Curtis McCrady

Updated Mar 3, 2026
← Back to Product Log

Multi-tab Session Replay

Fullstory logoFullstoryApril 2023

Problem Summary

New users didn't trust the AI assistant enough to complete setup. They were asked to connect their codebase and configure settings before experiencing any value. This 'work before reward' flow led to 77% drop-off in the first session.

Opportunity

How might we demonstrate AI value before asking users to do work?

Solutions

Solution Image 1
Redesigned action nodes to include aggregate run data and object creation (e.g. conversations) data. Improved the run history to be filterable by each specific action.
Solution Image 2
Updated the ouput of individual runs to be consice and account for every possible edge case so users are not left in the dark about what happened and why.