Hiring had become the bottleneck. Time-to-fill crossed 70 days, and the old system screened on exact keywords, so a strong engineer who wrote "server-side development" was auto-rejected because the role said "back-end", while weaker candidates who gamed the wording sailed through.
The screening was keyword-deep. It matched the exact words in a job description, so it rejected excellent people for a synonym and passed mediocre ones who knew which terms to paste in, and recruiters still had to read the pile it let through by hand.
That manual load was the tax. Recruiters spent around 60% of their time reading CVs that were never relevant, and the genuinely good applications were lost in the same pile.
And human bias crept in at every step. Recruiters leaned, often unconsciously, on which school someone attended, their age or where they lived, which quietly narrowed the diversity of who got hired.
Match on what a person can actually do, strip what shouldn't matter, and put the strongest in front of the recruiter first.
We replaced keyword screening with a natural-language engine that understands context, recognises equivalent skills, and infers what a candidate has actually done from how they describe past roles, so "server-side" and "back-end" finally count as the same thing.
A skills model maps related and adjacent capabilities, so a candidate is credited for the experience their work implies, not only for the exact phrases they happened to write down.
A layer hides identifying details, name, gender, age, photo and school, from the recruiter's view and presents a candidate summary built purely on skills and fit to the role's requirements.
The system scores every applicant for fit and pushes the top tier straight to the front of the recruiter's list for immediate outreach, instead of leaving them buried in the stack.
The model surfaces and ranks; recruiters and hiring managers make every interview and offer decision. It clears the pile so the human judgement lands on the candidates who deserve it.
Because recruiters meet only the closest matches, the whole pipeline moves faster, and vacancies stop holding up the business units waiting on the hire.
Half the time-to-fill, better interviews, a broader pool.
Average time-to-fill dropped from 70 days to about 38, which stopped vacancies from stalling the projects that depended on them, and the interview-to-offer ratio rose 35% because recruiters were only meeting the closest matches.
Removing identifying details from the first screen lifted workforce diversity 20%, as candidates advanced on skills and fit alone rather than on the signals that used to bias the decision.
Keyword screening measures who wrote the right words, not who can do the job, and it carries human bias straight into the shortlist. The value here is reading a CV for real skill, hiding what shouldn't count, and ranking by fit so the recruiter's judgement lands where it matters. It is the same score-and-rank discipline behind our lead-scoring work, aimed at candidates and skills rather than sales leads.
Reps burned time on leads that never converted. AI enrichment and predictive scoring with conversational qualification lifted lead-to-deal conversion 47.6% and cut strategic-lead response from 48 hours to about 5 minutes.
Read the case → Onboarding & training · SaaSA SaaS company's new hires took four months to ramp on knowledge buried in old docs and video. AI turned it into adaptive, role-specific training with manager sign-off, cutting ramp to six weeks and lifting retention 25%.
Read the case →
Reef TRH