A new release has been on the service for a week. The model estimates its eight-week total from its features, fits the shape of its demand curve to the first days of sales, and forecasts the market's total with trend, weekday rhythm, holidays and the release calendar. The catalogue's average curve and a seasonal-naive forecast stand in for the spreadsheet.
Theatrical admissions, the gap between cinema and VOD release and genre predict a title's eight-week total from the back catalogue — tree ensembles in production, a log-linear regression here.
A Bass diffusion curve has two parameters: how many buy on their own initiative and how many follow others. Fitting them to the first week, with the feature-based total as a prior, gives the whole trajectory with its weekend ripples.
Total daily purchases decompose into trend, weekly seasonality, holiday effects and the release calendar — Prophet in production, the same additive structure fitted by least squares here — and the components explain the forecast.