[drop_cap]B[/drop_cap]y this stage you have a funnel. Visitors land, some sign up, some of those activate, some of those pay, some of those stay, and a few expand. Every stage has been built with intent — the trial length, the onboarding checklist, the pricing page, the win-back email. What none of that guarantees is that you know which piece is actually holding the business back.
This is Stage 10, the closing chapter, and it is not a new lever to pull. It is the discipline that tells you which of the levers already built is worth tightening next, and which ones to leave alone. A SaaS company that never measures its own funnel is running on instinct about which feature request matters most, which trial-length change will move the needle, which churn cohort to chase — and instinct, applied to six stages at once, mostly guesses wrong.
The Funnel, Stage by Stage
A SaaS funnel has six seams, and each one is a division: how many made it through, divided by how many arrived.
Visitor → Signup: signups ÷ visitors = X%
Signup → Activated: activated ÷ signups = X%
Activated → Paid: paid ÷ activated = X%
Paid → Retained: retained-at-90-days ÷ paid = X%
Retained → Expanded: expanded-seats-or-plan ÷ retained = X%
Fill in your own X at each seam, for a fixed cohort and a fixed window — not “this month’s average,” a specific group of people who all crossed the same line in the same week. “Activated” needs its own definition before this table means anything: the action a user takes that correlates with staying, whatever that is for your product — importing their first dataset, inviting a teammate, hitting a specific screen three times in seven days. Write that definition down once, put it in your product analytics, and never redefine it mid-quarter, or every number downstream stops being comparable to the number before it.
Finding the Binding Constraint, Not the Busy One
The busy stage is the one everyone’s already looking at — usually paid conversion, because it’s the one with a dollar sign attached. The binding constraint is the stage that, if you fixed it, would move the whole business. Those are not always the same stage, and confusing them is the single most common way a SaaS team burns a quarter improving a number that was never the problem.
Here’s a worked example, filled with placeholder numbers you should replace with your own:
Say 1,000 visitors sign up at a 10% rate, giving X = 100 signups. Of those, 20% activate — X = 20 activated users. Of those, 50% convert to paid — X = 10 paying customers. Of those, 90% are still paying at day 90 — X = 9 retained. That’s a lot of stages, and it’s tempting to fix all five at once.
Don’t. Instead, ask what happens if you move each seam by the same relative amount — say, a 20% lift at each stage, one at a time, holding the others fixed — and compare the effect on the final number. A 20% lift in signup rate (10% → 12%) still starts from 1,000 visitors and still loses most of them to the same downstream leaks; it moves the final paid-and-retained count only a little. A 20% lift in activation (20% → 24%) moves the same downstream math from a much smaller base but a stage that’s already leaking heavily, and the effect compounds through every gate after it. Run this arithmetic on your own numbers before you decide what to fix — the stage worth fixing is whichever one, moved by the same percentage, produces the biggest change at the bottom of the funnel. It is very often not the stage with the lowest raw percentage. A stage can convert at 90% and still be the binding constraint, if almost nothing reaches it; a stage can convert at 20% and be irrelevant, if it’s already fed by so much volume that a leak there barely matters downstream.
Leaky vs. Small — Two Different Diagnoses
A weak conversion rate and a weak absolute count are not the same problem, and they call for opposite fixes.
A leaky stage has healthy volume arriving and a bad rate getting through — plenty of signups, few of them activating. The fix lives inside the stage: the onboarding flow is confusing, the first-value moment is buried, the checklist asks for too much before showing anything back. You improve the experience of the stage itself.
A small stage has a fine conversion rate but almost nothing arriving to convert — activation looks strong because the ten people who signed up were highly motivated, but ten people is not a business. The fix lives upstream, at whichever stage is failing to deliver volume, not inside the stage that’s merely starved. Fixing a small stage’s own experience changes nothing, because the experience was never the problem — the fix is fixing acquisition or top-of-funnel filtering, and letting the healthy conversion rate downstream do its job on more people.
Read both numbers — the rate and the raw count — every time you look at a stage, or you will “fix” a stage that was never broken.
NRR and Gross Retention — Definitions, Not Targets
Two numbers get treated as universal benchmarks that they were never meant to be. State them as what they measure, and set your own bar.
Gross retention is the share of starting revenue you keep from an existing cohort over a period, counting only losses — downgrades and churn — never expansion. It caps at 100%. It tells you how well you hold onto what you already have.
Net revenue retention (NRR) takes the same cohort and adds back expansion — upsells, seat growth, plan upgrades — so it can exceed 100% if your existing customers grow faster than they shrink. NRR above 100% means your revenue would grow even with zero new signups; below 100% means the business is a leaking bucket that new signups have to keep refilling just to stand still.
Neither number has a law-of-the-trade threshold that applies to every SaaS business — your churn profile, contract length, and expansion motion are yours alone. What matters is that you compute both, on the same cohort, on a fixed cadence, and that you know which one your growth actually depends on this quarter.
Cohort Reading, Not Monthly Averages
A monthly average blends customers who signed up last week with customers who signed up eleven months ago, and reports one number that describes neither. If your activation rate has been improving steadily, a monthly average hides the improvement inside a lump that also contains the bad old numbers. If it’s been getting worse, the average hides that too, diluted by better history.
Read by cohort instead: group customers by the week or month they entered a stage, and track how each cohort’s number moves at a fixed interval past entry — day 7 activation, day 30 retention, day 90 expansion. Line the cohorts up next to each other and the trend becomes visible in a way a single blended number never shows. This is also the only honest way to answer “did the change we shipped in March actually help” — compare the March cohort’s curve to February’s, not this month’s average to last month’s.
Test Discipline for Small Numbers
If your business runs a hundred signups a month, you will rarely have the volume to reach statistical significance on a funnel test within a reasonable window, and waiting for significance means waiting months per test. That is a real constraint, not a reason to skip testing.
Design the test the same way regardless: one variable, one metric, a defined window, a pass/fail bar set before you look at the result. But be honest about what a small sample can tell you — a 100-signup-a-month business testing an activation change should expect to make a judgment call, informed by the direction and size of the shift, rather than a statistically airtight verdict. Set the bar in advance (“this needs to beat the trailing period by a clear enough margin that a reasonable person looking at the raw numbers would agree it moved”), watch qualitative signal alongside the number — did support tickets about the same friction point drop, did users mention the change unprompted — and accept that you are trading certainty for speed. That trade is often correct at this volume. Waiting for a sample size you’ll never reach is not caution, it’s a decision by inaction.
Back to Stage 0, With What You Now Know
You started this playbook at Stage 0 not knowing where your funnel leaked. You now have six seams measured, a method for finding which one is binding rather than which one is loudest, and a cadence for testing one change at a time. That is the loop: measure the funnel, find the binding constraint, run one disciplined test, and come back here next cycle with a new set of numbers. Refine is not the last stage because the work is finished — it’s the last stage because it’s the one that sends you back to Stage 0 with better information than you had the first time through.
AI Earns Its Place Here
This is exactly the kind of recurring, structured calculation a model handles well and a founder has no time to redo by hand every week. Feed it your six-stage numbers by cohort, your current NRR and gross retention, and ask it for two things: the binding-constraint arithmetic above run against your real figures, and a ranked list of which seam moving by an equal percentage would produce the biggest change at the bottom of the funnel.
Treat the output as a first draft, not a verdict. The model can do the division and rank the levers; it can’t tell you whether a dip in activation was a real product problem or a batch of low-intent signups from a conference booth. Read its ranking, check it against what you know happened on the ground that period, and then pick the one test that runs next cycle. The math is generated. The judgment about what to test next is still yours.