About

What if Zerve could identify a user’s path to success before success happens?

Zerve Early Warning System transforms anonymized event data into an actionable prediction system that answers: Which early behaviors predict sustained adoption, and how early can we detect them?

We define success as 2+ distinct active days during the following 7 days, then reconstruct user journeys and measure persistence, workflow breadth, activity, and product-area adoption.

The key discovery is striking: successful users don’t simply do more—they return and expand.

Users active for 4+ days show a 58.33% success rate versus just 0.45% for one-day users. Users engaging with 3+ product areas reach 11.99% versus 1.08% for 0–1 areas. The strongest observed combination— 4+ active days + 3+ product areas—reaches 68.75% success.

We then convert these behavioral signals into an interpretable Logistic Regression Early Warning Model. Across four chronological future periods, it achieves 0.405 mean PR-AUC and 0.904 mean ROC-AUC.

Operationally, the model is highly actionable: the top 10% of users achieve 15.48% success, delivering 6.08× baseline lift and capturing 60.94% of observed successes.

The result is more than a prediction model—it is a product-growth framework:

Observe → Measure → Predict → Explain → Act

Identify users at risk, understand their behavioral trajectory, and intervene before disengagement becomes permanent.