91.3 vs. 87: Why Candidate Match Scores Mislead Hiring Managers

A candidate scored 91.3. Another scored 87. On a dashboard, the first one looks like the better hire. In reality, no hiring model can show you that a gap of four points means anything at all.

Candidate match scores have become a standard feature of recruiting software. They are fast to read and easy to sort. They also hide most of what a hiring manager needs to know: how uncertain the number is, what it was trained on, and why it came out the way it did. This post explains what a match score actually claims, what the research says about predicting job performance, and what good decision support should show instead.

What a candidate match score actually claims

A match score compresses a judgment into a single number. A model reads a résumé, a profile or an assessment, compares it with the job, and outputs a value, often to one decimal place.

The decimal is the problem. It suggests a measurement, like a temperature or a weight. A measurement has a known error. A thermometer that reads 21.3 degrees is accurate to a stated margin. A candidate score of 91.3 almost never comes with one.

People trust precise numbers more than round ones. In consumer research, people judged estimates expressed in finer detail as more precise and had more confidence in them (Zhang and Schwarz, 2012). In another study, people giving more precise answers were judged more confident, and participants preferred the more precise advisors (Jerez-Fernandez, Angulo and Oppenheimer, 2014). A score of 91.3 borrows that credibility whether or not it has earned it.

Three things a match score hides

1. Its uncertainty. Even the best hiring methods predict job performance only partly. The most current meta-analysis of selection methods, by Sackett, Zhang, Berry and Lievens (2022), ranks structured interviews first:

Selection method	Corrected validity
Structured interviews	0.42
Job knowledge tests	0.40
Work samples	0.33
Cognitive ability tests	0.31
Unstructured interviews	0.19
Years of job experience	0.07

A validity of 0.42 means the method explains about 18% of the differences in job performance. Most of what makes someone succeed is not captured by any single predictor. A score built on a résumé is unlikely to beat the top of that table, so a four-point gap between two candidates sits well inside the noise.

2. Its training bias. A model learns from past decisions, including their biases. Reuters reported in 2018 that Amazon scrapped an experimental tool that rated candidates from one to five stars and penalized résumés containing the word "women's". It had been trained on ten years of résumés that came mostly from men. Amazon said recruiters never used it to evaluate candidates. The lesson stands: a score reflects whatever its training data rewarded.

Vendors rarely make this visible. A study of 18 hiring assessment vendors found that their websites did not make clear whether or how they validate their models, and that vendors rarely release rigorous validation studies (Raghavan, Barocas, Kleinberg and Levy, 2020).

3. That it is a verdict, not a measurement. A score does not tell a hiring manager what to look into. It tells them what to conclude. Two candidates can reach the same 87 for opposite reasons: one with deep experience and a thin track record of leading, the other the reverse. The number erases the very differences the interview should explore.

Why adoption outpaces trust

Hiring teams have adopted AI faster than they have learned to trust it. HireVue's 2026 report, based on more than 3,100 hiring managers, found that 77% of HR leaders now use AI weekly or daily, while only 40% trust AI-driven recommendations.

Speed has not followed either. GoodTime's 2026 Hiring Insights Report, a survey of more than 500 U.S. talent acquisition leaders, found that 60% of organizations saw time to hire increase in 2025, and only 1 in 9 managed to hire faster.

The gap is a scoping problem. When a tool delivers a verdict, people either follow it blindly or ignore it. When it delivers information, they can use it.

Regulators are moving in the same direction. New York City's Local Law 144 requires a bias audit within one year before an automated employment decision tool is used, a public summary of the results, and notice to candidates. The EU AI Act lists AI used to analyse and filter job applications and evaluate candidates as high-risk, with human oversight and transparency duties. Those obligations were pushed back to December 2027 by the 2026 Digital Omnibus.

What decision support should show instead

A hiring manager does need a fast way to think about a candidate. The question is what kind of summary helps them decide well. A useful one shows:

• The evidence, not just the conclusion. Which parts of the candidate's background match the role, and which do not, with the source of each point.
• What is still unknown. The gaps a résumé cannot answer, turned into questions for the next interview.
• A category, not a decimal. A small number of clear categories says honestly how much the evidence supports. It does not pretend to tell 91.3 from 87.
• The reasoning in plain words. Anything the hiring manager cannot explain to a colleague or a candidate should not drive the decision.
• A human making the call. The tool prepares the decision. A person makes it and owns it.

Structured interviews top the validity table for a reason: they make every interviewer test the same things and write down what they found. Our post on structured interviews vs. gut feel covers how to run them. Our guide to AI candidate screening without bias covers how to audit the tools themselves.

How HyreMynd does it: no black-box score

HyreMynd is an end-to-end recruiting platform with a built-in TRM, built on executive search rigour. We built it as human-in-the-lead recruiting. AI prepares people to decide; it does not decide for them.

That shapes what we do not build. HyreMynd does not score people with a ranking nobody can explain. It does not run your interviews, because candidates deserve a human conversation. And it does not make the final call on who gets hired.

What it does instead:

Team fit in four categories. HyreMynd assesses each candidate against how your team actually works and places them in one of four categories: Strong alignment, Fills a gap, Worth a conversation, or Lacks alignment. Four honest categories instead of a decimal, and a read you can explain to your hiring manager.

Custom interview guides. Every interview comes with a guide built for that candidate and that role. The gaps the team fit read surfaces become the questions the panel asks, so nothing important goes untested.

Launch alignment. Before any candidate is evaluated, HyreMynd compares the hiring manager's launch notes against the job description and flags every mismatch. A summary is only as good as the brief it is measured against.

The full picture before every stage. HyreMynd pulls together what is known about a candidate before each step, so the person deciding has more to go on, not a single number.

FAQ

Are candidate match scores accurate?
They are rarely as accurate as they look. Even the best hiring methods, such as structured interviews, explain only about 18% of differences in job performance. A score without a published error margin cannot show that 91.3 and 87 are different.

Should hiring managers rely on AI candidate scores?
Not as a verdict. Use AI to gather evidence, flag gaps and prepare interview questions, and keep the decision with a person who can explain it.

What is the most predictive hiring method?
According to Sackett and colleagues' 2022 meta-analysis, structured interviews have the highest corrected validity, at 0.42, ahead of job knowledge tests and work samples.

Is AI in hiring regulated?
Yes, in some places. New York City requires annual bias audits and candidate notice for automated employment decision tools. The EU AI Act classifies AI used in recruitment as high-risk, with obligations now set for December 2027.

What should replace a candidate score?
A short summary of the evidence for and against the match, a clear category instead of a decimal, the open questions for the next interview, and a human who makes the call.

Decide with more, not less

A single number feels decisive because it removes the work of thinking. That is exactly what makes it risky in hiring. The better hiring tools give people more context, ask sharper questions and leave the decision with someone who can defend it. That is how HyreMynd is built.

Book a demo to see how HyreMynd fits your team.