01 the business
Regional logistics company
400 staff, hiring 80 people a year
A single warehouse supervisor role can bring in 300 applications. Hiring managers skim, strong candidates get missed, and nobody can explain afterwards why one person got an interview and another did not.
02 before
The problem
A single job ad can bring in far more applications than a hiring manager can read properly. People skim, strong candidates get missed, and nobody can explain afterwards why one person got an interview and another did not.
- Applications are skimmed in minutes.
- Criteria shift from one candidate to the next.
- No written record explains decisions.
03 the build
What the AI does
- Step 1: Turns the job description into a short list of must-have and nice-to-have criteria that the hiring manager confirms before screening starts.
- Step 2: Reads each resume and application answer from Greenhouse, Workable or your inbox.
- Step 3: Scores each candidate against every criterion and writes a sentence of evidence for each score, quoting the resume.
- Step 4: Leaves out names, photos, ages and other details that should not affect the decision when showing the summary.
- Step 5: Sorts candidates into a shortlist, a maybe pile and a decline pile, with the reasons visible for each.
paused: waiting for a person to approve
How it was put together
- Turned the job description into must-have and nice-to-have criteria the hiring manager confirmed.
- Scored each application against every criterion with a sentence of evidence quoted from the resume.
- Hid names, photos, ages and other details that should not affect the decision.
- Sorted candidates into shortlist, maybe and decline, with reasons visible.
04 human in the loop
Where a person stays in charge
A hiring manager reviews the shortlist and the written reasons and decides who moves forward. No candidate is rejected or contacted without a person approving it.
- A hiring manager decides who moves forward; no one is rejected without a person approving it.
- Outcomes are checked for adverse impact across groups, and a bias audit is run where the law requires it.
- Candidates are told how automated tools are used, as local law requires.
05 rollout
From kickoff to live in 5 weeks
- Weeks 1–2Legal review for your locations; agree criteria templates and the audit approach.
- Weeks 3–4Score past applications and compare with real hiring decisions.
- Week 5Use on live roles with the hiring manager reviewing every decision.
06 after
What changes
Every applicant is read against the same criteria, and the team can explain each decision. Hiring managers spend their time on the shortlist instead of the whole pile.
Assume 300 applications for a role and 3 minutes to read each properly, against 30 minutes to review a scored shortlist with reasons.
300 × 3 minutes = 15 hours to read every application properly. Reviewing the scored list takes about half an hour, plus time with the strongest candidates.
What to measure
- Hiring manager time per role.
- Time from application to first contact.
- Shortlist-to-hire rate.
- Selection rates across groups, reviewed regularly.
07 evidence
What the research says
Employers in New York City may only use automated hiring tools that had a bias audit within the past year, must publish a summary of it and must notify candidates.
AI used to recruit or select people, or to make decisions about promotion, termination, task allocation or performance, is classed as high-risk.
EU AI Act, Annex III: high-risk AI systems (opens in a new tab) · 2024
NIST's framework organises AI risk work into four functions, Govern, Map, Measure and Manage, with a profile for generative AI added in July 2024.
NIST, AI Risk Management Framework 1.0 and Generative AI Profile (opens in a new tab) · 2023
08 lessons
Mistakes to avoid
- Letting the tool reject people. Keep a human decision.
- Criteria that proxy for protected traits, such as gaps in employment.
- Ignoring the law. Hiring tools are regulated in more places every year.
09 questions
Common questions
- Is this legal?
- It can be, with care. New York City requires a bias audit and candidate notice for automated hiring tools, and the EU AI Act classes recruitment AI as high-risk. We design for both.
- Will candidates know?
- They should. We include a plain notice on the application page.