Every SaaS company tracks its funnel until the exact moment the funnel starts mattering. Marketing measures visits, campaigns, and signups with three tools' worth of precision. Then the user enters the product, where revenue is actually decided, and the measurement goes dark until the invoice either arrives or does not.
If your board deck has a detailed acquisition slide and a churn number nobody can explain, this is why. Here is how to instrument the whole journey.
Define the funnel in business terms first
Before any events, write the stages as sentences a founder would recognize:
- Visit: someone with the problem lands on the site.
- Signup: they create an account. Cheap, and it predicts nothing by itself.
- Activation: they experience the product's core value once. This is the stage that predicts everything.
- Paid: trial converts, card charges.
- Retained: they are still using it in month two, three, six.
- Expanded: more seats, higher tier.
Most teams instrument stages one, two and four, because those have obvious events. Activation and retention, the two stages that actually predict revenue, go unmeasured because they require a decision about what to measure. Which brings us to the only hard part.
Finding your activation metric
Activation is the early action that separates users who stick from users who evaporate. You do not pick it in a workshop; you find it in your data. The procedure:
- Take two cohorts from the last few months: users still active after 60 or 90 days, and users who vanished inside two weeks.
- List what each group did in their first week: features touched, objects created, invites sent, integrations connected.
- Look for the behavior with the biggest gap between the groups, that is achievable in the first session or two, and that plausibly causes the value rather than merely accompanying it.
Then be honest about what you have: a correlation. "Users who connected a data source stayed" might mean connecting causes commitment, or that serious users both connect and stay. You firm it up by pushing onboarding toward the candidate action and watching whether retention follows. If it does not, your correlation was a mirror, not a lever. Either result is knowledge.
One number, not five. A composite activation score is a way to avoid deciding.
The identity handoff, where funnels go to die
The marketing site sees an anonymous visitor. The product sees user 8317. Unless someone deliberately connects the two at signup, your funnel is two half-funnels that cannot be joined, and "which channel produces customers who retain" becomes permanently unanswerable.
The fix is mechanical but must be explicit: capture the marketing context (first touch, campaign, landing page) before or at signup, attach it to the account at creation, and set the same user ID in every tool that watches the product. Test the join like a feature, because it is one. When I audit SaaS setups, this handoff is broken more often than any other single thing, and it is why the acquisition report and the revenue report describe two different companies.
Instrument against the funnel, nothing else
With stages and activation defined, the event list writes itself: one event per stage transition, plus the handful of product actions that feed your activation definition. For most SaaS products that is ten to fifteen events, named consistently (convention here), each with an owner and a definition.
Resist the everything-tracker. Autocapture and "log all the things" produce a swamp that nobody queries, and the swamp's existence becomes an argument against ever cleaning it. Events chosen against decisions stay clean because every one has a reason to exist.
Tooling matters less than this structure. GA4 plus BigQuery can do it; PostHog and its peers do it more comfortably (the comparison); the deciding factors are in web analytics vs product analytics.
The weekly readout
A funnel that predicts revenue produces a short, recurring readout: signups by channel, activation rate by channel, trial-to-paid by activation status, and cohort retention. Four views. The third one is the money view: paid conversion split by activated versus not-activated is the clearest proof of whether your activation metric is real, and it converts onboarding from a design debate into an optimization problem.
If you cannot produce that readout today, the gap is almost never the tool. It is a missing activation definition, a broken identity join, or events nobody trusts. All three are findable in a week of honest audit work, and finding them is considerably cheaper than another quarter of arguing about whose churn number is right.