What demand forecasting means for FBA sellers
Demand forecasting takes your historical sales velocity and projects it forward to answer one question: how many units of each SKU will you sell over a given future period? The output feeds directly into your reorder point calculation, your purchase order quantities, and your cash flow planning.
For Amazon FBA sellers sourcing from Asia, the forecast window matters more than in most retail businesses. A 75-day ocean freight lead time means you’re placing orders today based on what you expect to sell three months from now. Get the forecast wrong by 30% and you either stockout (losing the Buy Box and organic rank) or sit on excess inventory (paying aged inventory surcharges and tying up capital).
The two most common forecasting methods for FBA are the simple moving average and the weighted moving average. Both use trailing sales data. The difference is whether recent weeks count more than older weeks.
Demand forecasting formulas
Example: forecasting a $42 garlic press
You sell a garlic press at $42 ASP, sourced from Guangzhou with a 68-day lead time. Here are your last 4 months of daily sales averages:
| Month | Avg daily units |
|---|---|
| January | 22 |
| February | 26 |
| March | 31 |
| April | 28 |
Simple moving average (4-month): (22 + 26 + 31 + 28) / 4 = 26.75 units/day
Weighted moving average (weights: Jan=1, Feb=2, Mar=3, Apr=4):
(1×22 + 2×26 + 3×31 + 4×28) / (1+2+3+4) = (22 + 52 + 93 + 112) / 10 = 27.9 units/day
The WMA is higher because it gives more weight to March and April, which had stronger sales. For a 68-day lead time order, you’d forecast: 27.9 × 68 = 1,897 units of lead time demand. Add safety stock on top of that for your actual PO quantity.
If you used the SMA instead: 26.75 × 68 = 1,819 units. The 78-unit difference at $42 ASP represents about $3,276 in potential lost sales if the WMA turns out to be more accurate.
How Amazon's ecosystem changes the forecast
Raw sales data on Amazon is noisy. Stockouts suppress demand (you can’t sell what you don’t have), so any period where you were out of stock understates true demand. PPC spend shifts velocity up or down artificially. Lightning deals and coupons create spikes that may not repeat. And competitor stockouts can temporarily inflate your numbers.
Before feeding data into a moving average, clean it. Replace stockout days with your average in-stock velocity. Strip out one-time deal spikes unless you plan to repeat them. Account for any planned advertising changes in the forecast window. Amazon’s receiving delays (often 5 to 14 days at peak) also need to be layered into your lead time, which extends the forecast window you’re predicting for. Pairing your forecast with adequate safety stock buffers the inevitable forecast errors.
Common mistakes
- Using raw sales data during stockouts. If you were out of stock for 12 days last month, your average daily sales will be understated. Backfill those days with your in-stock average or your forecast will be too low every cycle.
- Ignoring seasonality. A flat 90-day moving average will underorder for Q4 and overorder for Q1 every year. Apply a seasonality index to adjust the baseline forecast.
- Setting the forecast window too short. A 30-day average reacts fast to trends but whipsaws on noise. For products with 60+ day lead times, a 60 or 90-day window smooths out the noise while still capturing real trend changes.