The right expert, not just more résumés
A US boutique legal staffing firm: 50 people, 170 placements a year.
- faster searches
- 10x
- pilot searches matched or beat experienced recruiters
- 20 of 20
- annual return
- $450K
- ranked shortlist, out of 10,000 profiles
- Top 50
Challenge
Keyword search kept missing strong candidates for niche legal counsel roles, and off-the-shelf tools are built for volume, not precision. The firm expected more work but didn't want to hire another recruiter.
What we built
AI candidate search built around the firm's own process and wired into its ATS. We designed it with the recruiters through ride-alongs and trained it on the firm's past hires and its best recruiters' judgment.
How it works
- 01
Search
An embedding model searches all 10,000 profiles by the meaning of the role, not keywords, and keeps the best 1,000.
- 02
Vote
Several reranking models score each candidate, and their votes are fused into a shortlist of 200.
- 03
Check
A fast, smaller model checks the must-have requirements and drops weak matches, leaving 100.
- 04
Rank
A larger model with agents reads each profile in depth and ranks the top 50 for the recruiter to review.
Results
- 10xfaster searches
- 20 of 20pilot searches matched or beat experienced recruiters
- $450Kannual return
- Top 50ranked shortlist, out of 10,000 profiles
Same team, more wins. Better matching adds about 8 placements a year, and the firm handles more work without a new hire and loses no deals at peak: about $450K a year.
