On this page
A recruiter screening 300 CVs for one backend role spends roughly 20 to 25 hours before a single interview is booked (our own estimate: 300 CVs at 4 to 5 minutes each). Automating that step is easy. Automating it without quietly baking in yesterday's hiring patterns is the hard part.
Short answer: AI screening shortens time-to-hire by removing the queue in front of the first human read. It stays fair only if it scores each CV against the written job requirements, strips proxy signals such as name, photo and address, and shows the quoted evidence behind every score. Across twelve EvalCV beta customers, the median time to first interview fell from 11 to 4 working days over eight weeks.
The real cost of slow screening
Across twelve EvalCV beta customers, the median time from application to first interview was 11 working days. Most of that was queueing: CVs waiting for someone with enough context to read them properly. Strong candidates rarely wait that long. By day eight, a third of the shortlisted applicants had already accepted another process.
Where bias hides in automated screening
Most screening tools learn from past hiring decisions. If those decisions favoured certain universities, postcodes or career shapes, the model learns to favour them too, and presents the result as an objective number.
Proxies are the problem, not the model
Names and photos are obvious. Career gaps, graduation years and home addresses are not, but they correlate just as strongly with protected characteristics. EvalCV strips them before scoring and logs that it did.
“A score nobody can explain is just a faster way to make the same mistakes.”
Scores you can explain
Instead of training on past outcomes, EvalCV scores each CV only against the requirements written in the job description. Every point in the score maps to a requirement and a quoted line of evidence from the CV.
In the recruiter portal, the score breakdown lists each requirement with the sentence in the CV that satisfied it. Recruiters can click through from a number to the words behind it, and disagree with it if they want to.
What the API returns
The same evidence is available programmatically, so teams plugging EvalCV into their ATS can show it to hiring managers too. This is the same pattern we apply in AI integration projects: the evidence travels with the score.
POST /v1/score
{ "job_id": "job_8f2c41", "cv_id": "cv_31d0a7" }
200 OK
{
"score": 78,
"matched": ["Spring Boot", "PostgreSQL", "5+ years backend"],
"missing": ["Kubernetes in production"],
"evidence": [{
"requirement": "Spring Boot",
"quote": "Led migration of 14 services to Spring Boot 3"
}],
"excluded_signals": ["name", "photo", "date_of_birth", "address"]
}
The excluded signals list
The excluded_signals field is there for auditors. It records which fields were removed before the model saw the CV, on every single request. Regulators treat this as audit territory: the EU AI Act classes AI systems used to filter or evaluate job applications as high-risk (Annex III), and New York City Local Law 144 requires a bias audit of automated employment decision tools.
Measuring time-to-hire honestly
After eight weeks, median time to first interview dropped from 11 to 4 working days. More useful: the share of interviewed candidates from non-traditional backgrounds rose slightly rather than fell, which is the opposite of what we feared.
Try it on your own roles. EvalCV's free tier lets you score CVs against a live job description. Plans and limits are on evalcv.com.
What we'd do differently
We shipped scoring before we shipped the evidence view. Recruiters trusted the numbers less in those first weeks, and they were right to. Next time, the explanation ships first.
Found this useful? Share it with your hiring team.
Share on LinkedInRelated product
EvalCV
AI Hiring
AI resume screening API and recruiter portal that scores candidates against a job description in seconds.



