Forecasting demand when your catalogue is too small for statistics

Small catalogues do not have enough signal for machine learning to beat a well-tuned moving average with seasonality applied at category level.

Use a 13-week weighted average, apply a category seasonality index rather than a per-SKU one, and overlay known events manually.

Spend your effort on forecast error tracking instead. Knowing your MAPE by SKU band tells you how much safety stock to hold, which is the decision that actually costs money.