Every TA leader I talk to in 2026 can name the five stages of their recruitment funnel. Most can recite the conversion rates between each stage off the top of their head. Far fewer can tell you, with any specificity, why a conversion moved last month. That gap, between tracking a funnel and explaining it, is the most expensive thing in recruiting right now. It is also the gap that lets the same leaks keep happening, quarter after quarter, under different names.

The audit on the original version of this article asked us to do one specific thing: stop describing the funnel in the abstract and start describing the data layer underneath it. We agree. The five stages are well known. The five places they leak are well known. The part almost no team has wired correctly is the structured interview signal that explains what changed at each stage, in time to do something about it. That is the layer this guide is about.

What you will get below: how the recruitment funnel actually measures (and the thing it hides), the five stage-by-stage leak patterns we see most often, how to map your funnel without losing two weeks to ATS exports, a side-by-side comparison of manual, generic AI, and Metaview at each stage, the feedback loop that closes the gap between ‘conversion dropped’ and ‘this is why,’ and a 30-60-90 instrumentation plan a TA leader can hand to a hiring ops partner on Monday.

What the recruitment funnel actually measures (and what it hides)

Define the recruitment funnel in functional terms and the picture is the same on every team. It is a structured pathway candidates move through, from application or outreach to hire, with a measurable conversion rate at each handoff. Sourcing into screening, screening into interviewing, interviewing into offer, offer into hire. Every applicant tracking system draws it the same way. Every TA leader has the chart somewhere on a dashboard.

What the chart measures is movement. A candidate left stage 2, a candidate entered stage 3, and the ratio of those two numbers is your conversion rate at the screen step. What the chart does not measure is anything about the substance of that movement. Why did this candidate move and that one not? Who asked the questions? Were they the right questions for the role? Did the panel hear the same answers? Was the rejection driven by the rubric or by the interviewer’s memory of how the call felt?

That is the gap the auditor on the original draft flagged. The earlier version described the funnel in stages and offered generic AI advice for each. The actual question a TA leader has is more specific: when conversion at stage 3 drops six points month over month, which two interviewers, on which competency, gave us the new pattern? You cannot answer that with the funnel chart alone. You answer it with the interview signal underneath.

The case for instrumenting the layer below the funnel is now data-backed, not hypothetical. According to Metaview’s 2026 AI & Hiring Alignment Report - surveying 505 recruiting leaders and hiring managers across North America and EMEA, 85% of companies exceeding their hiring goals use AI in hiring, and the gap between AI-core teams and no-AI teams shows up most clearly at the start of the funnel. Only 49% of searches at non-AI teams start with high alignment on requirements; that delta cascades into every stage that follows.

The 5 stages and the 5 places they leak

Funnels leak the same way across most teams. The labels change, the numbers vary, but the patterns are remarkably consistent. Here is the stage-by-stage map of what we see most often, paired with the data signal that makes each leak visible before it shows up as a missed quarter.

Stage 1: Sourcing

The leak: top-of-funnel volume is fine, but the share of qualified candidates is below expectations. The cause is rarely sourcing tactics. It is misalignment between what the job ad says, what the recruiter is screening for, and what the hiring manager will actually accept on a debrief. The data signal that exposes it: structured comparison of intake-call requirements against screen-stage rejection reasons. If rejections cluster around criteria that were not in the intake notes, the funnel is leaking at the spec, not the source.

Stage 2: Screening

The leak: screen-to-interview conversion looks healthy until you check it by recruiter. One recruiter is converting at 12%, another at 38%. The chart averages to 24%, and the team-level funnel report does not catch the variance. The data signal that exposes it: screen-call notes structured against the same rubric across the team. Without rubric capture, you cannot tell whether the high-converting recruiter is finding hidden gems or shipping marginal candidates forward.

Stage 3: Interviewing

The leak: interview-to-offer conversion swings stage over stage and quarter over quarter, and nobody can explain it. The most common pattern is that panels are running different versions of the same interview. One interviewer asks for live code, another asks for past experience, the third asks for behavioural examples. They are all assessing ‘the engineering bar,’ but they are not pulling the same signal. The data signal that exposes it: interviewer-level reports against the question rubric. When one interviewer is asking three of the seven competencies and another is asking five, the conversion drift becomes a coverage problem, not a quality problem.

Stage 4: Offer

The leak: offers go out, candidates take longer than expected to respond, and acceptance rates drop without an obvious cause. The pattern is almost always debrief drift. The hiring manager forms the offer recommendation from memory of the panel, not from the recorded signal of who said what. By the time the offer letter goes out, the framing has subtly shifted, and the candidate hears a pitch that does not match the conversation they had. The data signal: post-interview multi-source summaries that the hiring manager can pull from at offer time, instead of recall.

Stage 5: Hire-to-Quality

The leak everyone admits has no good data: do the people we hired actually do the job we hired them for. The connection between funnel performance and hire quality is usually broken, because the panel signal that produced the offer is unrecoverable six months later. The data signal: the same structured interview record, queryable when a hiring manager flags a performance pattern. The team that can ask ‘show me the panel notes for everyone we hired into Sales Engineer in 2025 who did not survive ramp’ has closed the loop. Most teams cannot.

How to map your funnel without losing two weeks to ATS exports

Most funnel-mapping exercises drown in ATS export work. The data lives in three different objects (candidates, applications, interview events), the joins are inconsistent, and the conversion math has to be redone every time someone renames a stage. By the time the spreadsheet is clean enough to read, the question the TA leader was trying to answer has aged out.

Here is the version of the mapping exercise that does not lose two weeks. Define the five stages first, the way your team actually uses them, not the way the ATS labels them. Pull conversion rates at each handoff for the last 90 days. For each stage, identify the single most important signal capture point: intake notes for stage 1, structured screen notes for stage 2, panel transcripts for stage 3, panel debrief notes for stage 4, the same panel record retrieved at 90 days for stage 5. That is the audit map.

The shortcut on this exercise is to skip the months of spreadsheet engineering and start at the signal capture layer instead. If every interview is being captured at the stage it happens, the funnel report builds itself out of the interview data, instead of requiring a quarterly archaeology project. That is the part of the workflow most teams have not turned on yet, and it is also why the original draft of this article called the AI section ‘automate’ instead of ‘instrument.’ They are not the same thing.

Manual, Generic AI, and Metaview at each stage

The auditor on the original draft specifically asked for a side-by-side comparison table, and they were right. The five-stage breakdown reads as three different workflows depending on whether you are running it manually, with a stack of generic AI tools, or with Metaview as the structured signal layer. Here is the comparison.

The pattern under the table is the boring one. Manual is fine when volume is low and the team has senior interviewers everywhere. Generic AI helps at specific moments, but it does not connect them. The Metaview argument is the same signal at every stage, every interviewer, queryable later. The compounding is in the connection, not in any one cell.

Closing the feedback loop: stage data should drive stage decisions

A funnel report that does not change behaviour is a screenshot. The point of the data layer is to close the loop between ‘conversion at stage 3 dropped’ and ‘we ran a coaching conversation with the two interviewers who account for it on Tuesday.’ The feedback loop is what makes the funnel a learning system instead of a periodic disappointment.

Siadhal Magos, our co-founder, made this point cleanly in a recent short video: conversations are becoming the differentiator in recruiting, and the recruiters who can read what is happening inside a conversation in time to act on it will out-recruit the ones who cannot. The funnel sits downstream of those conversations. If you can read them, you can read the funnel.

The loop has three pieces. First, the data captured at stage has to be specific to the question a TA leader will ask later. A conversion percentage on its own does not pass that test. Interviewer-level coverage against a rubric does. Second, the data has to be available before the quarterly review. Weekly is the cadence we see working at Metaview customers, because that is the cadence at which an interviewer can actually change behaviour. Third, the action has to be at the interviewer level, not the funnel level. Funnels do not change behaviour. Coaching conversations with named interviewers do.

Catawiki’s TA team described this shift cleanly when they wrote about how Metaview Reports changed the way they advise hiring managers. The reframe is small, but it is the whole game: from feelings about the panel to evidence from the conversation. That is what closing the loop sounds like in practice.

Once the loop is closed, a conversion drop at stage 3 stops being a mystery. It becomes a specific question with a specific owner. That is the difference between a funnel that gets reported on and a funnel that gets fixed.

A 30-60-90 instrumentation plan

A TA leader does not need a six-month project to wire the data layer underneath the funnel. The 30-60-90 plan below is the version we see working most often on Metaview customer rollouts, and it is the version I’d hand to a hiring ops partner on a Monday morning to get started.

1. Days 1 to 30: turn on live interview capture across every interview at the screen stage and the panel stage. Confirm the rubric for each role is loaded into the capture surface, so the AI Notes feed the right competencies. By day 30, every interview is producing structured data, and the back-loaded transcription project goes away.
2. Days 31 to 60: wire Application Review onto inbound for the two highest-volume roles. The ATS gets autofilled scorecards from the panel notes. Pull the first 60-day funnel report off the new data layer and compare it to what the old ATS export was telling you. The differences are the leaks the old report was hiding.
3. Days 61 to 90: switch on interviewer-level dashboards. Run the first two coaching conversations using interviewer reports as the agenda, not recall. By day 90, the feedback loop is closed: stage data is driving stage decisions weekly, and the funnel report is being used to fix things, not just to read them.

That sequence is the cheapest insurance you can buy against the five leaks above. Each phase produces a specific artefact, each artefact is the input the next phase needs, and at the end the funnel is something you can read in real time rather than autopsy in a quarterly review.

Is hiring based on gut feel quietly breaking your funnel?

Deel used Metaview Reports to turn structured interview conversations into behavioural funnel data. Result: a redesigned GTM hiring process and sharper, evidence-backed hiring decisions instead of surface-level interview impressions.

LinkedInMetaview Technologies