An onboarding funnel is not a conversion funnel
A conversion funnel ends at a transaction. An onboarding funnel ends at the moment a new user gets the thing they came for — the first report generated, the first teammate invited, the first payment collected. The two overlap, but they answer different questions, and measuring one while calling it the other is the most common reason onboarding numbers look fine while retention does not.
The practical difference is what counts as success. A signup is a conversion. It is not activation, and a funnel that stops at signup will tell you nothing about whether the product worked.
Define the steps from the product, not the org chart
Teams often build the funnel around internal stages — signup, onboarding, activation, adoption — because that is how the work is divided. Users do not experience those categories. They experience screens and actions.
Write each step as something a person does: created an account, connected a data source, saw a chart with their own data in it, returned the next day. If a step cannot be observed as an event, it is a label rather than a measurement.
Four to six steps, and no more
A funnel with twelve steps produces twelve small drops and no clear answer. One with three hides the failure inside a step too wide to act on. Four to six is usually where a funnel stays readable and still isolates the problem.
Keep the steps in a strict order that every user must follow. If two steps can happen in either order, they are one step, or they belong in separate funnels.
Where onboarding funnels usually break
Three failures account for most of it. The empty state: the product works but has nothing in it yet, so the first screen demonstrates nothing. The setup cliff: a step that requires credentials, a teammate, or a decision the user is not ready to make. And the unmarked finish line: the user completes setup and is never told they succeeded, so the value never registers.
Each leaves the same footprint in the data — a single step where the drop is much larger than its neighbours. Open the recordings behind that step before theorising about why.
Measure time, not only rate
Two products can activate 40% of signups and be in completely different health. One does it in ten minutes; the other takes nine days, by which point most of the intent that brought the user has evaporated. Track how long the funnel takes to complete alongside how many complete it, and treat a rising time-to-value as a warning even when the rate holds steady.