A hiring manager asks why strong candidates keep withdrawing after the final round. The answer is probably sitting in a dozen recorded interviews your team never finds time to rewatch, so it comes from whoever remembers those calls best, and the meeting moves on.

Interview analytics means putting that question to every captured conversation at once, instead of to the person with the best memory. I’d start with four query patterns, and I’d read every result as a list of conversations to open before anyone acts on it, never as a number to report upward.

Each pattern works on interviews your team has already recorded and transcribed, so you can run it this week, and each one ends in a decision a recruiting leader already owns.

Read every interview analytics result as a reading list.

Every query in this guide returns conversations: the interviews that match, sometimes with a count beside them or a line pulled out of each one. The count is the part people screenshot for a leadership update. I’d treat it as the length of your reading list, because a count of conversations says how many interviews matched and nothing about what happened in them.

Three rules hold for any query you run:

• Scope it to one kind of interview and one window of time. A recruiter screen and a technical round are different conversations, and a query across both blends them into a result that describes neither.
• Write down the decision before you run it. If no answer would change what you do next, skip the query and save the hour.
• Read before you act. Open a handful of the conversations at each end of the result before anyone outside the recruiting team hears about it.

A query run out of curiosity produces a chart, and teams tend to present that kind of chart once and then drop it. That’s why I’d hold the second rule hardest.

Find the interviews where the candidate barely spoke.

The first pattern asks which interviews left the candidate the least room to talk. Talk time is the share of the conversation each side spent speaking, and the useful version of the query sorts interviewers by how much of the floor they kept.

Run it because an interviewer who holds the floor for most of a behavioral round collects less evidence than the scorecard then asks them to rate. The debrief weighs that rating as if it rested on as much as everyone else’s.

Keep the comparison inside one interview format. A hiring manager selling the role on a first call should talk more than a panelist running a case exercise, and a ranking across both puts your strongest closers beside your weakest listeners.

Then read three transcripts first, from the interviewer at the edge, before you say a word to them. A long answer to the candidate’s own questions about the team looks exactly like an interviewer talking over the candidate on a talk-time chart, and only the transcript tells the two apart. The other edge deserves the same read: an interviewer who almost never speaks may be letting the candidate run for most of the hour without a single follow-up.

Use the result to coach one interviewer at a time, with the transcript open between you. A chart can point you at the interview worth discussing, and the talk that follows goes better with both of you reading the same passage.

A published talk-time benchmark from a large set of captured interviews gives you a reference point for where your edges sit.

Pull what candidates say about pay and competing processes.

The second pattern collects what candidates have already told your recruiters about money and about the other companies they’re talking to. Candidates often raise both in a recruiter screen, and what they say tends to stay in that recruiter’s notes.

For one role and one window, ask the record two things: what each candidate said they expect to be paid, and whether they mentioned being in another process. Keep the two answers side by side, because a candidate with a high expectation and two late-stage processes is a different conversation from one with the same expectation and no other options.

This query feeds two decisions. If most of the finalists for a role named a figure above your band, take the band to your compensation team before the next offer goes out. If several finalists said they were close to an offer elsewhere, shorten the time between your final round and your offer for that role.

Read every answer as what a candidate chose to share with a recruiter, which is often an opening negotiating position. A candidate who never reached a screen told you nothing, so the result is silent on everyone your sourcing never reached.

Search past conversations before you open a new search.

The third pattern finds people your team has already interviewed who talked about the experience a new role needs. I’d run it during the intake call, against the profile the hiring manager has just described, because that’s when the answer can still change the plan.

Ask for every past conversation in which a candidate described that experience, such as running a team of a certain size or closing a certain kind of deal, and look at how each of those processes ended. A candidate who withdrew for timing reasons last spring is usually a warmer first message than a profile from an outside search.

The same list also tests the hiring manager’s profile against your own history. If your team met several people with exactly that experience and advanced none of them, the profile and the interview panel disagree about what the role needs, and the intake call is the cheapest place to settle it.

Note where those candidates were working when your team met them, too. It’s a short list of companies where the experience exists, built from people who already agreed to talk to you once.

Treat every match as a snapshot. A conversation from eighteen months ago records what someone could do then, so the first message asks what has changed since.

A search like this finds only what the records hold. If it keeps coming back thin, a separate guide covers building a candidate database recruiters search first.

Ask one question across every final-round conversation.

Some patterns exist only across many conversations: the question finalists ask most often, or the concern that keeps surfacing before candidates withdraw. A single debrief sees one candidate at a time, so it can’t see either.

Pick one stage and one window, and ask what candidates raised. I’d start with the final round, because that’s where a candidate’s questions turn into a decision about you.

Two decisions come out of the answer. If the same question keeps coming up, the pitch is missing the answer, so rewrite the hiring manager’s pitch to cover it before the next final round. If one competency on the scorecard draws almost no questions in some interviewers’ rounds, those interviewers need help asking about it, and that’s a coaching conversation like the first pattern’s.

Read the conversations behind the top themes first, then the ones behind a concern that appears only once or twice. A rare worry raised by your two strongest finalists can outweigh a frequent one raised by people who were never likely to join, because how often a theme comes up says nothing about who raised it.

Where Metaview fits in interview analytics.

The first pattern starts in Metaview Reports, which compares talk time and question coverage across interviewers and roles.

1

2

3

1. 1“Coding,” one of four competency headings the counts sit under.
2. 2Ashley K.’s recruiter screen, with five questions on “Culture” against a single one on “Collaboration.”
3. 3“Comp expectations,” a saved report built from the salary figures candidates gave in recruiter screens.

A long talk time in that comparison shows you which transcripts to read, and reading them tells you whether there is anything to coach.

Siadhal Magos, Metaview’s chief executive, runs the second pattern on an engineering pipeline in Episode 2 of Metaview’s Interview Data Masterclass, narrowing it to candidates who said they were interviewing with other companies and then adding a column for the salary expectations each of them shared.

Both halves of that query read what candidates said in recorded interviews. When an interview is recorded, Metaview’s Notetaker takes part in it as a visible participant, leaving the transcript and structured notes a column reads from. A candidate can say no to being recorded, and your team’s own consent rules decide how that question is asked. An interview no one recorded gives a query nothing to search. Because the Notetaker captures every spoken word, a column can pull what a candidate said about pay even when no one wrote it down.

An earlier version of the Reports page showed a pay column partway through filling in:

1

2

3

1. 1A heading being written for the new column, with the cursor still after “Compensation Expectations.”
2. 2An answer with equity added to the salary, so open the call before you set it beside the others.
3. 3Kevin W.’s “Intro Chat” and the call below it, both waiting on an answer.

That column is an “AI column”: you describe the data point you want, and Metaview’s AI extracts it from each transcript in your account. For the other half of this pattern, the data point can be mentions of competing interviews, and a new column backfills across your full history before it fills in for every new conversation.

The third pattern starts from a filter. Reports filters hiring conversations by dimensions such as interview stage or hiring manager, and it can find conversations based on what was discussed. That earlier version of the page also showed the filter on account executive interviews and talk of large deals:

1

2

3

1. 1A filter that reads like a sentence: large deals discussed, in account executive interviews.
2. 2A yes from a recruiter screen, with the contract value the candidate mentioned.
3. 3Timothy S.’s answer describes the job he held when your team spoke to him, the snapshot the third pattern warns about.

Each of those views ends where this guide says it should, on a short list of conversations someone on your team then opens and reads.

Start AI Sourcing from the intake call: give it one natural-language prompt built from the profile the hiring manager described, and it searches past Metaview conversations together with the web and the candidates your connected applicant tracking system already holds. Every candidate it returns carries the sourcing agent’s reasoning, and the third pattern’s habit applies to it: read each reason against the intake notes, then decide who to approach. Approaching a candidate and assessing one both wait for a recruiter, since the agent won’t start either on its own.

The fourth pattern is a plain question, and the Metaview Assistant answers questions about your captured interviews. It can summarize what final-round candidates for one role kept asking and recall specific details from those interviews.

Joel Baroody, vice president of talent at Brex, says Metaview “has allowed me as a talent leader to gain insights across meta themes like compensation or common candidate questions.” Those two themes are the second and fourth patterns in this guide, and they’re the two a talent leader can bring into a planning meeting before a problem shows up at the offer stage.

In an interview with TechCrunch, Magos described where the company started:

Each of the four patterns is a way of making those interviews less opaque to the people who weren’t in the room, starting with the recruiting leader.

Tie the next round of queries to your hiring review.

Before your next hiring review, write the decision each query will inform at the top of the agenda: whether a role’s band needs to move, say, or whether a final-round pitch needs rewriting. Then run only those queries, and leave any query without a line on the agenda for a later review.

Open the conversations behind each result before the meeting, so the review starts from what candidates said and treats the count as the length of the reading list.