Forecast Accuracy

Forecast accuracy is the metric that quantifies how close a prediction came to the actual result once the period has closed. It applies to demand, sales, revenue or inventory forecasts, and it is almost always reported as a percentage: the higher it is, the more reliable the forecasting model was. It is the flip side of forecast error: if the error was 12%, accuracy was 88%.

It matters because almost every planning decision (how much inventory to buy, how many sales reps to hire, what budget to commit) starts from an estimated number. Measuring its accuracy systematically is what turns the forecast into an instrument that improves over time instead of a guessing game. Building and monitoring this metric on consolidated data is part of what a Metrix dashboard solves, with the forecast and its error sitting right next to the actual data.

How it is calculated

The most widespread method starts from MAPE (Mean Absolute Percentage Error). Accuracy is simply 100% minus MAPE. For example, if you forecast selling 900 units of a SKU in a month and you sold 1,000, the absolute error is 100 units out of 1,000, or 10%, and accuracy was 90%. When there are many products or many periods, you average the error of each one to get an overall number for the operation.

There are variants worth knowing because they change how you read the result. WMAPE (volume-weighted MAPE) gives more weight to the best-selling products, preventing a small SKU with a huge error from distorting the average. Bias measures whether the model is always wrong in the same direction: over-forecasting month after month inflates inventory, while under-forecasting causes stockouts. A healthy operation looks at accuracy and bias together, not separately.

Why it is one of the metrics that moves the most money

Every point of accuracy you gain translates into less tied-up capital and fewer lost sales. A weak forecast pushes teams to over-buy "just in case," which inflates inventory and financing costs, or to fall short, which causes stockouts and customers who buy somewhere else. In chains with perishable products, the cost is double: whatever is left over expires. That is why, in consumer goods, retail and distribution, forecast accuracy is an indicator watched by both the commercial and the finance leadership.

What counts as a good number (with nuances)

There is no universal threshold: it depends on the horizon (forecasting next week is not the same as forecasting the next quarter), on demand variability and on the level of aggregation. As a general rule:

• At the individual product and monthly level, accuracy is usually much lower than at aggregated levels
• At the category or region level (more aggregated, so errors offset each other), accuracy goes up
• For new products or very erratic demand, accuracy drops sharply and calls for other methods

The rule of thumb: the more you aggregate and the shorter the horizon, the higher accuracy goes. Comparing your company's number against another company's only makes sense if the level of aggregation and the horizon are the same.

Forecast accuracy vs. sales forecast

They are related but different concepts, and it pays not to confuse them:

Put simply: the sales forecast is the prediction, and forecast accuracy is the grade you give that prediction once you know the result.

A concrete example in Argentina

A food distributor in Greater Buenos Aires forecasts selling 5,600 cases of a product in March. It sells 5,000. The absolute error is 600 cases, MAPE is 12% and accuracy is 88%. Reviewing its history, the company notices that it over-forecasts almost every month: there is a systematic positive bias. It corrects the model downward, adjusts its purchasing and frees up capital that was tied up on store shelves and in the warehouse. The number that changed the decision was not the forecast itself, but the measurement of its accuracy over time.

Common mistakes when measuring it

• Looking only at the average and not at bias: two models can have the same MAPE, but one makes you over-buy every time and the other doesn't
• Comparing apples to oranges: comparing accuracy at the SKU level against another company's accuracy measured at the category level
• Not weighting by volume: letting a marginal product with a huge error ruin the overall reading (hence WMAPE)
• Measuring it late or by hand: if the calculation isn't automatic on consolidated data, nobody keeps it up month after month and the metric dies

Forecast accuracy is not a number to report and forget: it is an improvement loop. You measure it, understand where and in which direction the model fails, adjust it, and measure again.