Resume-screening assist and structured scorecards
An AI assistant that reads each resume against a job-specific rubric and returns a structured scorecard with evidence, so your recruiters spend their time on real candidates instead of the first pass. A human still makes every decision.
Tools you'll use
Resume-screening assist is an AI assistant that reads each resume against a job-specific rubric and returns a structured scorecard with evidence, so recruiters spend their time on real candidates instead of the first pass. A human still makes every decision.
The volume problem is real and growing. Workday Recruiting customers processed 173 million job applications in the first half of 2024, up 31% year over year, while job openings rose only 7% (Workday Global Workforce Report, 2024). That means more piles to triage and more first-pass screening. Under that pressure, reviewers form quick impressions inconsistently. An eye-tracking study from Ladders found recruiters spend an average of 7.4 seconds on an initial resume scan (Ladders, 2018). Two reviewers looking at the same stack often rank it differently, and nobody can say exactly why a candidate was dropped.
The assistant scores each candidate on fixed criteria and shows its work: the criteria, the score, and the resume excerpts behind it. The win is twofold. You save the hours of triage, and you make the screen defensible, because every result is tied to job criteria with documented reasoning.
One caution: hiring is regulated, and automated rejection carries legal risk. This tool assists, it never decides. A person reviews every scorecard.
Moriva's take
Gate 1 (Real work): clear pass. Screening a resume stack is weekly, sometimes daily, work for any team that is hiring. Gate 2 (Owned): pass. One operator can build and run this with Claude Code or Codex, or run it as a workflow in Claude Cowork without writing code, and your team can adjust the rubric whenever a role changes. Gate 3 (Measured): pass. Track hours saved per req and time-to-first-interview; both move quickly. The reason this is CAREFUL and not GO: hiring is regulated, an automated screen that rejects candidates can create adverse-impact and disclosure liability, so a human must review every scorecard and the assistant must assist, never decide.
How do you resume-screening assist and structured scorecards?
- 1
Write the rubric before you touch a tool
Sit with the hiring manager and turn the job into a scorecard: the must-have skills, the nice-to-haves, and the years or kind of experience that actually predict success in this role. Assign weights (a common split is role-specific skills around 40 percent, relevant experience 30, credentials 15, and judgment or domain fit 15) and a 1-to-5 scale with a written definition for each score. Keep every criterion job-related and defensible. This document is the heart of the system; the tool just applies it.
- 2
Point the tool at one real, closed req first
Open Claude Code or Codex in a folder, or start a workspace in Claude Cowork, and give it the rubric plus the resumes from a role you have already filled. In plain English, ask it to score each resume against the rubric and return a scorecard with the evidence for each score. Using a closed req means you already know who was strong, so you can check the assistant's judgment against reality before you trust it on live applicants.
- 3
Make it output a structured scorecard, not a verdict
Tell the tool to produce, for every candidate, the score per criterion, the exact resume line that supports each score, and a short list of gaps or open questions. Ask it to explicitly avoid recommending hire or reject. The output should be a CSV or table your recruiters can sort, plus a one-page brief per candidate. Codex and Claude Code can build this as a script you own and rerun; Claude Cowork can produce it as a document set without code.
- 4
Build in blind screening and a bias guardrail
Have the tool ignore and not score name, age, photo, address, graduation year, and other signals unrelated to the job, and instruct it to base every score only on skills and experience. Ask it to flag its own reasoning if it ever leans on something off-rubric. This is a real bias-reduction practice, and because the assistant writes down its reasons, you can audit them, something a black-box screen cannot give you.
- 5
Calibrate with your recruiters
Run a short session where two or three reviewers score the same five resumes by hand, then compare against the assistant's scorecards. Where they disagree, fix the rubric wording, not the people. This calibration is what makes scores mean the same thing across reviewers and across weeks, and it surfaces criteria that were vague before the tool exposed them.
- 6
Keep the human as the decision-maker
Wire the workflow so the assistant produces scorecards and a human reviews every one before anyone is advanced or declined. Do not let the system auto-reject below a threshold; that is where legal and fairness risk concentrates. The recruiter sorts by score to prioritize their attention, reads the briefs, overrides where the assistant missed context, and makes the call.
- 7
Measure and adjust each cycle
After each role, record hours spent on first-pass screening before and after, time-to-first-interview, and how often recruiters overrode the assistant. A high override rate on one criterion means the rubric needs work. Because you own the rubric and the script, your team edits and reruns it for the next req without calling anyone.
What could go wrong (and how to handle it)
Learned bias and adverse impact. An assistant that infers patterns from your past hires can quietly favor signals that correlate with protected groups, the failure that forced Amazon to scrap its hiring AI when it learned to downgrade resumes mentioning 'women's'.
Score against an explicit, job-related rubric rather than 'who looks like our team'. Use blind screening to drop off-rubric signals. Periodically check selection rates by group against the EEOC four-fifths rule, and if you operate in a jurisdiction like NYC, treat a bias audit as mandatory, not optional.
Over-automation. The biggest legal and fairness exposure comes from auto-rejecting candidates below a score with no human review.
Never let the system decline anyone on its own. A human reviews every scorecard before any candidate is advanced or rejected. The assistant prioritizes attention; it does not make decisions.
Hallucinated or inflated scores. The model can occasionally invent a qualification or read more into a line than is there.
Require an exact supporting quote from the resume for every score. A recruiter spot-checks the quotes against the source. No evidence, no score.
Keyword gaming and AI-written resumes. Candidates stuff keywords or generate resumes to beat automated screens, degrading the signal.
Score on demonstrated experience and outcomes, not keyword presence, and have the assistant flag thin or generic claims as open questions for the human rather than rewarding them.
Sensitive data handling. Resumes are personal data and may carry protected-class information.
Treat data sensitivity as high. Use an account with appropriate data handling, keep resumes in controlled storage, limit access to the hiring team, and delete copies on the schedule your policy requires.
Disclosure and notice rules. Some jurisdictions require telling candidates an automated tool is used and publishing bias-audit results; the EU AI Act treats recruitment screening as high-risk.
Confirm your obligations before going live. Where notice is required, give it. Where an independent bias audit is required, run one annually. Keep the human-decision design so the tool is assistive, which lowers your exposure.
Prompts to get started
FAQ
Will this make our hiring decisions for us?
No, and it should not. The assistant reads resumes and produces scorecards with evidence so your recruiters can prioritize their attention. A human reviews every scorecard and makes every advance-or-decline call. That design is both fairer and lower-risk than letting an algorithm reject people on its own.
Is automated resume screening even legal?
Assistive screening with a human deciding is widely used and defensible. The risk lives in fully automated rejection and in disclosure rules. Laws like NYC's Local Law 144 require bias audits and candidate notice for automated decision tools, and the EU AI Act treats recruitment screening as high-risk. Check your jurisdiction, keep a human in the loop, and audit for adverse impact using the EEOC four-fifths rule.
How is this different from the screening built into our ATS?
Most ATS screens are keyword filters or black boxes; you cannot see why a candidate was ranked where they were. This produces a transparent scorecard with the reasoning and the supporting quote behind every score, against a rubric you wrote. You own the rubric and the logic, and you can change them whenever a role changes.
Do we need engineers to build and maintain it?
No. One operator can stand up a first version in roughly a week. A non-coder can run it as a workflow in Claude Cowork. If you want a reusable script, Claude Code or Codex builds one your team owns and reruns, and you edit the rubric in plain English without calling a consultant.
What stops it from just preferring people who look like our current team?
Two things. First, you score against an explicit job-related rubric rather than similarity to past hires, which is exactly the trap that broke Amazon's hiring AI. Second, blind screening drops off-rubric signals, and because every score carries a written reason, you can audit the logic and check selection rates by group for adverse impact, something a black-box tool will never let you do.
Sources
- Workday Recruiting customers processed 173 million job applications in the first half of 2024 - a 31% increase - while job requisitions rose only 7%, meaning applications grew roughly four times faster than openings. — Workday Global Workforce Report, 2024
- Recruiters spend an average of 7.4 seconds on an initial resume scan, according to an eye-tracking study. — Ladders, 2018
- Amazon scrapped its experimental recruiting AI after it learned to downgrade resumes that included the word "women's". — Reuters, 2018
More from HR & People
Job descriptions and structured interview kits
Turn a role's real requirements into a clean job description and a matching structured interview kit (questions plus a scoring rubric) that your team owns, reuses, and runs the same way for every candidate.
Onboarding docs and a policy Q&A bot
Turn your scattered onboarding and policy documents into a Q&A assistant that answers new-hire and employee questions with quoted sources, and escalates anything sensitive to a human.
Prep renewals and spot upsell openings
Build a repeatable agent that pulls usage, support, and account data into a per-customer renewal brief flagging churn risk and concrete upsell openings, 90 days ahead of each renewal date.
Want help shipping this?
We'll build it with your team on your real work — and leave you owning it, not renting it.