Predictive customer intelligence for a national subscription business

The problem

The business had years of customer, billing, engagement, and support history sitting in the warehouse, but no consistent way to act on it. Retention motions were reactive, upgrade conversations were untargeted, and the “right customer at the right moment” signal lived in the heads of a few veteran account managers.

The approach

We built a predictive intelligence layer on top of the warehouse that scores every customer continuously and surfaces three things to the right team, at the right time: who’s likely to upgrade based on engagement and account fit, who’s drifting toward churn before they actually leave, and which lifecycle moments (renewal, plan change, support escalation) are about to happen. The signals route into the team’s existing CRM and outreach tooling, so the action happens where the work already lives.

Architecture

[ Architecture sketch placeholder, replace with diagram when ready. ]

• Foundation: Built on top of the existing customer, billing, and engagement tables in the warehouse.
• Upgrade model: Calculates likelihood-to-upgrade per customer using engagement, plan fit, and account-level signals.
• Churn model: Flags drifting customers based on engagement decay, support patterns, and billing signals, before the cancel button gets clicked.
• Lifecycle layer: Surfaces renewal windows, plan-change opportunities, and service-load anomalies on the cadence each team needs them.
• Delivery: Signals pushed to CRM and outreach tooling, with confidence scores so teams know which signals to act on first.

Results

• Daily customer-level signals surfaced to retention and account teams.
• Upgrade conversations now happen because the data says so, not because someone happened to remember.
• Churn risks identified weeks earlier than the previous reactive process caught them.

Stack

Warehouse-native scoring (Snowflake + dbt), Python for model training and inference, integration with CRM and outreach tooling.