Candidate-Job Matching Score: How AI Evaluates (and Where It Goes Wrong)

The candidate matching score is the most-used and most-misunderstood feature on AI sourcing platforms. How does the AI assess the fit between a profile and a role? What does it base its judgment on? Where does it consistently get it wrong? This article opens the hood and explains what a matching score actually measures, using TrueCalling's TrueFit 360 score as a reference.

What is a candidate matching score?

A candidate matching score is a rating (often out of 100) that expresses the likelihood that a candidate fits a given role. The score combines several dimensions: skills, experience, company context, geography, availability. The score is not absolute truth — it's a useful ranking to prioritize outreach.

The 5 dimensions a good AI evaluates

1. Technical skills. Stack, languages, frameworks. The best AIs don't stop at the resume: they look at recent GitHub commits.
2. Experience trajectory. Seniority, progression, company size. A candidate who has shipped at 3 startups in a row is different from someone leaving 8 years at a large enterprise.
3. Company context. Sector, business model, stage (early, scale-up, public). Critical for cultural relevance.
4. Estimated availability. Tenure in the current role, weak signals (LinkedIn updates, conference activity, side projects).
5. Geography and mobility. Current location, openness to remote, time-zone compatibility.

How TrueFit 360 works

TrueCalling's TrueFit 360 score produces a rating out of 100 broken down by dimension, so a recruiter or hiring manager can understand where the recommendation comes from. For a Senior Data Engineer in Paris, you might see something like this:

• Technical skills: 92/100 (Spark, dbt, Airflow detected on GitHub).
• Trajectory: 88/100 (6 years, scale-up SaaS).
• Context: 75/100 (coming from a large enterprise, transition feasible).
• Availability: 80/100 (3 years in current role, openness signals).
• Geography: 100/100 (Paris, OK with remote).
• TrueFit 360 total: 89/100.

That explainability changes everything. Hiring managers trust a score they can read; a black-box score gets ignored.

Where the matching score always gets it wrong

1. Cultural fit

No AI reads company culture correctly. A great technical profile can be a relational nightmare. The candidate matching score is a relevance indicator, not a culture-fit indicator.

2. Real motivations

Why would the candidate change jobs? AI can detect signals, but it can't read minds. That's what the interview is for.

3. Atypical profiles

Someone who studied theater and then did backend at Google will rank low on the score. Yet that may be your best hire. Always review the outliers manually.

4. Data freshness

A score based on a LinkedIn profile that hasn't been updated in 2 years is wrong. Good platforms continuously enrich their data and surface freshness signals.

Matching score and bias: an ongoing watch

Any candidate matching score can reproduce biases present in training data. Gender, age, school, first name — all variables that can unduly weight the score if the AI isn't audited. Ask your vendor:

• Is the score auditable variable by variable?
• Are protected attributes excluded or controlled for?
• Is there an "anonymized resume" mode for the scoring phase?

How to use a matching score intelligently

1. Sort, don't exclude. Work the top 50 scores first, but keep an eye on the next 50.
2. Cross-reference score and human signals. A cover letter, a side project, a story told in an interview are worth more than a raw score.
3. Revisit the brief if too few candidates score above 80. The problem is often the brief, not the market.
4. Share the score with the hiring manager. Broken down by dimension, it becomes a discussion tool.

Matching score and recruiter productivity

Used well, the candidate matching score changes recruiter productivity. Instead of skimming 200 profiles, the recruiter works 30 in depth. To go further on recruiter productivity, read our article on the AI copilot for recruiters in daily practice.

How to evaluate a matching score before you buy

Before signing with a vendor, run this test: take 20 candidates you've recently hired. Have the engine score them. If the majority come out above 80, the score is learning well. Run the inverse test with 20 rejected candidates: most should land below 60. Without that double test, you're buying a score blind.

Conclusion: a matching score is a tool, not an oracle

A well-designed candidate matching score — like TrueFit 360 — speeds up triage and improves shortlist quality. But no score replaces your recruiter judgment. The golden rule: demand explainability, cross-check with human signals, audit for bias, and always keep an eye on the outliers.

To see TrueFit 360 in action on one of your briefs, See TrueFit 360 in action.