There is no single right way to forecast demand. The best method depends on how much data you have, how stable the product is, and what drives its sales. This guide walks through the methods that actually get used, grouped into qualitative (judgment-based) and quantitative (data-based), with a clear note on when each one fits.
Pick the method that matches your data: qualitative when you have little history, quantitative once you have clean sales data. Most teams run a quantitative baseline and adjust it with judgment. This is the toolbox behind demand forecasting.
Qualitative methods (when you lack data)#
Use these for new products, new channels, or anything without a sales history.
1. Expert judgment#
Your sales team and operators estimate demand from experience. Fast, cheap, and biased, so treat it as a starting point, not gospel.
2. Market research#
Surveys, pre-orders, and interviews to gauge interest before launch. Useful for a new SKU, but stated intent rarely matches real buying.
3. The Delphi method#
A structured panel of experts forecasts independently, sees the anonymized results, and revises over rounds until they converge. It removes the loudest-voice problem of a group meeting.
Quantitative methods (when you have sales data)#
Once you have a few months of clean daily sales, switch to these.
4. Moving average#
Average the last n periods to smooth out noise into a baseline.
Simple and stable, but it lags trends and treats old data the same as new.
5. Exponential smoothing#
Weight recent demand more heavily than old demand. The factor alpha (0 to 1) sets how reactive the forecast is.
A higher alpha reacts faster to change, a lower alpha stays smoother. Good for stable demand, and variants handle trend and seasonality.
6. Regression analysis#
Tie demand to the things that drive it: price, ad spend, season, weather. Regression estimates how each driver moves demand, so you can forecast under different plans.
7. Time-series models#
Methods like ARIMA decompose history into trend, seasonality, and noise. Powerful for strong seasonal patterns, but they need more data and tuning.
- Quantitative methods are objective and repeatable
- Regression explains the why, not just the what
- Time-series models capture seasonality well
- All need clean history, which new SKUs lack
- They miss one-off events you can see coming
- More complex models need tuning and maintenance
How to choose#
A quick guide:
| Situation | Best method |
|---|---|
| Brand-new product, no data | Qualitative (judgment, surveys, Delphi) |
| Steady demand, some history | Exponential smoothing |
| Demand driven by price or promos | Regression |
| Strong seasonality | Time-series (ARIMA) or seasonal smoothing |
Even the best statistical method only knows the past. Layer in upcoming promos, launches, and a viral moment you can see building before you trust the number.
Skip the spreadsheet math and forecast every SKU automatically
The bottom line#
The methods split into qualitative (judgment, for new or data-light products) and quantitative (moving average, exponential smoothing, regression, time-series, for products with history). Match the method to your data, adjust for known events, and feed the result into your reorder points and safety stock. Enough Stock runs the quantitative work for you across every channel so you can focus on the adjustments only you know.