How Do You Measure Forecast Accuracy? (MAPE, Explained for Ecommerce)
If you can't measure your forecast, you can't improve it. The standard metric is MAPE — but the number most teams chase isn't the one that actually protects them from stockouts. Here's how to measure accuracy, and what to do with it.
Quick Answer
Forecast accuracy is how close your demand forecast lands to actual sales, measured over a period. The most common metric is MAPE (Mean Absolute Percentage Error), and accuracy is usually reported as 100% − MAPE.
Example: forecast 100, sold 120 → error = |120−100|/120 = 16.7%. Average that across SKUs and periods. A blended **WAPE** (weighted by volume) and a **bias / tracking signal** (are you consistently over- or under-forecasting?) round out the picture. Measure per SKU, per channel — and remember the goal is better stocking decisions, not a prettier accuracy number.
The three numbers that actually matter
MAPE — the headline accuracy number
Mean Absolute Percentage Error: for each SKU/period, take |actual − forecast| ÷ actual, then average. Report accuracy as 100% − MAPE. It's intuitive and comparable across SKUs.
Watch out: MAPE breaks on low/zero-sales periods (dividing by a tiny actual explodes the percentage), so it flatters steady sellers and punishes intermittent-demand SKUs. Don't compare a slow mover's MAPE to a hero SKU's.
WAPE — accuracy weighted by what sells
Weighted Absolute Percentage Error: sum of all |actual − forecast| ÷ sum of all actuals. Because it weights by volume, a big miss on a hero SKU counts more than a rounding error on a long-tail SKU. For a portfolio number that reflects dollars at risk, WAPE beats a simple MAPE average.
Bias (tracking signal) — the direction of your error
MAPE and WAPE tell you how big the error is, not which way it leans. Bias = average of (forecast − actual). Consistently positive means you over-forecast (overstock risk + storage fees); consistently negative means you under-forecast (stockout risk). A model can have great MAPE and still be quietly biased one direction — that's the one that costs you.
Which metric for which job
| Metric | What it answers | Best for |
|---|---|---|
| MAPE | Average % error per SKU | Comparing SKUs of similar velocity |
| WAPE | Volume-weighted % error | One portfolio-level number that reflects dollars |
| Bias / tracking signal | Direction of error (over/under) | Catching systematic over- or under-forecasting |
| MAD | Average absolute units off | Sizing safety stock in real units |
What counts as a "good" forecast accuracy?
There's no universal number — it depends entirely on how volatile your demand is. A stable consumable might hit 85–90%+ accuracy (10–15% MAPE); a promo-driven or seasonal SKU might be 60–70% and that's genuinely good for its profile. New products with no history are worse still.
So don't benchmark against a vanity target. Benchmark against yourself over time (is accuracy trending up?) and against a naïve baseline (does your forecast beat "next month = last month" or "same month last year"? If not, the model isn't earning its keep). Rising accuracy vs your own baseline is the signal that matters.
The honest caveat
Accuracy is a means, not the goal. The goal is good stocking decisions — in stock on what sells, lean on what doesn't. A forecast that's 5 points "less accurate" but paired with the right safety stock will beat a higher-accuracy forecast you don't act on. Chasing the accuracy number for its own sake is how teams over-engineer models while still stocking out. Measure it, watch the trend and the bias, then spend your energy on the decisions it feeds.
Frequently asked questions
How do you measure forecast accuracy?
Compare forecast to actual sales over a period. The standard metric is MAPE (Mean Absolute Percentage Error) = the average of |actual − forecast| ÷ actual, expressed as a percentage; accuracy is reported as 100% − MAPE. Pair it with WAPE (volume-weighted) for a portfolio number and bias/tracking signal to see whether you systematically over- or under-forecast. Always measure per SKU and per channel.
What is a good MAPE for inventory forecasting?
It depends on demand volatility, so there's no universal target. Stable consumables can reach 10–15% MAPE (85–90%+ accuracy); seasonal or promo-driven SKUs may sit at 30–40% MAPE and still be good for their profile. The better benchmark is your own trend over time and whether you beat a naïve baseline (last month, or same month last year).
What is the difference between MAPE and WAPE?
MAPE averages the percentage error across SKUs equally, so a tiny long-tail SKU counts as much as a hero SKU. WAPE (Weighted Absolute Percentage Error) divides total absolute error by total actual sales, so high-volume SKUs carry more weight. WAPE is the better single portfolio number because it reflects the dollars actually at risk; MAPE is better for comparing similar-velocity SKUs.
What is forecast bias?
Bias measures the direction of your error, not its size: average of (forecast − actual). Consistently positive = you over-forecast (overstock + storage fees); consistently negative = you under-forecast (stockouts). A forecast can have good MAPE but still be biased one way, which quietly drives either dead stock or lost sales — so track bias alongside accuracy.
Why measure forecast accuracy per SKU instead of company-wide?
A company-wide accuracy number hides the truth: over-forecasting on some SKUs cancels under-forecasting on others, so the blend looks fine while individual SKUs stock out or pile up. The same is true per channel — a SKU can be accurate on Shopify and badly off on Amazon. Accuracy only drives better decisions when measured per SKU, per channel.
Does higher forecast accuracy mean fewer stockouts?
Not on its own. Accuracy only helps if you act on it, and a slightly-less-accurate forecast paired with the right safety stock beats a high-accuracy forecast you don't use. Watch accuracy and bias, then feed both into reorder points and safety-stock policy. See the full forecasting process here.