How does candidate matching work in an ATS?
How candidate matching works in modern recruitment software: the signals behind each score, skills versus titles, ranking, bias controls, and human oversight.
On this page
Candidate matching compares what a job requires with what each candidate offers and scores the overlap. Modern recruitment software weighs skills, experience depth, location, salary expectations, and availability rather than job titles alone, then ranks candidates so recruiters review the strongest profiles first. The score prioritizes work; a recruiter still decides.
What signals does candidate matching use?
A useful match score blends several signal families. The exact mix varies by product, but the core set looks like this:
| Signal | What it tells you | Watch out for |
|---|---|---|
| Skills | Whether they can do the work | Skills listed once but never used in a role |
| Experience depth | How long and how recently they did it | Years counted without context |
| Location and work mode | Whether they can realistically take the job | Stale relocation preferences |
| Salary expectations | Whether the role is winnable | Numbers from an application two years old |
| Availability | When they could start | Notice periods that changed |
| Engagement history | How they responded to you before | Penalizing people for your own slow replies |
Weak matching engines only do keyword overlap between the CV and the job description. That rewards candidates who happen to use the same vocabulary as the hiring manager and misses people who describe the same skill differently.
Skills or job titles: which matters more?
Skills, almost always. Job titles are inconsistent across companies: a marketing manager at a five-person startup and one at a multinational do different jobs, and plenty of skilled people carry vague titles like consultant. Title-based matching also punishes career changers whose skills transfer but whose titles do not.
Titles still carry some signal, mostly seniority and domain, so good engines use them as context rather than as the primary filter. A practical test when evaluating a vendor: feed in a strong candidate whose title does not match the vacancy and see whether the engine still finds them. How the profile got structured in the first place matters too, which is why CV parsing quality directly caps matching quality.
How does candidate ranking work?
Ranking turns individual signals into an ordered list. Each signal gets a weight, the weighted scores combine, and candidates sort from strongest to weakest fit for that specific role. Two properties separate trustworthy ranking from a black box:
- Explainability. Every score should show its evidence: which requirements matched, which were missing, what pushed someone up or down. A bare 87 you cannot interrogate is not a tool, it is a liability.
- Job-anchored scoring. A candidate is never strong in the abstract, only relative to a role. The same person should score differently against different vacancies.
Recruitifly scores each pipeline candidate on three axes: fit (skills and experience against the role), flight risk (how likely they are to drop out of your process), and readiness (how close they are to a decision). Together they tell you who is strong and who needs attention today. There is more detail in candidate scoring explained.
How do you keep candidate matching fair?
Four controls do most of the work:
- Score only job-relevant signals. Skills, experience, logistics. Protected characteristics such as age, gender, or origin have no place in a matching model.
- Keep evidence visible. Bias hides comfortably inside opaque scores; reviewable reasons expose it.
- Keep a human on every decision. Matching may order a list, but shortlisting and rejecting are human acts. In Recruitifly, the Fly assistant proposes and a recruiter confirms; nothing is auto-rejected on a score.
- Audit outcomes. Look periodically at who scores low and why. If a pattern looks demographic rather than skills-based, treat it as a defect and fix the model or the criteria.
An honest caveat: no vendor can promise bias-free matching, because job requirements themselves encode human choices. What software can do is make every step inspectable and keep accountability with a person.
What should you ask a vendor about matching?
Three questions sort the field quickly. Can I see why a candidate scored the way they did? Does the same candidate score differently per role? And what happens automatically when someone scores low? The only acceptable answer to the last one is: nothing, without human confirmation. You can see how Recruitifly handles all three across the features overview and in how the Fly assistant proposes every action before it runs.
Recruitifly is currently in private beta. If you want to see fit, flight-risk, and readiness scoring on your own roles and your own candidates, talk to us and join the beta.
Frequently asked questions
What is candidate matching in recruitment software?
Candidate matching is the feature that compares a job's requirements with each candidate's profile and produces a ranked list or score. It weighs signals such as skills, depth of experience, location, salary expectations, and availability, so recruiters review the most promising people first instead of reading every CV in arrival order.
Can candidate matching be biased?
Yes, it can be. Matching encodes human choices, so careless systems replicate past bias. Good systems reduce the risk by scoring only job-relevant signals such as skills and experience, excluding protected characteristics, showing the evidence behind every score, and keeping a recruiter responsible for each shortlist and rejection decision.
Do recruiters still make the final decision?
Yes, and they should. Matching is a prioritization tool, not a decision-maker. In Recruitifly, scores order the pipeline and the Fly assistant can propose actions such as advancing or rejecting a candidate, but nothing happens until a recruiter confirms it. The system never rejects anyone automatically on a score alone.
What is a flight-risk score?
A flight-risk score estimates how likely a candidate is to drop out of your process or accept a competing offer, based on signals like engagement, response patterns, and how long they have been waiting at a stage. Recruiters use it to prioritize follow-ups and speed up scheduling for in-demand candidates, not to filter anyone out.
Recruitifly Editorial
Editorial
Related reading
Adding LLM screening to your ATS without creating duplicate records
Layer LLM screening over your ATS without splitting your candidate data: one source of truth, stable ID sync, scores written back as fields, not copies.
Sourcing with adjacent job titles and skills
Searching one job title misses most of the market. A worked SRE example plus a repeatable method for mapping adjacent titles and skills for any role.
AI Act candidate disclosure: notice template
What to tell applicants when automated screening is used, under the EU AI Act and GDPR Articles 13, 14 and 22, plus a copy-paste disclosure notice template.
Want to see how this looks on your own data?
No hard promises. Just a straight conversation about exports, stages, and your current stack.
Contact us