Skills-based hiring is any process where the decision to advance a candidate rests on evidence of what they can do, such as a work sample, a structured interview or a validated assessment, rather than on proxies for ability like a degree, a former employer, or years of experience.
- A proxy is any signal you accept instead of evidence: degree, brand-name employer, years served, referral.
- Removing a degree requirement is not skills-based hiring. Replacing it with a measurement is.
- The evidence must be collected the same way for every candidate, or you have moved subjectivity rather than removed it.
- It is a measurement discipline first and a sourcing strategy second. Get the measurement right or you scale noise.
- Skills-based does not mean unstructured, and it does not mean nobody decides. A human should always make the call.
The one-sentence version
Skills-based hiring means the decision to advance someone rests on evidence of what they can do, gathered the same way for every candidate.
That is the whole idea. Everything else people attach to the phrase, whether that is dropping degree requirements, widening sourcing, or running work samples, is either a consequence of taking that sentence seriously or a distraction from it.
The thing you are actually removing
A proxy is any signal you accept in place of evidence. Proxies are not stupid. They are cheap, they are fast, and they correlate with ability often enough to feel useful, which is precisely why they survive.
| Proxy | What you hope it tells you | What it actually measures |
|---|---|---|
| Degree from a named university | Can think, can finish hard things | Access to that university, and the ability to pay for it |
| Brand-name former employer | Was good enough to be hired there | That someone else's hiring process said yes, once |
| Years of experience | Depth | Time elapsed, which is not the same as time spent learning |
| Job titles | Scope and seniority | A former employer's internal naming conventions |
| Referral from a colleague | Vouched-for quality | The shape of your existing team's network |
Each of these carries a second, quieter cost: they inherit whoever had access. A signal that tracks access means filtering on it filters on access. The classic demonstration is Bertrand and Mullainathan's résumé-callback experiment, where otherwise identical applications drew materially different callback rates depending on the name at the top of the page. The résumé did not need to contain a biased statement. It only needed to carry an identity signal a human reader could act on.
The four things a skills-based process needs
A definition of the skill that two people would score the same way. "Strong communicator" is not a skill definition, it is a vibe. "Can take an ambiguous bug report, ask the two questions that narrow it, and write a reproduction case" is a skill definition. If your wording cannot be scored consistently by two different reviewers, you have not defined a skill yet.
Evidence, collected identically. The same task, the same time, the same information, the same rubric, for every candidate. The identical part is not bureaucracy. It is the entire source of comparability, and two candidates given different problems have not been compared at all.
A rubric written before you see any answers. Rubrics written afterwards describe the candidate you already liked. Write the scoring bands first, including what a mediocre answer looks like.
A human who decides, and is recorded as having decided. Measurement narrows the field and explains itself; a person still makes the call and owns it. This is also, increasingly, what regulators expect of any automated tool in the loop, which we cover in what "meaningful human oversight" actually requires.
Four things it is not
Not "no requirements." You have more requirements than before. They are simply about the work rather than the person's history.
Not automatically fairer. A badly designed assessment can produce a cleaner-looking, more defensible and equally discriminatory outcome, which is arguably worse because it now has a number attached. If you measure, you also have to check the outcomes across groups.
Not "let the model decide." Anything that can auto-reject a person is a policy decision disguised as a threshold. Keep the human.
Not unstructured. "We just have a good chat and get a feel for people" is the least skills-based process available, because it maximises the influence of everything except the work. Structured interviews exist to fix exactly this.
Where it goes wrong in practice
The take-home that measures free time. A six-hour unpaid exercise measures who has six unclaimed hours, which is a proxy for caring responsibilities and second jobs. Cap the time and mean it.
The assessment that tests the wrong thing. Algorithm puzzles for a role that is 80% integration work and stakeholder negotiation. The test is rigorous and irrelevant, and it produces confident bad hires.
Structure at the top, vibes at the bottom. Teams often assess rigorously and then hold a final "culture" round with no rubric, quietly restoring every bias the process removed. Whatever your last stage is, that is your real selection criterion.
Measuring, then ignoring it. If a hiring manager can override a score with "I just have a feeling," you have added a step rather than changed a decision.
How to tell whether it is working
Four things are worth tracking deliberately, because each fails differently.
Score-to-performance relationship. Do the people who scored well actually do well six months in? This is the only question that ultimately matters and the one most teams never check. Start recording it now so you can answer it later.
Pass-through rates by group. Compare selection rates across groups at every stage. A stage that looks neutral can be the one doing the damage.
Stage-level drop-off. Candidates abandoning a stage is data about your process, not about their commitment.
Reviewer agreement. Two reviewers scoring the same evidence should land close together. If they do not, your rubric is decorative.
Notice what is missing from that list. Time-to-hire and cost-per-hire are throughput metrics; they tell you how fast the machine runs, not whether it makes good decisions. That distinction is worth being explicit about.
What this looks like on CalHire
We built the platform so the skills-based version is the default rather than a discipline you have to maintain by hand.
Candidates take one cheat-resistant assessment, a skills test plus a text interview, and earn a verified composite score from test, interview and role-fit. It is valid for 90 days and portable across every application. There is no résumé-match component in the score at all.
Employers see verified skills and scores, never a name, photo, school or former employer. Identity is stripped at an architectural boundary before any model sees a candidate, so the score cannot be influenced by a signal that never reached it. Ranking runs on a composite whose weights you set per role, and below-threshold candidates are flagged for human review rather than auto-rejected. No code path can auto-reject anyone.
You can see the mechanics on the features page, the employer view on for employers, and the candidate side on for candidates.
Start with one role
Do not convert your whole process. Take one open role, define two or three skills precisely, write the rubric before you see a single submission, run every candidate through the same evidence, and record who decided and why. Then compare that cohort's outcomes against the last group you hired the old way.
One role, honestly measured, will teach you more than a policy announcement.
Frequently asked questions
- Is skills-based hiring the same as dropping degree requirements?
- No. Dropping a degree requirement removes a filter. Skills-based hiring replaces that filter with a measurement of the ability the degree was standing in for. If you remove the requirement without adding the measurement, you widen the funnel and make the decision more subjective, not less.
- Does skills-based hiring work for senior roles?
- Yes, but what you measure changes. For senior roles the relevant skills are judgement, prioritisation and design under ambiguity, which are better assessed through a realistic scenario and a structured conversation about trade-offs than through a coding exercise. The principle is unchanged: define the skill, then collect comparable evidence for every candidate.
- Is a resume still useful in a skills-based process?
- As a source of context, sometimes. As a basis for a decision, no. It is self-reported, unverifiable, and carries identity signals that are irrelevant to the work. On CalHire a resume can only pre-fill a candidate’s declared skills; it is never stored as the record and never shown as one.
- How do we know our skills assessment is any good?
- Check that it is job-related, that it is administered the same way for everyone, and that its results relate to on-the-job performance. Then check its outcomes across groups. An assessment nobody has validated is a proxy with better branding.