Predictive audiences

Predictive audiences score every customer on three dimensions: churn risk, purchase propensity, and LTV bucket. Use these scores as segment conditions, journey branches, or reporting dimensions.

Scores

Score	Range	Interpretation
Churn risk	0.0 – 1.0	Probability the customer will lapse in the next 30 days.
Purchase propensity	0.0 – 1.0	Probability of a purchase event in the next 14 days.
LTV bucket	low / medium / high	Predicted lifetime-value quintile relative to your workspace.

How scoring works

The baseline algorithm uses:

Recency of last event (last_seen_at).
Count and recency of purchase events.
Total monetary value of purchases.

Scores are updated on demand from the Predictive audiences tab of Trust & Premium, or nightly for Business-plan workspaces.

Worked example

A customer who ordered 3 times in the last 30 days → low churn, high propensity, medium or high LTV.
A one-time buyer last seen 90 days ago → high churn, low propensity, low LTV.

Using predictions

Suppress discounts from high-propensity customers ("they'll buy anyway").
Focus win-back on high-LTV + high-churn combinations ("can't lose them").
Report ROI segmented by predicted LTV bucket.

Predictions are directional, not guarantees. Use them as tie-breakers between lookalike cohorts, not as the sole gate on critical sends.