On a role with real volume, the same question gets answered in several places at once. A sourcer reads profiles and flags the ones that fit, while a recruiter works through the applications. Then someone pulls a former finalist out of the applicant tracking system because their interview went well. Each of them calls their picks a strong match, and the hiring manager opens a shortlist where every name passed a different test.
AI candidate matching speeds up every one of those reads, and the shortlist still arrives built on different tests. Each tool works from the brief it was given. When the sourcing prompt and the inbound profile were written by different people, the tools hand back confident picks on different terms, only sooner.
I think the scoring layer in candidate matching is something your team writes and keeps, whatever runs the search. It is a rule the team writes and the hiring manager signs: the lines a candidate must clear, and how the rest trade off. Write it once for the role and have every motion mark its candidates against it. Every tool’s output then gets checked against that same rule, so the shortlist rests on one bar.
Decide what a match means for this role.
Start with one page the hiring manager signs before any motion runs. It holds two kinds of line. Gates are the two or three things a candidate must have, and weighted lines are the three to five that make a candidate stronger without being required. Under them sits a single rule saying how the weighted lines combine. On a page for a first account executive who sells to finance teams at mid-sized companies, that looks like this:
- Gate: has carried a new-business quota.
- Gate: has sold to finance leaders, such as a chief financial officer.
- Weighted: has closed deals that took longer than a quarter.
- Weighted: has built their own pipeline without a sales development team behind them.
- Weighted: has worked somewhere small enough that sales had no playbook yet.
- Weighted: has sold into companies the size of your customers.
- Rule: both gates, plus at least two weighted lines, one of them the pipeline line.
That rule is the scoring layer, and I’d keep it written out as a sentence. Points let a strong weighted line buy back a missing gate, and a total of seven tells the hiring manager nothing about which trade the recruiter made to reach it. Written rules show the trade, so the hiring manager can disagree with it before a single candidate is marked.
Most of these lines are first said out loud on the intake call, so draft the page straight after it, while the hiring manager can still correct your wording.
Mark every motion against the same rule.
Three motions feed one shortlist on this kind of role, and each finds people with different evidence in hand. A sourced profile carries what someone chose to list in public, and an application adds what they wrote for this role. A candidate pulled back from an earlier search brings an interview record, which holds what they said when an interviewer asked.
The rule only works across those motions if every match arrives in the same form. I’d have each one written as a mark per line, with the place the evidence came from in brackets, and the trade spelled out at the end:
“Quota: met (application). Finance leaders: met (application). Long cycle: met (past interview). Own pipeline: met (past interview). Company stage: not met. Customer size: met (profile). Trade: pipeline, long cycle, and customer size.”
Written that way, a sourced candidate and an applicant can sit side by side on one shortlist, and the hiring manager can argue with one mark at a time. The bracket counts as much as the mark: a line met in a past interview, in the candidate’s own words, is stronger than the same line inferred from a job title. This guide to candidate quality sorts out what a profile can carry and what waits for a conversation.
Tally the trade each motion makes.
With every match carrying its trade, you can see which weighted lines each motion leans on. Once a month, I’d count, for each motion, how often each weighted line was part of the trade that got a candidate through.
Expect the counts to differ, because each source makes some lines cheap to mark. Outbound will lean on company stage, since headcount is usually easy to find, and past candidates will lean on the lines interviewers asked about. Watch the rule’s open slot: a motion that fills it with the same easy line almost every time deserves a closer look. On the example page, that would be outbound matches that clear the pipeline line and then always add company stage.
That pattern means the motion is matching on a narrower standard than the page describes. I’d put it to the hiring manager as a question about the rule itself: either company stage deserves less weight, or the rule should name which second line it wants.
Change the standard in one place, then re-mark.
The page will change. After the first interviews, the hiring manager usually wants something different, and a typical edit promotes a weighted line to a gate. Make that edit on the page itself, and keep old marks visible beside the new ones for each candidate, so you can see whom the change moved.
Then re-mark the candidates already in process and rerun the sourcing search from the new page. Past candidates waiting on a rediscovery list need the same re-mark, and it is the easiest one to skip. Skip it, and a former finalist gets turned down for a line that no longer exists, or waved through without ever being marked on a line added after their interview. Bringing past candidates back into a live search has its own method, set out step by step in a guide to candidate rediscovery.
Where Metaview fits in AI candidate matching.
Each motion brings a different kind of evidence to the same page, and Metaview contributes something different to each one. None of it counts as a mark until someone on the team writes the mark down. A keyword filter cuts everyone who described the work differently before a person has read them, and a page of gates and weighted lines exists so that no motion has to work that way.
I don’t think the answer to an information problem is making decisions with even less information.”
Marks taken from a recorded interview are the strongest a past candidate can carry. That record exists only if the call was recorded, and a call is recorded only where consent was given under your team’s own process. On those calls the Metaview Notetaker joined as a visible participant and recorded and transcribed the conversation, and the record holds structured notes beside the transcript.
In Reports, your team can put questions to its own interview records without leaving the product. Metaview captures every spoken word of the interview, so a question can ask what the screens discussed as well as what was decided, and the illustration below puts one to a sales team’s screens:
- 1The role selector at the top right, which keeps the report to one sales team’s screens.
- 2“Consultative selling” at the top of “Coverage by competency,” discussed in most of the screens.
- 3“Collaboration,” a bar partway along the chart, between the most and the least discussed competencies.
Set a chart like this beside your page. Where a line the rule requires maps to a scorecard competency, its bar says how often the screens took it up. A short bar means the “(past interview)” brackets on that line deserve a second look before anyone counts them.
On outbound, AI Sourcing starts from one prompt a recruiter writes in natural language. Its search covers the open web, the candidates synced from your connected applicant tracking system, and the conversations your team has already recorded in Metaview. Its reasoning explains why each person came back. It leaves every gate unmarked, so a recruiter writes the page’s marks, bracket included, before any sourced name counts as a match.
On inbound, Application Review puts each application in one of four named fit buckets, with the reasoning for the placement and no numerical score. The buckets are Great, Good, Okay, and Poor. I’d lay that reasoning beside the page’s marks for the same applicant, and read the application in full wherever the two disagree. Accepting or rejecting the applicant stays a person’s decision.
The bucket comes from the role’s ideal candidate profile (ICP), which Application Review drafts from the job description and a recruiter reviews and signs off before use. From each accept and reject decision your team makes, Application Review suggests refinements to the ICP, and the candidate list is re-evaluated once someone approves one. I’d check each suggested refinement against the written rule before approving it, and carry any change the team accepts onto the page as well.
Take the first tally to the next intake call.
The first month’s trade tally shows the rule as each motion has been reading it, which no single shortlist can. Bring it to the next intake call for a similar role, open on the weighted line that filled the rule’s open slot most often, and ask the hiring manager whether that line should be doing so much of the work. Their answer is the first edit to the next role’s page.
Read a sourcing agent’s reasons against your gates.
In a demo, give AI Sourcing a prompt for one role, then set the reasoning on its first candidates beside the gates on your page before any of them counts as a match.
Frequently asked.
Should candidate matching use a points score?
It can, provided the gates sit outside the points. Award points only once every gate is met, and write the rule beside the total so the hiring manager knows which lines earned it. I’d still start with the sentence form, because it makes the team agree on the trade before anyone argues about a number.
How many lines should a match standard have?
Few enough that every reviewer can mark them from memory. I’d keep two or three gates and three to five weighted lines. Beyond that, lines start to overlap, and two reviewers file the same evidence under different lines.
Does AI candidate matching make screening fairer?
It makes screening more consistent, which is a different thing. A tool applies each line the same way to everyone. A line that stands in for background, like the name of a school, gets applied consistently too. Read every line for that before the standard goes into any tool.
Who should own the match standard for a role?
The recruiter running the role writes it and keeps it current, and the hiring manager signs it. Give it one owner even when several people source for the role, so two edits never pull it in different directions.
How do internal candidates fit a match standard?
Mark them on the same page as everyone else. They often show more lines than outside candidates, because their work happened inside the company. What their current manager says counts as evidence for a line, and it still needs a mark of its own like any other source.
