Automate·Advanced·45 min·Updated Sep 30, 2026

Build a recruitment assistant agent

Build a Copilot Studio agent that ranks candidates against a job's actual requirements to speed up the first screening pass.

Download PDF

Microsoft 365

Works With

Prerequisites

A well-specified job requirements document and a pool of candidate resumes/profiles to screen

Business Outcome

Faster time-to-shortlist for recruiters without lowering candidate quality at the top of the funnel.

Handle with care

Candidate data

Resumes carry candidates' personal information (contact details, full work history). Restrict the agent's knowledge source (step 2) to the active requisition's candidate pool, not a shared company-wide resume repository, and limit who can see individual scores to recruiters with a legitimate need.

Workflow Overview

The apps run in this order.

Word
Excel
Copilot Studio

Step 1: Write requirements the agent can actually checkWord

Rewrite the job description as a numbered checklist of concrete, verifiable items — "5+ years SQL," "active PMP certification," "built financial models in Excel" — and cut anything like "strong communicator" or "team player" that can't be confirmed from a resume's text.

Every item on this list should be something a human reviewer could also check and agree on; if two recruiters would score it differently, the agent will too.

Step 2: Build the ranking agentCopilot Studio

In Copilot Studio, have the agent read each candidate profile against the requirements checklist and return a score with the specific evidence from the resume backing each matched requirement.

Prompt idea:

Score this candidate against the requirements for the Senior Data Analyst role: 5+ years SQL experience, Power BI proficiency, and a degree in a quantitative field. For each requirement, state whether the resume shows evidence of it and quote the relevant line.

Step 3: Keep the agent to screening, never final selectionExcel

The agent's output is a ranked, evidence-backed shortlist for a recruiter to review — it should never auto-reject or auto-advance a candidate without human sign-off, given the legal exposure around biased or opaque hiring decisions.

Step 4: Audit for bias in what gets ranked highlyExcel

Export a batch of scored candidates into an Excel Table and sort by score against fields like school name, employment gaps, or anything else that correlates with protected characteristics.

If top scores cluster suspiciously around one pattern unrelated to the stated requirements, that's a signal the checklist or the agent's reading of it needs adjustment, not something to wave off as coincidence.

Check the work

  • Spot-check the quoted resume evidence behind a sample of scores — confirm the agent isn't inferring a qualification that isn't actually stated.
  • Have a recruiter independently review a few candidates the agent ranked low, to catch requirements the checklist missed.
  • Document the checklist and scoring logic used for each requisition, in case a hiring decision is later questioned.

Source: Microsoft Copilot Scenario Library — HR (2026)

Expected Outcome

A ranked shortlist of candidates against a specific role's requirements, with the reasoning for each ranking stated.

✓

AI is the right call here

A requirement phrased many different ways across resumes ("5+ years SQL" vs. a project description that implies it) needs reading comprehension to score with evidence — keyword matching alone misses the paraphrased cases.

Non-AI alternative: A basic keyword filter can pre-screen for an exact-phrase requirement like a named certification, but evidence-backed scoring against requirements stated in varied language needs the agent's comprehension.

Related Workflows

WORK WITH LIMINALS

Ready to roll this out beyond one person?

Workflows like this tend to raise real governance and licensing questions once more than one person is using them — that's exactly what we help with.