Sellrise

An online casino operator was doing retention on cohort blasts. Player events sat in the platform's own database, unusable for real-time decisions. VIPs went silent before anyone noticed, bonuses landed on the wrong cohorts, and the analyst team could not build the picture from raw data fast enough to act.

Built the retention engine from the event layer up. Redpanda ingests millions of events per day into ClickHouse. Per-player scoring runs continuously and produces a VIP desk worklist, a fraud queue, and a reactivation channel. Bonus economics are computed against actual player LTV, not cohort averages. LLM call analysis converts recorded conversations into structured signals. Reactivation chains fire per player based on the model, not a schedule. Later rebuilt as a multi-tenant SaaS - Revenue Autopilot / Retivo - for subscription businesses.