AI CV Scoring: How It Works and Why It's Reliable
AI reads your CVs in seconds across four key dimensions. But how does the scoring actually work — and can you trust it? We asked the question and answer it here.
Jean Toselli · · 9 min read
CV screening: a problem of volume and subjectivity
A frontline recruiter receives 80 to 150 CVs on average for a single frontline role. There isn't time to assess each one objectively and pinpoint 'the ideal candidate.' They have to read every CV and narrow the field to 20, or even 10.
The result: decisions made out of fatigue, cognitive biases around names, schools, spelling. And the nagging sense that good candidates were passed over simply because their CV wasn't 'well formatted.'
AI scoring answers this problem. But how does it really work? Is it genuinely reliable? Here's our explanation, no marketing.
The four dimensions of Rafa scoring
Rafa's AI scoring analyzes each application across four complementary axes, combined into an overall score out of 100.
1. Professional experience
Rafa analyzes the consistency and depth of each candidate's track record: time spent in the trade, recurrence, progression, similar job titles. A warehouse worker with 8 continuous years in logistics scores well on skills, comparable experience, and the stability signals expected in a 'production operations' type of role.
This isn't just a keyword 'match' on the job title. It's a contextual analysis that accounts for the profile you're looking for, the industry, and the level of the role.
2. Education
Level of qualification, fit with the role, additional training, certifications. For frontline jobs, formal education often matters less than experience — Rafa weights it accordingly, based on the specifics of the role.
3. Motivation (NLP analysis)
This is where AI makes the difference. Using natural language processing (NLP), Rafa analyzes cover letters, free-text answers, and candidate messages to gauge motivation beyond a simple 'this is my dream job.'
It detects:
- Extrinsic vs. intrinsic motivation signals: how much the candidate actually knows about the company and the market
- Consistency between what a candidate says and what their track record shows
- Warning signs, or signals of a poor fit with the job or the profession
A candidate with an average CV but solid motivation scores higher than a purely going-through-the-motions candidate — because motivation predicts engagement and staying power.
4. Custom criteria
Every role is unique. Beyond the three dimensions above, each client sets their own specific criteria: driver's license, technical skills (e.g. CACES forklift certification, safety clearances), product or sector experience, availability, languages. These criteria are weighted separately, according to your priorities.
Why NLP changes everything when it comes to motivation
'Motivation' is the hardest thing to convey in a CV. It's also the worst reason to drop a candidate — gut-feel judgments are misleading.
NLP (Natural Language Processing) lets AI understand the meaning of a text, not just its form. Concretely:
- 'I'm passionate and I've always given it my all' → weak signal (generic)
- 'I just found out your Lille location is a 15-minute drive from my home' → strong signal (specific, shows real knowledge)
It's this kind of nuanced sorting, across thousands of applications, that only AI can support. Screening by motivation still matters. But it isn't reliable at scale, because feelings and impressions are, by definition, inconsistent.
Score transparency: understanding the why
A score without an explanation is a black box. At Rafa, every score is explained.
For a score of 75/100, here's what that actually means:
- Experience: 85/100 — '3 years in materials handling, steady progression'
- Education: 60/100 — 'High-school level, no sector-specific training'
- Motivation: 90/100 — 'Personalized letter, references the company's plans and specific interest in the role'
- Custom criteria: 70/100 — 'Valid driver's license, partial availability'
This gives the recruiter — and their manager — decision support based on tailored data.
AI scoring vs. manual screening: an honest comparison
AI scoring isn't 'better than a human.' It's faster. Here's an honest comparison:
| Criterion | Manual screening | Rafa AI scoring |
|---|---|---|
| Speed | 5 to 8 min per CV | < 1 second |
| Consistency | Variable (fatigue, mood) | 100% consistent |
| Bias | Unconscious but very real | Reduced (algorithmic bias audited) |
| Application volume | Unworkable at 150+ CVs | Scalable |
| Cost | High (HR time) | Included |
In practice, the two work best together. AI screens the 80 CVs and flags the 10–15 worth a manual review. Your HR team goes from 8 hours to 45 minutes of screening, and the screening is structurally more objective — while staying in the hands of your operational managers and human recruiters.
Rafa Core: scoring in service of the frontline
Rafa Core embeds AI scoring across the whole chain, from the social ads campaign to the booked interview. The candidate's score is updated continuously:
- The candidate applies → Rafa scores them in real time
- The strongest profiles are flagged for outreach by SMS and/or email
- Qualified candidates receive a booking link to grab an interview slot
- Your recruiters only review the top profiles in the process — continuously, without any manual HR overhead
In summary
- AI scoring is reliable when it's transparent, contextual, and tied to the frontline. It doesn't replace human judgment — it augments it.
- Four dimensions: experience, education, NLP-based motivation, custom criteria
- Two goals: move faster, but not in the wrong direction
- Maximum consistency, reduced bias
- A two-column HR view: the score and the ways to improve it
See AI scoring in action
Try Rafa Core on your next hires. Results guaranteed or your money back.