How to forecast sales
You don't need a data-science team to forecast sales well — you need clean data, the right method, and the humility to check yourself. Here's the whole process in six steps.
1. Gather clean history
Collect sales by period (monthly is common) for as long as you can — 12+ periods unlocks trend and accuracy estimates, 24+ unlocks reliable seasonality. Keep the periods evenly spaced and fix obvious data-entry errors first.
2. Look before you model
Inspect the series. Is it trending up or down? Is there a repeating yearly pattern? Any one-off spikes (a promotion, an outage) you should note or exclude? This tells you whether a trend or a seasonal model fits.
3. Choose a method — or let it choose
A steady trend suits linear regression; a repeating cycle suits Holt-Winters. If you're unsure, let the tool backtest both and auto-select the one that predicts your recent past best.
4. Read the interval, not just the number
Always look at the 95% confidence interval. A wide interval means high uncertainty — plan for the range, not a single figure. The interval widens the further out you forecast.
5. Validate with a backtest
Trust is earned out-of-sample. Check how the model would have predicted the last few periods (its MAPE). A backtest error under ~10% is solid for most businesses; much higher means treat the forecast as a rough guide.
6. Update on every new period
A forecast is a living estimate. Re-run it whenever fresh actuals land — recent data improves both the point forecast and the intervals.
Salesdamus does steps 3–5 for you automatically: it backtests candidate models, picks the best, and shows the forecast with confidence intervals and accuracy — free, in your browser.