How do I measure forecast accuracy?

Why measure forecast accuracy?

If you do not measure forecast accuracy, you cannot improve it. Tracking accuracy over time builds credibility with the board, identifies where your planning is strongest and weakest, and drives continuous improvement in your FP&A process.

Key accuracy metrics

MAPE (Mean Absolute Percentage Error): The most widely used metric. Calculate |Actual - Forecast| / Actual for each period, then average across periods. A MAPE of 8% means your forecasts are, on average, 8% away from actuals.

Forecast bias: The average signed error. Positive bias means you consistently over-forecast; negative means you under-forecast. Bias reveals systematic tendencies that MAPE alone cannot detect.

Weighted MAPE (WMAPE): Weights each period's error by its actual value. This prevents small months from disproportionately affecting the overall accuracy score. More appropriate when there is significant volume variation between periods.

What to measure

• Revenue forecast accuracy (overall and by segment)
• Cost forecast accuracy (overall and by category)
• EBITDA or operating profit accuracy
• Cash flow forecast accuracy
• Headcount forecast accuracy

Implementation

1. Create a tracking table. For each month, record the forecast value (captured at the time the forecast was made), the actual value, the absolute error, and the percentage error.

2. Calculate rolling MAPE. Compute a trailing 6 or 12-month MAPE to see trends. Is accuracy improving or degrading?

3. Analyse by line item. Which revenue streams or cost categories have the worst accuracy? These are your improvement targets.

4. Analyse by horizon. How does accuracy degrade as the forecast horizon extends? This tells you the optimal reforecasting cadence.

5. Report and act. Include accuracy metrics in your monthly FP&A report. Set improvement targets and hold the team accountable.

Accuracy improvement techniques

Focus on the line items with the largest errors first. Improve data quality for key inputs. Shorten feedback loops with more frequent reforecasting. Use driver-based models that capture cause-and-effect relationships. Incorporate leading indicators.