> ## Content Index
> Fetch the complete content index at: https://www.theplanningnerd.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# How to Measure Forecast Accuracy: MAPE, WAPE and Bias Explained
- URL: https://www.theplanningnerd.com/how-to-measure-forecast-accuracy-mape-wape-and-bias-explained/
- Published: 2026-09-10T12:10:10.000Z
- Updated: 2026-09-14T13:23:54.000Z
- Description: You can't improve a forecast you're not measuring properly. And a surprising amount of forecast accuracy measurement in the real world is quietly broken.
- Author: Shane Fernando
- Tags: Demand

You can't improve a forecast you're not measuring properly. And a surprising amount of forecast accuracy measurement in the real world is quietly broken, because the most commonly used metric — MAPE — has a flaw that distorts the picture, especially in businesses with a mix of high and low volume SKUs.

## MAPE: the popular one, with a real problem

**MAPE (Mean Absolute Percentage Error)** averages the percentage error across all SKUs:

`MAPE = Average of |Actual - Forecast| / Actual`

Simple to explain, which is why it's everywhere. The problem is what happens with low-volume SKUs. If a SKU forecast to sell 2 units actually sells 4, that's a 100% error — even though in absolute terms it's a rounding error to the business. Average that alongside your highest-volume SKUs, and a handful of tiny, noisy items can dominate the headline number, making your overall accuracy look far worse than the business actually experienced.

## WAPE: usually the better choice

**WAPE (Weighted Absolute Percentage Error)**, also called WMAPE, fixes this by weighting errors by volume:

`WAPE = Sum of |Actual - Forecast| / Sum of Actual`

Instead of averaging percentage errors equally across SKUs, this sums up the absolute errors and divides by total actual demand. A large SKU being slightly wrong now correctly outweighs a tiny SKU being wildly wrong in percentage terms but negligible in volume terms.

For most FMCG-style businesses with a real spread of SKU volumes, WAPE gives a far more honest read of how the forecast is actually performing where it matters commercially.

## Bias: the metric that catches what accuracy metrics miss

Both MAPE and WAPE measure the *size* of error, but they treat over-forecasting and under-forecasting the same way — a miss is a miss, direction ignored. That hides a common and costly problem: a forecast that's consistently high, or consistently low.

**Bias = Sum of (Forecast - Actual) / Sum of Actual**

A positive bias means you're systematically over-forecasting — building excess inventory over time. A negative bias means you're systematically under-forecasting — running into shortages repeatedly. Either can produce a perfectly respectable-looking WAPE while quietly causing real operational pain, because the errors are compounding in one direction rather than cancelling out.

Checking bias alongside accuracy is one of the most underused habits in demand planning. A forecast can look "accurate enough" on average while being wrong in the same direction every single period.

## How to build this in Excel

For each SKU/period:

- `Absolute Error = ABS(Actual - Forecast)`
- `WAPE = SUM(Absolute Error) / SUM(Actual)`
- `Bias = SUM(Forecast - Actual) / SUM(Actual)`

Track both over a rolling window (8-13 periods is common) rather than a single month, since any individual period can be noisy. What you're looking for is the trend — is WAPE improving or worsening over time, and is bias drifting in one direction.

## What to actually do with these numbers

Don't chase a single "good" accuracy number in isolation — accuracy targets vary enormously by industry, SKU volatility and lead time, so an 80% accuracy that would be disappointing for a stable staple item might be genuinely strong for a highly promotional, short-lifecycle product.

Instead, use accuracy and bias together as a diagnostic:

- **Low accuracy, no clear bias** — demand is genuinely volatile; consider whether more safety stock is the right response rather than chasing a tighter forecast
- **Reasonable accuracy, but clear bias** — the method itself has a systematic problem worth investigating (a growth trend not being captured, a promotion effect not being unwound afterward)
- **Accuracy or bias getting consistently worse over time** — something has changed in the market or the process that the model hasn't caught up with yet

Accuracy measurement isn't about assigning blame for a wrong number. It's the feedback loop that tells you where the forecasting process actually needs attention.

---

*Next:* *Why most S&OP processes don't work* *— or track WAPE and bias yourself using the* *Nerd Foods dataset* *, which includes forecast and actuals data across 104 weeks.*