Reach forecasting for TV & OTT inventory

Problem

Reach forecasts were built by hand, with no statistical or ML methods. Labour-intensive and inaccurate; most forecasts were never updated because capacity ran out.

What I did

Initiated, raised resources for, and built a forecasting engine on the new unified data foundation.

Result

Accuracy up 8%: latency from 6 months to 1 day, 2 FTEs freed. The engine became the foundation of the ~10M CHF Replay business case.