Case Study · Greenhouse Blog · 5/25/2026

HR Team Screens 10,000 Resumes in 2 Hours With an AI Agent

HR 团队用 AI Agent 在 2 小时内筛选 1 万份简历

# hr⚡ automation⚡ decision-supportClaude🟡 Low-code
Why it matters
The real unlock isn't speed — it's consistency. Human screeners are inconsistent across 10,000 resumes (fatigue, bias, different interpretations of requirements). The agent applies the same rubric to resume #9,847 as it does to resume #1.

The Problem

A fast-growing tech company posted a senior engineer role and received 10,847 applications in 72 hours. Their two-person recruiting team would typically spend 3 weeks doing first-pass screening — by which point top candidates had already accepted other offers.

The Agent Solution

They built a screening agent using Claude that takes each resume as input alongside a structured job rubric (required skills, preferred experience, deal-breakers). The agent outputs a score from 1–10, a one-paragraph rationale, and a "proceed/skip/maybe" recommendation. The whole batch ran overnight.

Results

The Rubric Design

The rubric was the hardest part. They spent a full day with the hiring manager converting vague requirements ("strong communicator") into measurable signals ("has given a conference talk, written a technical blog post, or managed cross-functional projects"). Garbage rubric = garbage shortlist.

Bias Mitigation

The agent was explicitly instructed to ignore name, graduation year (a proxy for age), and university prestige. They ran an A/B audit comparing AI-screened shortlists to manually screened ones — the AI shortlist had 23% more diverse candidates by gender and educational background.

Can You Do This Without Code?

Yes. Lever, Ashby, and Greenhouse all offer AI screening natively. For a DIY approach, a simple Claude or GPT-4 prompt with your rubric and resume text works immediately — no infrastructure needed for under 500 applications.

Related Cases