What forecast accuracy means in FBA
Forecast accuracy, commonly measured using MAPE (Mean Absolute Percentage Error), tells you how well you predict future demand for each SKU. For Amazon FBA sellers, forecast accuracy directly determines whether you order the right quantity at the right time. Poor forecasts lead to one of two expensive outcomes: stockouts (lost sales, BSR decay, wasted PPC spend) or overstock (storage fees, aged inventory surcharges, tied-up capital).
MAPE is the most common way to quantify forecast accuracy because it expresses error as a percentage, making it comparable across SKUs regardless of volume. A SKU selling 10 units/day and a SKU selling 500 units/day can both be evaluated on the same scale.
Most FBA sellers do not formally track forecast accuracy, which means they cannot improve it. If you do not know your current MAPE, you cannot tell whether a new forecasting approach (moving average vs. exponential smoothing vs. a tool like Profit Hawk) actually performs better than your gut feeling.
MAPE formula
Example: forecast accuracy across 4 SKUs
A private label seller reviews last month’s forecast accuracy across four SKUs (average selling price $52, total catalog revenue ~$3.1M/year):
| SKU | Forecast | Actual | |Error| | % Error |
|---|---|---|---|---|
| Widget-A | 400 | 450 | 50 | 11.1% |
| Widget-B | 300 | 260 | 40 | 15.4% |
| Widget-C | 180 | 120 | 60 | 50.0% |
| Widget-D | 550 | 580 | 30 | 5.2% |
MAPE = (11.1% + 15.4% + 50.0% + 5.2%) ÷ 4 = 20.4%
Forecast Accuracy = 100% – 20.4% = 79.6%
Widget-C is the outlier at 50% error. Investigating why (a competitor launched, a listing was suppressed, a coupon ran unexpectedly) will improve future forecasts more than tuning the other three SKUs.
FBA-specific considerations
Forecast accuracy for FBA sellers faces unique challenges that warehouse-based retailers do not:
Stockouts contaminate your actuals. If a SKU was out of stock for 10 of 30 days, the “actual” sales number understates true demand. Using stockout-contaminated actuals to measure MAPE makes your forecast look worse than it really was (you under-sold because you ran out, not because demand was lower). Clean your data by excluding or adjusting periods with stockouts before calculating MAPE.
Amazon’s algorithm changes shift demand unpredictably. A search algorithm update, a new competitor getting the top organic spot, or a change in PPC auction dynamics can shift demand 20% to 40% overnight with no signal in your historical data. This puts a floor on achievable MAPE for most FBA sellers at around 15% to 25%.
Promotional spikes need separate treatment. Lightning Deals, Prime Day, and coupon stacks create demand spikes that are not representative of baseline demand. Either exclude promotional periods from your MAPE calculation or forecast them separately with a lift multiplier. Combining promotional and baseline data leads to inflated safety stock levels and unnecessary days of supply buffers.
Common mistakes
- Not measuring forecast accuracy at all. Most FBA sellers reorder based on gut feel or simple rules of thumb and never compare their predictions to actual outcomes. Without tracking MAPE, you cannot identify which SKUs are unpredictable and need larger safety stock buffers.
- Using MAPE on very low-volume SKUs. A SKU that sells 2 units/month can show 100% MAPE just from selling 1 or 3 units instead of 2. MAPE is unreliable for low-volume products. Use absolute error or weighted MAPE (WMAPE) for long-tail SKUs.
- Measuring at the wrong granularity. Monthly MAPE will always look better than weekly MAPE because averaging smooths out variation. Match your measurement period to your decision cadence. If you reorder weekly, measure weekly forecast accuracy.