About

Every Zerve user leaves a trail. Most analyses ask what users do. UserDNA asks when, in what sequence, and what that sequence predicts.
This project reverse-engineers the behavioral trail of 4,774 Zerve users into a single Composite Momentum Score — a 0–100 signal built from four dimensions: upgrade conversion, deployment depth, retention cadence, and workflow complexity. A Random Forest model trained exclusively on first-48-hour events predicts that score with CV R² = 0.917, meaning user fate is largely determined before the end of day two.
The analysis surfaces four user archetypes — The Builder, The Explorer, The Wanderer, and The Ghost — each with a distinct behavioral fingerprint, churn risk score, and a single product intervention that would move them toward success. 82.5% of users are Ghosts. 46.1% have never made a single AI agent call. One collaboration event in week 1 lifts median success by 166%. These are not trends — they are cliffs.
The project was built entirely inside Zerve: Python and SQL notebooks for feature engineering and cohort analysis, Zerve's AI agent for archetype narration and methodology auditing, and a live deployed API endpoint that accepts any user's first-48-hour events and returns their predicted archetype, momentum score, and top 3 behavioral interventions in real time.
The core finding: Users who collaborate once in their first week and make at least one AI agent call score 2.7× higher on average than those who do neither — and both behaviors can be triggered by a single product decision made on day zero.