You don't need a data science team to forecast demand. A 4–8 week moving average of each item's sales, adjusted up or down for known seasons and planned promotions, gets most small retailers most of the way there — and it feeds straight into your reorder points.
Demand forecasting sounds like something only large retailers with data teams do. For routine reordering decisions, a much simpler approach gets you most of the value — and it's a big improvement on ordering from memory.
Step 1: Start with a moving average
Take an item's sales for the last several weeks and average them — that's your baseline expected demand for the next week. Each week, drop the oldest week and add the newest, so the forecast keeps adjusting to recent trends.
| Week | Units sold | 4-week moving average (forecast for next week) |
|---|---|---|
| 1 | 38 | — |
| 2 | 42 | — |
| 3 | 35 | — |
| 4 | 45 | (38 + 42 + 35 + 45) ÷ 4 = 40 |
| 5 | 41 | (42 + 35 + 45 + 41) ÷ 4 ≈ 41 |
| 6 | 48 | (35 + 45 + 41 + 48) ÷ 4 ≈ 42 |
Divide the weekly forecast by your trading days to get the average daily sales figure your reorder point needs.
Step 2: Layer in known seasonality
If an item reliably sells more in a particular month or around a festival, a flat moving average will underestimate demand right before the peak and overestimate right after. A simple fix is a seasonal factor: if last Diwali week sold 1.8× a normal week, multiply this year's baseline by 1.8 for that week.
Step 3: Account for planned promotions
A moving average is built from normal sales — it won't predict a spike from a promotion you're about to run. If a discount or marketing push is coming, raise the forecast for that period manually, then leave the promotion weeks out of next month's average so they don't inflate normal demand.
Step 4: Check how accurate it was
Each week, compare the forecast with what actually sold. A simple accuracy check is the average percentage error over the last few weeks. If an item is regularly off by more than 30–40%, its demand is probably erratic — a larger safety buffer will serve you better than a cleverer forecast.
What this is (and isn't) good for
This approach works well for routine reordering of items with reasonably steady demand. It won't predict a genuinely unpredictable spike — XYZ analysis helps identify which items fall into that category. For new items with no history, start from the sales of a similar item and adjust after a few weeks of real data.
StashBill Pro's demand forecasting builds these projections automatically from your sales history and turns them into a suggested reorder plan.
Frequently asked
How far back should my moving average look?
4–8 weeks is a reasonable default for most retail items — long enough to smooth out noise, short enough to reflect recent trends. Use a shorter window for fast-changing items.
Do I need historical data to start forecasting?
Yes, at least a few weeks of sales history. For a brand-new item, borrow the pattern of a similar item until its own data builds up.
Is a weighted moving average better?
It can be — giving recent weeks more weight reacts faster to trends. For most small shops, a simple average reviewed weekly is accurate enough, and easier to trust.