The AI assistant went unnoticed. We found out why
A large CRM for sales teams. We set up Amplitude, walked the journey step by step and found two causes of churn: people were stalling during setup, and the AI assistant sat off to the side
Diagnosis and changes to the user flow
To the first insights out of Amplitude
Mapped the product's key flows
Growth in weekly use of the assistant
The numbers grew. The causes of churn stayed invisible
The company could see sign-ups, subscriptions and activity. What it could not answer were the three questions that decide whether a user stays or leaves
- Sign-ups and subscriptions climbing
- Overall activity looking healthy
- By the metrics, the product looked fine
- Where do users get stuck during setup?
- Why does almost no one open the AI assistant?
- Which features actually keep people?

We measured every step users took
Not a sum of clicks but the real routes: where people backtrack, where they lose the next step, where they leave for good
180+ actions · funnels · paths · return visits · first insights in 2 weeks

We found the step where setup was abandoned
Most users left while importing contacts and deals, not across onboarding as a whole
Half dropped out importing contacts and deals, one long form
Split the import into short steps with auto-mapped fields, so no one quit halfway
More users finish setup and reach their first result 90-day cohort against the previous quarter
Prompts moved into the deal card
Amplitude showed the assistant living off the main routes. You could reach it, but only if you went looking

Prompts and a ready-to-send email right inside the deal card, not in a separate section you have to open on purpose
Only a handful found the assistant: it sat away from the deal card
Built the prompts into the flow, and replaced the empty field with ready-made questions about the deal
The assistant is used far more often Weekly active use of the feature, 60 days
Return visits doubled
We counted return visits weeks after launch, not the spike right after it, that is where a durable effect shows

What we did
Set up Amplitude
Mapped the product's key flows so we could see them in numbers
Ran the diagnosis
Rebuilt the real paths and found the exact points of loss
Changed the flows
The AI assistant and setup: from the causes we found, not by eye
Verified the result
Compared how users behaved before and after the changes