Skip to content
All writing
Product5 min read

What we got wrong about onboarding

We spent a quarter making the first five minutes beautiful. The number that moved was in week two, and it was a documentation problem.

Maya Okonkwo

Co-founder & CTO

Our activation rate was flat for two quarters, and we were confident we knew why. The first setup took too long. So we built the thing every team builds: a guided flow, sensible defaults, a sample workspace, a first result that was useful in under four minutes.

It worked exactly as designed. Time-to-first-result dropped from nineteen minutes to four. Activation moved by less than a point.

The drop-off was not where we were looking

When we finally looked at the cohort curve properly rather than the funnel we had built, the shape was obvious and had been for months. Almost nobody dropped out during setup. They dropped out between day four and day nine, after the sample worked and before their own thing did.

We had optimised the part of the journey we could see, because it was the part we had instrumented, because it was the part we had built.

The second one is the hard one

The first is a guided path with a known answer. The second is the user's actual problem, with their own systems, their own edge cases and their own idea of what correct means. Everything we had built helped with the first and nothing helped with the second.

  • Twelve of the fifteen support conversations in that window were 'how do I do X', not 'X is broken'.
  • The most-visited page in the docs had no code sample on it.
  • Four of our six most common integrations had no worked example at all.

What actually moved the number

Nine worked examples, each one solving a real task end to end, with the integration configured and the failure cases handled. Not snippets. Whole things a person could copy, run, and then edit into their own problem. That shipped over five weeks and moved activation eleven points.

The guided onboarding was not wasted — it is still there and people still use it. But it solved a problem we had assumed rather than measured, and the measurement was available the whole time.

Keep reading

  • Engineering

    A log is not an explanation

    A complete log tells you everything the system did and nothing about why it did it. The gap between those two is where the hard afternoons go.

Give the queue to the agents.

Connect one system, run a week in shadow mode, and look at the numbers. If they don't move, you've lost an afternoon.

Free for 14 days · No card required

Live demo of Vantis · see all five brands