Predictive Forecasting & Operational Alerts

Predictive Per-Head Merch Sales Forecasting

The forecasting model was trained on roughly 45,800 consolidated show-level rows enriched with venue capacity, weather, and artist demand signals. On a held-out test split it...

~45,800 show-level rows
Training data
~0.71 R-squared on held-out split
Model signal
~10 units MAE
Prediction error benchmark
Upcoming-tour forecast support
Planning use case
The problem

When a band announced a tour, coordinators estimated how much merch to print and ship largely from experience and a handful of comparable past tours. Over-ordering risked dead inventory; under-ordering risked selling out. There was no consistent, data-backed way to estimate dollars-per-head for a show that hadn't happened yet.

System designed

Defined the forecasting workflow around the merch decisions coordinators actually make: capacity, artist history, show context, comparable performance, and upcoming-tour planning. The model translates historical sales and external context into per-head estimates that can support print quantities, ship quantities, and risk conversations before a tour begins.

Outcome

The forecasting model was trained on roughly 45,800 consolidated show-level rows enriched with venue capacity, weather, and artist demand signals. On a held-out test split it reached about 0.71 R-squared with a mean absolute error of roughly 10 units. Forecasts were generated for upcoming tours as a validation set, positioning the workflow as a data-backed decision aid rather than a replacement for coordinator judgment.

Proof to show

Forecasting flow from historical show data and venue context to per-head planning output.