3 comments

  • roncron 2 days ago
    This sounds great in theory. It's been my job to review screen recordings of user sessions and it's often inconclusive.

    You can diagnose users not using some features or encountering bugs,, but churn seems out of reach. for example, a champion leaving/being laid off, budget cuts, lack of functionality, etc.

  • thsvrrck 2 days ago
    Hey, great idea. Main question that I have: how do you let an LLM determine causality?
    • kiranjohns 2 days ago
      Great question! To be clear, the LLM isn't discovering causality. What I do is feed it structured event data and session replays. Then we cross-check those findings across multiple sessions to filter out noise. The advantage here is that it dramatically reduces the time you’d otherwise spend guessing or digging through raw replays.
  • skyzouwdev 2 days ago
    [dead]