予測精度

Forecast accuracy quantifies error between forecast and actual values for sales, demand, inventory, hiring, or cash flow. It should be used to improve assumptions, detect bias, and manage uncertainty, not simply to reward lucky guesses.

Common example: MAPE = average(|actual - forecast| / actual). Some teams report accuracy as 1 - error rate. Formula | Common example: MAPE = average(|actual - forecast| / actual). Some teams report accuracy as 1 - error rate. | Use it as the primary operating calculation Bridge | Prior error + demand volatility - model improvement - data freshness improvement +/- bias correction = current error | Use it to explain changes between reviews Segment | Split by customer, product, channel, and period | Use it to find deterioration hidden by averages

This metric is comparable only when inclusion and exclusion rules stay stable. Include | Forecast, actual, horizon, error metric, outlier rule | These make comparisons stable Exclude | Forecasts edited after actuals, changed actual definitions, unexplained exclusions | They inflate accuracy Define explicitly | MAPE, MAE, RMSE, weighted error | The right metric depends on use case

Breaking the metric into drivers clarifies what action should follow the review. Data freshness | Stale data misses change Demand volatility | Turbulence increases error Forecast bias | Systematic optimism or pessimism distorts plans

Use Forecast Accuracy to decide setting inventory and staffing levels because it highlights data freshness and the complexity versus reliability tradeoff. It changes prioritization by forcing teams to state the horizon, boundary conditions, and controllable drivers. It informs adjustments when model bias or seasonality shift, so decisions stay grounded in current conditions.

• Use Forecast Accuracy to decide setting inventory and staffing levels because it highlights data freshness and the complexity versus reliability tradeoff.
• It changes prioritization by forcing teams to state the horizon, boundary conditions, and controllable drivers.
• It informs adjustments when model bias or seasonality shift, so decisions stay grounded in current conditions.

• Define the unit and horizon before comparing data freshness across options.
• Keep the primary driver separate from secondary noise and one-off shocks.
• Document data sources, estimation steps, and confidence ranges for review.
• Translate the tradeoff into thresholds that can be monitored over time.
• Revisit assumptions when the market boundary or policy setting changes.

Do not decide from the number alone; align assumptions, period, segments, and companion metrics. Accuracy alone can hide the range of risk. MAPE can behave badly with small or near-zero actuals. Editing forecasts after results destroys trust.

• Accuracy alone can hide the range of risk.
• MAPE can behave badly with small or near-zero actuals.
• Editing forecasts after results destroys trust.

Companion metrics turn a good-or-bad reading into a discussion of causes and actions. Sales Pipeline Coverage | Evidence for sales forecasts | Explains accuracy drivers Business Plan | Budget and plan | Feeds error learning into planning Scenario Planning | Range of uncertainty | Complements point forecasts

A quarterly revenue forecast of $10M lands at $9M, a 10% miss. Because forecasts have been optimistic for three quarters, the team revises stage probabilities and separates base, downside, and upside cases. After the review, the owner did not treat the metric in isolation. They compared it with companion metrics, checked segment differences, documented assumption changes, and verified data quality before changing the plan. Whether the number improved or deteriorated, the team identified the driver, assigned an owner, and fed the learning into the next budget, operating review, or experiment cycle.

Budget attainment | Actual versus target | Forecast accuracy is actual versus forecast Demand forecast | Estimate of future demand | Accuracy evaluates forecast quality Confidence interval | Uncertainty range | Accuracy evaluates past performance

• Forecast Accuracy is not a universal rule; results depend on boundary assumptions and data quality.
• A single metric like data freshness is not sufficient without considering model bias and seasonality.
• Short term movements can mislead when responses happen with lags.