Which forecasting method should I use?

Answer five quick questions about your data and operation. You’ll get a recommended method — and the reasoning — based on the “climb only when the data asks” principle. Whatever it suggests, always benchmark against naive out-of-sample before you trust it.

1. How much clean history do you have?

2. Is there a clear trend (steadily growing or shrinking)?

3. Strong repeating shape (seasonality, day-of-week, intraday)?

4. Do external drivers move your volume (weather, marketing, campaigns, pricing)?

5. Scale and complexity?

Recommended starting point

Rule of thumb: start at the simplest rung that captures the patterns you actually have, add complexity only when the data asks for it, and prove every step out-of-sample against naive (see the forecast value add lesson).

How this works

Forecasting methods form a ladder — naive at the bottom, machine learning at the top — and the skill is stopping at the lowest rung that captures the patterns present in your data. Each rung up adds the ability to handle one more thing: a moving average smooths noise, exponential smoothing weights recent data, Holt-Winters adds trend and seasonality, regression brings in external drivers, and ML finds complex multi-driver interactions. This tool maps your answers onto that ladder. It’s a starting point, not gospel: the honest test is always which method wins on data it hasn’t seen.