What is forecast bias and how do I correct it?

Understanding forecast bias

While MAPE tells you the magnitude of your forecast errors, bias tells you the direction. A company with 8% MAPE might have zero bias (errors are random) or significant positive bias (consistently forecasting too high). Bias indicates a systematic problem in your forecasting methodology or culture.

Types of bias

Optimistic bias (positive): Forecasts consistently exceed actuals. Common causes: - Sales teams over-weighting pipeline probabilities - Management anchoring to ambitious targets - Insufficient adjustment for known headwinds - Failure to account for seasonal slowdowns

Conservative bias (negative / sandbagging): Actuals consistently exceed forecasts. Common causes: - Budget holders underpromising to ensure they beat targets - Risk-averse culture that penalises misses more than it rewards beats - Not accounting for upside scenarios in the base forecast

Measuring bias

Calculate the signed percentage error for each period: (Forecast - Actual) / Actual. Average these over 6-12 months. A positive average indicates optimistic bias; negative indicates conservative bias.

Example: If your trailing 12-month average signed error is +7%, you are consistently forecasting 7% above actuals. This is meaningful optimistic bias that should be corrected.

Correcting bias

1. Make bias visible. Include bias metrics in your monthly accuracy reporting. When people see the pattern, they become more conscious of it.

2. Identify the source. Is bias coming from revenue, costs, or both? Is it specific to certain departments, products, or time horizons? Drill down to find the root cause.

3. Adjust methodology. If sales pipeline is consistently over-weighted, reduce stage probabilities based on historical conversion data. If costs are consistently under-forecast, apply a "cost creep" factor.

4. Calibrate assumptions. Compare each key assumption's forecast vs actual over time. If you consistently assume 10% growth but deliver 7%, adjust your growth assumption baseline.

5. Separate forecasts from targets. When forecasts are tied to compensation, bias is inevitable. Create a clean separation between the forecast (what you expect) and the target (what you aspire to).

Cultural considerations

Bias often reflects incentive structures. If managers are punished for missing forecasts but not rewarded for beating them, they will sandbag. If there is pressure to show optimistic numbers, forecasts will be biased upward. Address the incentives, not just the methodology.