Free Tool
Measure how good your demand forecast actually is: paste the actual sales and the forecast for the same periods, and get MAPE (average percentage error), WAPE (volume-weighted error), and bias (systematic over/under-forecasting) — the three numbers every inventory decision quietly depends on.
WAPE
10.8%
MAPE
10.3%
Bias
-4.5%
Headline: 89.2% forecast accuracy (100 − WAPE)
A store-level accuracy number hides the products that are bleeding: overall WAPE can look fine while your top seller runs 30% under-forecast. Track WAPE and bias per SKU monthly, and always compare against a naive baseline ("next period = last period") — a forecast that can't beat naive isn't earning its keep. Yeer does this scoring automatically and switches models per product when accuracy degrades — see how Yeer forecasts demand.
Get the free Shopify Inventory Playbook (PDF)
Every formula on this site with worked examples, plus reorder, pricing, and forecast-accuracy checklists you can run monthly.
No spam — just the playbook and an occasional inventory tip. Unsubscribe anytime.
The two most used metrics are MAPE (mean absolute percentage error — the average of |actual − forecast| / actual per period) and WAPE (total absolute error divided by total actual units). Accuracy is often quoted as 100% minus the error. WAPE is usually the better headline number for stores because it weights busy periods more.
It depends on aggregation level. Store- or category-level monthly forecasts often reach 10–20% MAPE; individual SKU weekly forecasts of 20–40% are common and still useful. Comparing your number against a naive forecast (e.g. 'same as last period') matters more than any absolute benchmark.
MAPE averages the percentage error of each period equally, so slow periods with tiny denominators can explode the number (and it's undefined when actuals are zero). WAPE divides total absolute error by total actual volume, weighting periods by how much actually sold — more robust for intermittent, spiky ecommerce demand.
Bias measures whether you systematically over- or under-forecast: (sum of forecasts − sum of actuals) / sum of actuals. A forecast can have decent MAPE yet run +15% biased — which quietly builds overstock. Aim for bias near zero; investigate anything beyond ±5%.
Forecast per product rather than store-wide, use models matched to each product's pattern (seasonal vs trending vs steady), refresh forecasts frequently, and correct with real-time signals (demand sensing). This is Yeer's approach: multiple models per SKU, auto-selected, refreshed up to hourly.
Yeer tracks per-product accuracy continuously and switches models when one degrades. Free up to 50 products.
Install Free on Shopify