Is MAPE Still the Right Forecast Accuracy Metric for Demand Planning?

A familiar metric with structural limitations

Mean Absolute Percentage Error, or MAPE, remains one of the most widely communicated indicators in Demand Planning. It is easy to express, easy to compare and familiar to executives. Yet its simplicity can create a misleading picture of forecast performance when product portfolios contain intermittent demand, new products, seasonal items or a large long tail.

The central problem is not that MAPE is mathematically useless. It is that organizations frequently ask one metric to serve several different purposes: diagnosing item-level problems, evaluating planner performance, tracking portfolio improvement and explaining forecast quality to leadership.

Where MAPE can distort the picture

MAPE becomes undefined when actual demand is zero. Depending on the implementation, these observations may be excluded, replaced or handled through special rules. That is not a marginal issue for intermittent demand, new product introductions and granular channel forecasts.

The metric also magnifies errors on low-volume items. Missing demand by two units on an item that sells two units produces a very large percentage error, while a much larger absolute miss on a high-volume product may appear modest. When results are averaged across SKUs, the slowest-moving products can therefore dominate the headline figure even when their operational or financial impact is limited.

Finally, traditional MAPE gives every item the same weight. A critical high-volume SKU and an immaterial long-tail SKU contribute equally to the average, despite having very different effects on inventory, service, capacity and revenue.

WMAPE offers a business-weighted view

Weighted Mean Absolute Percentage Error addresses part of this problem by weighting forecast errors according to actual volume. It generally provides a clearer portfolio-level view of the error affecting the business and is often more appropriate for executive reporting or aggregate performance monitoring.

However, WMAPE introduces its own blind spot. Strong performance on high-volume products can conceal poor forecasting across the long tail. A clean aggregate result does not necessarily mean that planners have control of the full portfolio, particularly when low-volume items still create service issues, obsolete stock or operational complexity.

Different decisions require different metrics

The practical answer is rarely to replace MAPE with a single alternative. A more useful forecast accuracy framework combines several measures according to the decision and audience.

• Use WMAPE to monitor the volume-weighted portfolio impact and communicate a high-level business view.
• Use MAPE selectively for item groups where percentage error remains meaningful and zero demand is not dominant.
• Use forecast bias to identify systematic over-forecasting or under-forecasting that may drive excess inventory or service risk.
• Use MAE or absolute error to preserve the operational scale of misses.
• Segment results by volume, value, variability, lifecycle stage or ABC-XYZ class rather than averaging the entire portfolio indiscriminately.
• Use Forecast Value Add to determine whether planning interventions improve the baseline forecast or merely add activity.

Metrics are part of planning governance

A forecast KPI is not merely a reporting choice. It influences planner behaviour, model selection, exception prioritization and executive perception. If planners are evaluated against an unsuitable metric, they may optimize the score rather than the underlying inventory, service or capacity decision.

Organizations should therefore document how each metric is calculated, how zero-demand periods are treated, at which aggregation level it is reviewed and which business decision it supports. Definitions should remain consistent across the APS, analytics layer and management reports to prevent multiple versions of forecast accuracy from circulating.

The better question

The question is not whether MAPE is universally right or wrong. It is whether the selected metric represents the economic and operational consequences of forecast error for the decision being made.

A mature planning organization does not rely on one headline percentage. It uses a controlled metric hierarchy that connects statistical error, business importance, forecast bias and decision outcomes.

The original discussion is available in Jon Nichols’ LinkedIn post.

Dataleo perspective

The choice between MAPE and WMAPE is a decision-architecture issue, not only a statistical debate. The metric must be aligned with the decision, portfolio segment and audience it is intended to support.

Before automating forecast-performance monitoring in an APS, BI dashboard or AI agent, teams should define who owns the calculation, how exceptions are handled and what action each threshold is expected to trigger. Otherwise, automation simply scales an ambiguous KPI.