadly, and nothing in the current public data confirms that skills-based AI screening has avoided it.

The position here is that AI's effect on this specific debate is genuinely unproven rather than settled in either direction. It could be the tool that finally operationalizes the 37 percent of firms who are serious about skills-based hiring, giving them cheap, consistent assessment at scale. It could equally become the mechanism that reintroduces the credential bias the posting-language reset was supposed to remove, this time hidden inside a model rather than visible in a job description. Both outcomes are consistent with the 35 percent and 26 percent adoption figures above; neither is confirmed by them.

The reversal underneath the debate

This case study is one facet of a larger pattern running through this series. Prior waves of automation, industrial robotics and early enterprise software, hit routine manual and blue-collar work hardest, while leaving the credentialed, cognitive professions largely untouched. Generative AI inverts that pattern. Pew Research finds 27 percent of workers in bachelor's-or-higher occupations sit in the jobs most exposed to AI, against 12 percent of workers whose education stops at high school. OpenAI's "GPTs are GPTs" analysis finds a similar gradient at the task level: about 80 percent of the US workforce has at least 10 percent of its tasks exposed to large language models, and about 19 percent has at least half its tasks exposed, concentrated in higher-wage, higher-education occupations. The IMF estimates about 40 percent of global employment is exposed, rising to 60 percent in advanced economies where cognitive, credentialed work is a larger share of the labor market. Anthropic's own usage data shows 57 percent of observed tasks being augmented and 43 percent automated outright, again concentrated in white-collar, degree-coded work.

That reversal is exactly why the degree is under this much pressure at this particular moment, rather than in some earlier era of automation. A four-year credential has functioned for decades as a blunt, portable signal that a worker can perform a broad category of white-collar cognitive tasks. Generative AI is the first general-purpose technology capable of performing meaningful slices of that same cognitive work directly, which means the blanket signal the degree provides is worth less than it used to be, independent of how any individual employer chooses to write a job posting. The posting-language reset, in other words, is a downstream symptom of the exposure reversal, not an independent trend that happened to arrive at the same time.

The comparison worth naming is the United Kingdom, where the credential gate has already thinned much further than in the United States: only about 4.6 percent of UK job postings explicitly require a bachelor's degree. Whether that is the United States nine years from now, or a market with structurally different hiring norms that will not transplant cleanly, is not something the current data can settle. It is included here as a data point about how far the reset can plausibly go, not as a prediction of where the US market is headed.

Three transparent scenarios, not a forecast

Two organizations have published forward-looking figures on this trend, and both are presented here explicitly as scenarios rather than forecasts, because neither is a measured outcome. McKinsey projects that skills-based hiring becomes "standard" practice at large organizations, rising from about 25 percent of large organizations in 2024 to a projected 60 percent by 2027. That is a steep three-year climb, and it is a projection built on an assumed adoption curve, not an observed trend line the way the 2017-to-2024 posting data above is.

Deloitte's research adds a useful, more cautious counterweight. Skills-based organizations do perform measurably better on multiple internal metrics, which is a real finding, not a projection. But fewer than one in five organizations have actually adopted skills-based hiring at real organizational scale, even though 63 percent of work inside the average company already falls outside the boundaries of any formal job description. That last figure is the quiet argument for why the shift is likely to keep moving in one direction over time: the work itself has already outgrown the credential-shaped job description that is supposed to describe it, whether or not the hiring process has caught up.

Held together, the two projections describe a plausible but unconfirmed path: continued growth in stated adoption, continued gap between the largest, most sophisticated employers and everyone else, and a multi-year lag between the moment a practice becomes common in survey answers and the moment it shows up in the Harvard Business School kind of outcome audit. Nothing in the current evidence justifies treating 60 percent by 2027 as anything more than McKinsey's stated assumption.

The limits of this reading

Several constraints bound everything above. The 85 percent skills-based hiring figure is a self-reported vendor survey from a company that sells skills-assessment software, TestGorilla, and it should be weighted as employer rhetoric rather than independently verified behavior; it is included because it is the clearest public statement of the claim being tested, not because it is treated as reliable on its own. The Harvard Business School and Burning Glass Institute outcome study, by contrast, covers 2014 to 2023, which means its most recent data predates the wide deployment of generative AI screening tools in recruiting; whether AI adoption since 2023 has widened or narrowed the 3.5-percentage-point gap is not yet measured anywhere in the public record.

The state-level outcome data, Pennsylvania and California in particular, is the strongest evidence available that policy can move hiring, but it describes public-sector hiring specifically and should not be read as evidence about what private employers will do under similar policy pressure. The sector-by-sector "where it is real, where it is not" breakdown draws on separate studies covering different populations and time periods rather than one uniform national survey, so the comparisons across rows are directional rather than a single controlled measurement.

The AI-and-recruiting figures, the 35 percent and 26 percent time-reallocation numbers, come from an early, single-source dataset and describe stated recruiter behavior rather than a measured hiring outcome; treat them as a signal of where attention is moving, not as proof that AI screening improves or worsens skills-based hiring. And the McKinsey and Deloitte figures cited in the projections section are exactly that, projections and self-reported organizational surveys, not verified changes in who gets hired. What remains solid after those caveats is narrow but load-bearing: the posting-language reset happened, it happened mostly before the AI moment, and the hiring-outcome shift it was supposed to produce has, so far, mostly not followed it.