How do you measure forecast accuracy and detect and correct forecast bias?

Measure accuracy with error metrics computed over time: MAPE (mean absolute percentage error) for magnitude of error, and a separate bias/tracking measure — the mean (signed) error or a tracking signal — to detect systematic over- or under-forecasting. Accuracy is about how far off you are; bias is about consistently leaning one way. To detect bias, trend the signed errors by business unit, forecaster, and line item: a persistent positive or negative mean error reveals sandbagging (deliberate underforecasting to beat target) or optimism. To correct it: feed the analysis back to owners, adjust for known structural bias, separate the forecast from the target/incentive (people bias forecasts when comp depends on them), use driver-based methods to reduce judgement, and hold forecast-accuracy reviews. The key insight is that a forecast can be 'accurate on average' yet badly biased, or unbiased yet noisy — you manage both, and bias is usually the more damaging because it's systematic.