Choose a forecast model

Before you start

• A variant whose sales pattern you know well.

Steps

1. Open the Forecast page → click the variant title to open variant detail.
2. Go to the Forecast tab.
3. Look at the History chart — note the demand shape (flat / drifting / trending / seasonal).
4. Use the decision tree:
   • Flat → Average.
   • Slow drift → Moving average or Exponential smoothing.
   • Up / down trend → Linear regression.
   • Annual cycle → Seasonal (Enhanced plan). Pick it by hand — Auto isn’t currently choosing Seasonal or Holt.
   • Two cycles → Double seasonal (Enhanced plan).
   • Last year’s actuals are the plan → History replay, which repeats the view’s history window instead of fitting a curve. Pair it with a growth amplifier for “last year, plus 10%”.
   • You don’t know → Auto.
5. Pick the model from the dropdown.
6. Save.

Worked example

My winter coats sell heavily Oct–Feb and almost nothing Apr–Sep. The History chart shows clear yearly seasonality. Pick Seasonal (Enhanced plan) — you need to select it yourself, since Auto is not choosing Seasonal at the moment.

Common failure modes

Symptom	Likely cause	Fix
Switching model didn’t change suggestions	Variant has less than 30 days of history — Persistence is forced.	Wait for more history.
Auto picks a model you disagree with	Auto sees patterns you don’t.	Override manually if your business knowledge trumps the data; otherwise trust Auto.

See also

• Forecast → Forecast models
• Forecast → Automatic model optimization

Last updated: 14 September 2026