The staffing firm had a good problem: more job orders than their recruiting team could handle. Applications were coming in faster than recruiters could screen them, qualified candidates were waiting days for a response, and the firm was losing placements to competitors who moved faster. They needed to scale their screening capacity without proportionally scaling headcount.

What we built: an AI screening system that handles first-pass qualification, automated async interviews, and recruiter routing — fully automatically, for 70% of applications. Here's what it does and how it works.

The screening pipeline we built

Stage 1: Resume parsing and minimum qualification check

When an application comes in, the system parses the resume and checks it against minimum qualification criteria for the job: years of experience, required certifications, location requirements, work authorization status, and any hard requirements specific to the client. This eliminates the bottom 30–40% of applications immediately — not by AI judgment, but by factual screening against objective criteria.

This is the most important design principle: the first filter is binary and objective. AI judgment comes in later, where it's more reliable.

Stage 2: AI-powered async screening interview

Candidates who pass minimum qualification get an automated async screening link — typically three to five questions specific to the role, recorded via video or text depending on the role type. The questions are written by the recruiting team for each job type, not generated by AI. The system sends this within minutes of the application, regardless of time of day.

Two things happen with those responses: an AI model scores them against a rubric developed with the client (relevant experience, communication quality, motivation indicators, red flags), and the responses are queued for recruiter review ordered by score.

The async interview step was the single biggest driver of screening time reduction. Recruiters were spending 15–20 minutes per candidate on phone screens. The async interview captures equivalent or better information in 10 minutes of the candidate's time and 3 minutes of the recruiter's review time.

Stage 3: AI scoring and recruiter routing

The AI doesn't make hiring decisions. It produces a score and a brief summary for each candidate — what they said, why it scores well or poorly against the rubric, and any flags for the recruiter to probe. Recruiters review candidates in score order, starting with the most qualified. The ones who need a human phone screen get one. The ones who don't pass AI scoring get an automated polite rejection.

What the 70% reduction actually means in practice

Before: a recruiter processing 100 applications would spend roughly 40 hours on screening — reading resumes, making initial calls, scheduling, documenting. After: the system handles the first pass on all 100 applications. The recruiter reviews roughly 25–30 AI-scored async interviews (the qualified pool after filtering), spending about 8–10 hours total.

The time savings compound: because qualified candidates get a response within minutes of applying (versus hours or days), acceptance rates on second-stage calls improved significantly. The firm was losing candidates to faster-moving competitors. That stopped.

What we deliberately didn't automate

Reference checks, final interviews, offer negotiation, and any role requiring specialized judgment (senior management, technical roles requiring skill verification) remained with human recruiters. The system was designed to maximize automation on the high-volume, lower-judgment work while preserving human touch where it genuinely matters.

We also built in explicit bias monitoring: the AI scoring rubric is reviewed quarterly against placement outcomes to check for any pattern of systematically under-scoring demographic groups. Good automation requires this kind of oversight built in from the start, not as an afterthought.

The honest limitations

The system performs best on roles with clear, definable qualification criteria. Highly creative roles, senior executive positions, and roles where "culture fit" is the primary qualifier aren't good automation candidates — at least not at the screening stage. Know your use case before automating. The right system for high-volume coordinator roles looks completely different from the right system for director-level hiring.