B2B Funnel Metrics: Definitions That Hold Up Under Pressure

Ask four people in a B2B company what an SQL is and you get four answers. The SDR says a lead that booked a meeting. The AE says a meeting that happened and had budget in it. Marketing says anything that passed the score threshold and got routed. RevOps says whatever the picklist says, and the picklist has nine values, three added by someone who left in 2024.

None of them is wrong. There is no industry definition of MQL, SQL or SAL, and waiting for one wastes a quarter. What matters is whether yours is written down, dated, owned and computed in one place.

The definitions are arbitrary, the drift is not

Inconsistent funnel terminology is not an embarrassment. Definitions are local conventions, like a fiscal calendar, and a good local convention beats a bad universal one.

Drift does the damage. Someone adds two behaviours to the scoring model in March; in April the MQL number is up 30 per cent and three people explain it as campaign performance. The trend line crosses a silent redefinition, and every decision from that chart compares two different things.

So treat definitions as infrastructure, not documentation. Reporting, routing, forecasting and compensation sit on top of them, and a change to one is a schema migration, not a wording tweak.

The five-part template

A definition that survives a real organisation has five parts.

Entry criteria. The machine-checkable condition that makes a record become this thing. Not "shows buying intent" but "score above 70, or requested a demo, or attended a live product session, and matches ICP firmographics". If an engineer cannot turn it into a query, it is not a definition.

Exit criteria. What removes the record from the stage, in every direction: promotion, disqualification, decay. Decay is the part nearly everyone omits, and it is why stage counts inflate forever. An MQL with no expiry is a permanent MQL.

Owner. One named human, not a team: the person who approves changes and answers edge cases.

Where it is computed. One named system: the warehouse view, the CRM report, the signal ledger. A metric computed independently in two places is not one definition but two implementations, and they will diverge on a Tuesday.

What it is used for. Routing, board reporting, agency compensation, forecasting. This is the change control: knowing a definition feeds a bonus tells you how carefully to alter it and who has to be in the room.

Add a version and a date, and keep the superseded ones: an undated definition cannot be reconciled against last year's chart.

A count without a denominator is a press release

"We generated 640 MQLs" contains almost no information. Out of how many, converting at what rate, against what last quarter? The count rises when the market improves, when spend goes up, when the threshold drops, or when a list is imported. Only the ratio separates those.

Every volume metric should carry its denominator in the same visual element: MQLs with MQL-to-SQL rate, opportunities per target account, meetings per thousand contacted. The denominator is what makes a number falsifiable, and an unfalsifiable number cannot support a decision on a command center screen. The same goes for populations: "win rate" must say of what.

Cohort by entry month, not close month

Take every deal that closed in Q3 and look at where the leads came from. That is a close-month cohort, weighted toward whatever produced fast-closing deals, and it says nothing about leads generated in Q3, most of which have not resolved yet.

Now take every lead that entered in January and follow it forward: how many became opportunities, how many closed, over how long. That is an entry-month cohort, the only view that measures what marketing did in January.

The two disagree systematically. Close-month cohorts flatter short-cycle channels and make strategy changes look instantaneous. Entry-month cohorts show the real conversion shape and are always incomplete for recent months, which people find uncomfortable. Mark cohorts younger than your median sales cycle as still maturing, rather than letting a reader take the last bars for a collapse.

A ten-minute check. Pull last year's leads by entry month and compute, for each, the percentage that became an opportunity within 90 days. If that moves more than a few points month to month without a known cause, either lead quality is unstable or the definitions changed. Both are worth knowing before the board meeting.

Deduplication decides what a number counts

Three people from one company download three assets. Three MQLs or one engaged account? Both are defensible, and the count differs threefold depending on the choice, so pick one and write it down. In account-based motions the account is the honest unit, with person-level counts as a secondary measure of committee depth.

Then the plain duplication problem: one human as three contact records, from a work address, a personal one and a misspelling. Email-only matching misses these. Matching on company domain plus normalised name catches most; the rest need a rule for which record survives and what happens to the activity on the others. Some records will still merge wrongly. The goal is not a clean database, which does not exist, but a consistent rule, so the number means the same in March as in September.

The review that stops it rotting

Definitions decay because the business changes and nobody updates the text. A quarterly review, one hour, is the cheapest maintenance there is.

Four things happen in it. Each owner confirms or amends their definition. Every amendment gets a date and a note, and any chart spanning the change gets an annotation. Definitions nobody used that quarter get deleted, because unused ones are where contradictions hide. And any metric that appeared in a report without a definition either gets one or comes out.

What breaks without it is predictable. The scoring model drifts and MQL volume stops meaning anything. Two teams build competing versions of pipeline and the meeting becomes an argument about arithmetic. New hires inherit conventions nobody can explain and invent their own. The board sees a trend line spanning three silent redefinitions. None of it arrives as a crisis, which is why it persists: an audit usually finds three or four such changes in eighteen months, unrecorded.

Where written definitions stop helping

They do not resolve genuine judgement. Whether an account is a fit is an assessment no entry criterion captures fully. Over-specify and reps game the fields; under-specify and you get inconsistency. Most mature teams settle on machine-checkable criteria plus a documented human override with a logged reason.

They do not fix thin data. If forty leads a month enter the funnel, precise stage definitions will not make conversion rates stable, and governance around numbers that are mostly noise is ceremony. At that volume, count opportunities and name accounts.

They cost something to keep, too. Companies that write two hundred definitions maintain none. Twelve that are current, owned and dated beat a dictionary nobody has opened since it was written.

Frequently asked questions

What is the difference between an MQL, an SQL and an SAL?

There is no universal answer, and that is the honest starting point. By common convention, an MQL is a lead marketing judges qualified enough to hand over; an SAL is one sales has accepted as worth working; an SQL is one sales has worked and confirmed as a genuine opportunity. What matters is not the convention you adopt but that each stage has written entry and exit criteria, a named owner, one place where it is computed, and a date on the current version.

How should a funnel metric be defined so it does not drift?

Write five things for each metric: machine-checkable entry criteria, exit criteria including decay, one named owner, the single system where it is computed, and what it is used for downstream. Version and date it, and keep superseded versions so historical charts reconcile. The part most often missing is decay, without which stage counts inflate forever.

Why should B2B funnel metrics be cohorted by entry month?

Because an entry-month cohort measures what marketing actually did in that month, while a close-month cohort measures deals that happened to finish recently. Close-month views flatter short-cycle channels and say nothing about recent activity, most of which has not resolved. Entry-month cohorts show the true conversion shape, at the cost of being incomplete for any month younger than the sales cycle, so mark those as still maturing.

What happens if nobody reviews metric definitions?

Definitions drift silently. Scoring models get adjusted, picklist values accumulate, and lead volumes move for reasons nobody records, so trend lines end up comparing different things. Teams build competing implementations of one metric and meetings become arguments about arithmetic. An hour of review each quarter prevents almost all of it: each owner confirms or amends their definition, amendments are dated and annotated on affected charts, unused definitions are deleted, and undefined metrics either get a definition or come out of the report.