Forecasting Math

How to Calculate Forecast Accuracy, Bias, and Demand Variability in S&OP

The core S&OP and demand planning formulas worked end to end: forecast accuracy, bias, demand variability, and the safety stock they feed.

Start with forecast accuracy because every downstream number depends on it. The workhorse metric is MAPE: for each SKU-period, take the absolute value of (actual minus forecast) divided by actual, then average across periods. If June forecast was 1,200 units and actual shipped was 1,050, the absolute percent error is |1,050 - 1,200| / 1,050 = 14.3%. Average that across 12 months and you have annual MAPE. Accuracy is simply 100% minus MAPE, so a 14.3% error month reads as 85.7% accurate. Pull actuals from shipment history in your ERP, not open orders, and the Forecast Accuracy calculator does this per item and weighted by volume.

Weight MAPE by volume or it lies to you. A 60% error on a SKU that sells 3 units per month should not outweigh a 4% error on a SKU that sells 40,000. Weighted MAPE (WMAPE) uses the sum of absolute errors divided by the sum of actuals: add up every |actual - forecast| across the portfolio, divide by total units shipped. If total absolute error was 42,000 units against 600,000 shipped, WMAPE is 7.0%. That single number is what an S&OP review should track, and it maps cleanly onto inventory and service outcomes in a way that unweighted MAPE never does.

Bias is a different question than accuracy, and confusing the two is common. Accuracy measures how far off you are; bias measures which direction you lean. Compute bias as the sum of (forecast minus actual) divided by the sum of actual, keeping the sign. If over six months you forecast 7,200 and actual was 6,400, bias is (7,200 - 6,400) / 6,400 = +12.5%, meaning you systematically over-forecast by an eighth. A tracking signal, the running sum of errors divided by mean absolute deviation, flags bias when it exceeds plus or minus 4. The Forecast Bias calculator returns both the percentage and the tracking signal.

Demand variability sets how much buffer you actually need, and the input is the coefficient of variation. Take the standard deviation of period demand and divide by mean demand. A SKU averaging 500 units per week with a standard deviation of 75 has a CV of 0.15, which is smooth and forecastable. One averaging 500 with a standard deviation of 450 has a CV of 0.90 and is effectively lumpy. Use at least 24 to 52 weekly buckets of shipment history so the standard deviation is stable. The Demand Variability calculator computes CV and flags SKUs above 0.5 as candidates for a different planning method.

Safety stock ties variability and lead time together. The classic formula is Z times the standard deviation of demand during lead time. Z comes from the target service level: 90% is 1.28, 95% is 1.65, 98% is 2.05. If weekly demand standard deviation is 75 units and lead time is 4 weeks, demand-during-lead-time standard deviation is 75 times the square root of 4, which is 150 units. At a 95% target, safety stock is 1.65 times 150 = 248 units. The Inventory Buffer from Forecast Error calculator drives this off residual error rather than raw demand, which is more honest when your forecast already removes seasonality.

Convert forecast error into the buffer it justifies, because raw demand variability double-counts what your forecast already predicts. The cleaner input is the standard deviation of the forecast error itself: the spread of (actual minus forecast) over the trailing 12 months. If those residuals have a standard deviation of 90 units per week over a 3 week lead time, error during lead time is 90 times the square root of 3, which is 156 units, and a 95% buffer is 1.65 times 156 = 257 units. Feeding residual sigma rather than demand sigma typically cuts safety stock 15% to 30% when a forecast adds real value.

Close the loop by checking demand plan attainment, which tells you whether the agreed plan actually happened. Attainment is actual demand divided by the consensus plan for the period, expressed as a percentage. If the S&OP consensus locked 18,000 units for the quarter and 16,650 shipped, attainment is 92.5%. Track it alongside accuracy so you can separate a good forecast that the business overrode from a bad forecast the business followed. The Demand Plan Attainment calculator splits this by product family, and pairing it with the Supply Demand Gap calculator shows where the plan broke on the supply side versus the demand side.

Published 2026-07-01.