MAPE vs WMAPE: Which Forecast Error Metric to Use
MAPE vs WMAPE explained for demand planners: formulas, worked examples, and why WMAPE is the honest forecast error metric for mid-market manufacturers.
For almost every executive dashboard and S&OP use case, use WMAPE, not MAPE. WMAPE weights each item's error by its actual demand, so the SKUs that fund your company carry the most weight, while a cheap, low-volume part can't blow up the average. MAPE treats every item equally and divides by the actual, which means a 3-unit miss on a slow mover can read as a 100% error and tank a metric your warehouse would tell you is fine.
I ran demand planning at a $250M furniture manufacturer. Switching our primary metric from MAPE to WMAPE changed which SKUs the team fought over in the planning meeting. That changed where we put our working capital. The math looks like a nitpick. It isn't.
What each metric actually measures
Both metrics start from the same raw ingredient: absolute error, the gap between forecast and actual with no regard to direction. The difference is how they aggregate that gap across a catalog.
MAPE (Mean Absolute Percentage Error) computes the percentage error for each item, then averages those percentages with equal weight:
MAPE = (1/n) × Σ ( |Actual − Forecast| / Actual ) × 100
WMAPE (Weighted Mean Absolute Percentage Error) sums the absolute errors and divides by total demand, so each item is weighted by its volume:
WMAPE = Σ |Actual − Forecast| / Σ Actual × 100
The simplified WMAPE formula above is the one the Wikipedia entry on mean absolute percentage error (2026) and the Institute of Business Forecasting glossary (2026) both give. One line: MAPE values a $40 lamp and a $4,000 sectional equally; WMAPE values them by how much they actually move.
The example that ends the debate
Take five SKUs across one week. Two are high-value A items, one is a mid-tier B, and two are cheap C items with tiny volume.
| SKU | Value tier | Forecast | Actual | Abs error | APE (for MAPE) |
|---|---|---|---|---|---|
| Sectional | A | 200 | 180 | 20 | 11.1% |
| Dining set | A | 150 | 165 | 15 | 9.1% |
| Accent chair | B | 80 | 70 | 10 | 14.3% |
| Table lamp | C | 6 | 3 | 3 | 100.0% |
| Throw cushion | C | 4 | 2 | 2 | 100.0% |
MAPE = (11.1 + 9.1 + 14.3 + 100 + 100) / 5 = 46.9% → accuracy 53.1%
WMAPE = (20 + 15 + 10 + 3 + 2) / (180 + 165 + 70 + 3 + 2) = 50 / 420 = 11.9% → accuracy 88.1%
Same forecast. Same actuals. MAPE says you're failing at 53% accuracy. WMAPE says you forecast the items that fund the business within 12%. The entire 35-point gap comes from two cheap items where a 3-unit miss reads as 100% error.
What this does to your team
Run your S&OP meeting on MAPE and your planners will spend the week chasing the lamp and the cushion to drag the average down. The sectional that pays the mortgage gets less attention. That's a metric driving the wrong behavior, week after week.
Why MAPE breaks on low-volume items
MAPE has a structural flaw baked into the formula: it divides by the actual. As the actual approaches zero, the percentage error approaches infinity. The authoritative reference here is Hyndman and Koehler's "Another look at measures of forecast accuracy" (2006), the most-cited paper in the field on this question.
The free textbook from the same authors, Forecasting: Principles and Practice (2026), puts it plainly: percentage-error measures are "infinite or undefined if y_t = 0" and produce "extreme values if any y_t is close to zero." For any business with a long tail of slow movers, spare parts, or intermittent demand, MAPE is dominated by the least important items.
The asymmetry nobody mentions
MAPE is also biased in direction. Under-forecasts cap out at 100% error (you can't be more than 100% short of the actual), but over-forecasts have no upper bound. The Wikipedia entry summarizes the consequence: MAPE "puts a heavier penalty on negative errors... than on positive errors," so it "systematically selects a method whose forecasts are too low."
That bias is dangerous in a real planning org. A metric that quietly rewards under-forecasting will starve your service levels while the scorecard stays green. WMAPE has neither the division-by-zero blowup nor the same lopsided penalty, because small items contribute small numerators and small denominators.
When MAPE still earns its place
WMAPE wins the default, but MAPE isn't useless. A few cases where it's the right call:
- Single-SKU tracking over time. For one stable, high-volume item, MAPE is a clean, intuitive trend line. No weighting needed when there's one thing.
- Comparing to academic or vendor benchmarks. Most published forecasting research reports MAPE, so you may need it for an apples-to-apples comparison. See our forecast accuracy benchmarks for manufacturers for what "good" looks like by industry.
- Communicating to non-planners. "We're 10% off on this product" lands faster than explaining a weighting scheme to a sales VP.
For anything that aggregates across a catalog, switch to WMAPE. And for true intermittent demand — spare parts, long-tail SKUs that sell zero most weeks — neither percentage metric is ideal. Hyndman and Koehler recommend a scaled error like MASE instead, which we cover in forecasting intermittent demand for spare parts.
Quick decision table
| Situation | Use |
|---|---|
| Executive / S&OP dashboard | WMAPE |
| Mixed catalog, wide value range | WMAPE |
| Low-volume or intermittent demand | MASE or WMAPE (not MAPE) |
| Single high-volume SKU trend | MAPE is fine |
| Comparing to published benchmarks | MAPE |
| Want the financial view | WMAPE weighted by revenue or margin |
The upgrade most teams miss: weight by margin, not units
Standard WMAPE weights by unit volume. The sharper move is to weight by revenue, or better, by gross margin. A point of error on your highest-margin line costs more than the same point on a thin-margin commodity, and a margin-weighted WMAPE puts your accuracy effort exactly where the profit is.
What this looks like in practice
We ran a margin-weighted WMAPE alongside the unit version, and the two told genuinely different stories. The unit-weighted metric flattered us on high-volume, low-margin items. The margin-weighted version exposed that our profit center was the line we forecast worst. That's a finding that changes a quarter, not a dashboard.
The financial stakes are real. McKinsey's Supply Chain 4.0 research (2026) reports that better demand forecasting can cut forecast error by 30 to 50 percent, which translates into up to 75 percent lower lost sales and inventory reductions of up to 75 percent. You can't capture any of that if your error metric is pointing your team at the wrong SKUs. This connects directly to how you read your full scorecard, which we break down in demand planning KPIs and metrics that matter.
Neither metric tells you direction
Whatever you choose, MAPE and WMAPE both take absolute values. Neither tells you which way you're wrong. A team can post a strong WMAPE while systematically over-forecasting into stranded inventory.
Always run bias alongside error
Run a signed bias number next to your error metric, every cycle. Error tells you how wrong you are; bias tells you which way, and the second number is what fills the warehouse with dead stock. The standard cumulative bias and tracking-signal approaches are simple to compute and worth the discipline — we walk through them in forecast bias: how to measure and eliminate it.
AI-native planning platforms now let you slice WMAPE by unit, revenue, and margin and overlay bias on the same view, so you're not maintaining three spreadsheets to see the full picture. The metric choice still matters more than the tool, though. WMAPE, weighted by margin, with bias alongside, is the honest scorecard.
How to actually make the switch
Changing your primary metric is a process change, not a formula swap. A clean sequence:
- Compute both for one product family. Pull a quarter of forecast-vs-actual history and calculate MAPE and WMAPE side by side. The gap is your "how much was the old metric lying to me" number.
- Add a margin-weighted WMAPE. Use gross margin per unit as the weight. Compare it to the unit-weighted version and see where they disagree.
- Put bias next to it. A signed cumulative bias per family, refreshed every cycle.
- Rebuild the S&OP review around it. Sort the meeting agenda by margin-weighted error contribution, so the highest-stakes misses get discussed first. Our demand planning KPIs guide has a starter scorecard layout.
- Hold the line for a quarter. Metric changes feel worse before they feel better, because WMAPE stops letting you hide.
Frequently asked questions
Is WMAPE always better than MAPE?
No, but it's the better default for almost any catalog-level reporting. WMAPE wins whenever you aggregate across items with a wide range of volumes or values, because it weights errors by demand. MAPE is fine for a single high-volume SKU tracked over time, or when you need to match a published benchmark that reports MAPE.
How do you calculate WMAPE?
Sum the absolute errors across all items, then divide by the sum of all actuals: WMAPE = Σ|Actual − Forecast| / Σ Actual. Multiply by 100 for a percentage. Unlike MAPE, you do not average individual percentage errors, which is exactly why a near-zero actual can't blow up the result.
Why does MAPE give such high error numbers for slow-moving items?
Because MAPE divides each error by that item's actual demand. When the actual is tiny — say you sell 2 units against a forecast of 4 — the percentage error is huge (100%) even though the absolute miss is trivial. As actuals approach zero, MAPE approaches infinity, which is why Hyndman and Koehler warn against it for intermittent demand.
Should I weight WMAPE by units, revenue, or margin?
Start with units if that's all you have, but margin is the sharpest weight for an executive view. A point of forecast error on a high-margin line costs more profit than the same point on a thin-margin commodity, so a margin-weighted WMAPE directs your accuracy effort toward where the money actually is. Many teams report unit-weighted and margin-weighted side by side.
Do I still need to track forecast bias if I use WMAPE?
Yes. WMAPE and MAPE both use absolute values, so they tell you how wrong you are but not in which direction. You can post a strong WMAPE while consistently over-forecasting into excess inventory, so always run a signed bias metric alongside your error number every planning cycle.
Let's see what's worth building first.
A 15-minute call: tell me where your AI or planning is stuck, and I'll tell you the one thing worth building first — and whether it's worth doing at all.