Product analytics tools show you funnels you define in advance. But users take paths you never anticipated. They discover workarounds, skip steps you thought were mandatory, and drop off at points you considered frictionless. Understanding the actual journeys users take, rather than the ones you designed, reveals where your product truly works and where it does not.

The AI Approach

An LLM analyzes sequences of user events to identify common journey patterns, cluster users by behavior, and highlight paths that correlate with success (conversion, retention) or failure (churn, support tickets).

Three-Step Build

  1. Extract journeys - Pull event sequences for a cohort of users from your analytics platform. Represent each user’s session as an ordered list of page views, clicks, and actions.
  2. Cluster and summarize - Feed journey data to an LLM. Ask it to identify the most common paths, unusual paths that deviate from the expected flow, and points where users seem confused (rapid back-and-forth navigation, repeated failed actions).
  3. Insight generation - For each identified pattern, the LLM explains what it suggests about user intent and recommends product changes to address friction points.

Where It Breaks

High-traffic products generate more event data than an LLM context window can handle. Pre-processing to cluster and aggregate journeys before LLM analysis is essential. The AI infers intent from behavior, which is speculative - validate insights with actual user research.

The Production Path

Run weekly analysis on key user cohorts (new users, churned users, power users). Integrate findings into product planning. Track whether product changes informed by journey analysis actually improve the metrics they targeted. Build a library of journey patterns specific to your product.