Forecast Accuracy Metrics: MAPE, WAPE, and Bias Explained

If you've ever presented a forecast accuracy report to sales or executive leadership, you know the feeling. You report "90% accuracy," and they look at you like you have two heads because they remember the stockout last month or the warehouse overflowing with slow movers.

The problem isn't usually the number itself—it's the metric you chose to calculate it.

Choosing the right forecast accuracy metrics is one of the most critical decisions a demand planner makes. Pick the wrong one, and you’ll mask serious problems or incentivize the wrong behavior. Pick the right ones, and you gain a diagnostic tool that actually helps you improve.

In this guide, we’ll cut through the academic jargon and explain the "Big 3" metrics every planner needs: MAPE, WAPE, and Forecast Bias. We’ll show you exactly how to calculate them, when to use them, and crucially, when to avoid them.

The Problem with "Accuracy" as a Concept

"Accuracy" sounds like a single number, but in demand planning, it’s a composite of two distinct concepts:

1. Magnitude of Error: How far off was the forecast, regardless of direction? (e.g., "We were off by 500 units.")
2. Direction of Error: Did we over-forecast or under-forecast? (e.g., "We were 500 units short.")

Most organizations fixate on magnitude (how close were we?) and ignore direction (bias). This is dangerous. Consistently under-forecasting by 5% might look like "95% accuracy," but it leads to chronic stockouts and lost revenue. Consistently over-forecasting by 5% creates excess inventory and obsolescence risk.

To get the full picture, you need a combination of metrics. Let's look at the standard toolkit.

Mean Absolute Percentage Error (MAPE)

MAPE is the grandfather of forecast metrics. It’s likely the first one you learned and the one your boss asks for. It measures the average percentage error across your dataset.

The Formula

\text{MAPE} = \frac{1}{n} \sum \left| \frac{\text{Actual} - \text{Forecast}}{\text{Actual}} \right| \times 100\%

In plain English: For every item, calculate the percentage error. Then, take the average of all those percentages.

Example Calculation

Let’s calculate MAPE for a 5-month period.

| Period | Actual | Forecast | Error (Act - Fcst) | Absolute Error | % Error (|Error| / Actual) | | :--- | :--- | :--- | :--- | :--- | :--- | | Jan | 100 | 90 | 10 | 10 | 10% | | Feb | 120 | 130 | -10 | 10 | 8.3% | | Mar | 80 | 85 | -5 | 5 | 6.25% | | Apr | 150 | 140 | 10 | 10 | 6.6% | | May | 0 | 10 | -10 | 10 | Undefined (∞) |

Step 1: Calculate % error for each line. Step 2: Average them.

(Note: If we ignore May for a moment): Average of (10% + 8.3% + 6.25% + 6.6%) = 7.79% MAPE.

Pros and Cons

| Pros | Cons | | :--- | :--- | | Intuitive: Everyone understands "we were off by 8%." | The Zero Problem: As seen in May above, if Actual = 0, MAPE cannot be calculated (divide by zero error). | | Standard: Widely used and expected by executives. | Scale Sensitive: It treats an error on a slow mover (selling 10 units) the same as a fast mover (selling 10,000 units), which can skew your aggregate accuracy. | | | Symmetric: It penalizes over-forecasting more heavily than under-forecasting in some interpretations (though mathematically bounded at 0% for actuals). |

When to Use MAPE

Use MAPE for high-volume, stable items where demand never hits zero. It’s great for aggregate reporting (e.g., "Category A Accuracy") but often fails at the SKU level for intermittent demand.

Weighted Absolute Percentage Error (WAPE)

If MAPE is the flawed grandfather, WAPE (often used interchangeably with WMAPE) is the modern workhorse. It solves the two biggest problems of MAPE: the divide-by-zero error and the volume weighting issue.

The Formula

\text{WAPE} = \frac{\sum | \text{Actual} - \text{Forecast} |}{\sum \text{Actual}} \times 100\%

In plain English: Sum up all the absolute errors for every item/month, then divide by the total actual demand.

Example Calculation

Using the same data as above:

| Period | Actual | Forecast | Absolute Error | | :--- | :--- | :--- | :--- | | Jan | 100 | 90 | 10 | | Feb | 120 | 130 | 10 | | Mar | 80 | 85 | 5 | | Apr | 150 | 140 | 10 | | May | 0 | 10 | 10 | | TOTAL | 450 | 455 | 45 |

Calculation: Total Absolute Error (45) / Total Actuals (450) = 10% WAPE.

Why WAPE is Usually Better

Notice that WAPE handled the zero demand in May without crashing. It also naturally weights the result based on volume. An error of 10 units on a high-volume item matters less to the final score than it would in a straight average, but WAPE focuses on the total volume of error vs total volume of demand.

When to Use WAPE

WAPE is your "go-to" metric for supply chain performance. It reflects the operational reality: you have to service the total volume of demand. If you are shipping 1 million units and miss by 100,000, that’s a 10% error rate that reflects the warehouse impact.

Forecast Bias

Bias is the "hidden killer." You can have a great accuracy score (low error magnitude) but still be bleeding money if that error is always in one direction.

Forecast Bias measures the tendency to consistently over-forecast or under-forecast.

The Formula

\text{Bias} = \frac{\sum (\text{Actual} - \text{Forecast})}{\sum \text{Actual}} \times 100\%

Note: Some organizations flip this to (Forecast - Actual). Just ensure you know which direction implies over-forecasting.

Example Calculation

Using our table again:

| Period | Actual | Forecast | Error (Act - Fcst) | | :--- | :--- | :--- | :--- | | Jan | 100 | 90 | +10 | | Feb | 120 | 130 | -10 | | Mar | 80 | 85 | -5 | | Apr | 150 | 140 | +10 | | May | 0 | 10 | -10 | | TOTAL | 450 | 455 | -5 |

Calculation: Total Error (-5) / Total Actuals (450) = -1.1% Bias.

Interpreting Bias

• Negative Bias (in this formula): Forecast > Actual. You are Over-forecasting. This leads to excess inventory.
• Positive Bias: Actual > Forecast. You are Under-forecasting. This leads to stockouts and lost sales.
• Zero Bias: Errors cancel each other out perfectly.

In our example, a bias of -1.1% is quite good. Even though we had errors every month (reflected in the 10% WAPE), the direction of those errors flipped back and forth, meaning we aren't systematically wrong in one direction.

Other Metrics Worth Knowing

While the "Big 3" cover 90% of use cases, you might encounter these alternatives:

RMSE (Root Mean Squared Error)

• What it is: A standard statistical metric that squares errors before averaging them.
• Why use it: It penalizes large errors heavily. Being off by 100 units is treated as much worse than being off by 10 units ten times.
• Best for: Safety stock calculations and when big misses are disproportionately costly.

MAE (Mean Absolute Error)

• What it is: The average error in absolute units (e.g., "We miss by an average of 50 cases").
• Why use it: It’s easy to understand operationally.
• Best for: comparing accuracy across items with similar volumes.

Choosing the Right Metric: A Comparison Table

| Metric | What it Measures | Handles Zero Actuals? | Shows Direction? | Best For | | :--- | :--- | :--- | :--- | :--- | | MAPE | Average % error | No (Infinite) | No | Simple, high-level reporting on stable items. | | WAPE | Weighted % error | Yes | No | Operational decision making; the default for demand planning. | | Bias | Directional tendency | Yes | Yes | Detecting systematic over/under-forecasting issues. | | MAE | Absolute unit error | Yes | No | Understanding magnitude in real terms (cases, pallets). | | RMSE | Penalized error | Yes | No | Safety stock setting; when large errors are dangerous. |

Benchmarks: What is "Good" Accuracy?

Clients always ask us, "What accuracy should we target?" The honest answer is: it depends.

Benchmarks vary wildly by industry, product lifecycle, and lead time.

• FMCG / Food & Bev: High volume, stable demand. Target: 75-85% (15-25% WAPE).
• Consumer Electronics: fast lifecycles, volatile. Target: 60-70%.
• Fashion / Apparel: High seasonality, trends. Target: 55-65%.

Warning: Don't obsess over industry benchmarks. Focus on internal improvement. If you are at 50% today, getting to 55% is a massive win that translates directly to working capital savings.

How DemandPlan Approaches Accuracy

Measuring accuracy is useless if you don't use it to improve. At DemandPlan, we believe in Adaptive Hierarchy™.

Traditional tools force you to measure accuracy at a single level (e.g., SKU-Location). But noise at the SKU level often masks signal at the Category level.

DemandPlan automatically tracks accuracy at every level of your hierarchy simultaneously. We help you identify:

1. Systematic Bias: Are you always over-forecasting "New Products"?
2. Aggregation Benefits: Is your "East Region" forecast accurate even if individual "Store" forecasts are noisy?
3. Forecast Value Add (FVA): We track the accuracy of the statistical baseline vs. your manual overrides. If your adjustments are making the forecast worse (which happens about 50% of the time), we let you know.

Conclusion: Progress Over Perfection

Don't get paralyzed by the math. Start simple:

1. Calculate WAPE to understand your magnitude of error.
2. Calculate Bias to ensure you aren't systematically over-stocking.
3. Review these metrics monthly and look for trends.

Accuracy is not an endpoint; it's a process of continuous learning. Every error is a data point that tells you something about your business. Listen to the data.

---

Ready to stop fighting with spreadsheets? See how DemandPlan automates accuracy tracking and Forecast Value Add analysis. Or, read more about how we compare statistical vs machine learning models.