Models

The seven models behind Churned, what each predicts, what it needs and where it surfaces

Churned runs seven kinds of models. Each is described on its own page with the same structure: what it predicts, what data it needs, how it works, how often it runs, how to read the output, and where it surfaces in the journey.

Model	Predicts	Surfaces in
Health scoring	The probability that a customer with a subscription churns, expressed as a health score, risk score and level, with drivers	Understand health, Build audiences
RFM	Segments of one-off buyers or donors by recency, frequency and monetary value	Understand health, Build audiences
RFV	Segments of users by recency, frequency and volume of product usage	Understand health (SaaS users)
Audience clustering	Groups of customers that share a retention, reactivation, upsell or engagement opportunity	Build audiences
Sentiment	The sentiment of contact moments such as support tickets and emails	Understand health, Measure impact
Co-pilot	The next best action per customer, learning from outcomes	Automate with workflows, Optimise, Measure impact
Reactivation	The probability that a churned customer comes back	Build audiences, Measure impact

What they have in common

• Trained on your data only. Every model is trained per organisation on the history delivered by your integrations; nothing is pooled across customers.

• Refreshed daily, retrained periodically. Scores are recomputed every day from the latest data; the models themselves are retrained on a schedule and after data changes, with a check that the new model is at least as good as the current one before it replaces it.

• Explained, not just scored. Health, reactivation and Co-pilot outputs come with drivers or per-segment results, shown on the customer page and on Model Insights.

• Only as good as the connected data. Engagement, sentiment and RFV need behavioural or communication sources; without them those models and their metrics are hidden.