A smooth forecast line is not an accuracy check. Compare predictions with recent observations you deliberately keep beyond the forecast's information cutoff.

Use desktop Excel for Windows in Microsoft 365, Excel 2024 or Excel 2021. Preserve the original series. Choose a complete monthly history with a recorded value for each month and enough earlier history for the pattern you want to examine. Retain the final three known observations as the comparison period, and record the intended start date before inspecting the forecast.

Keep the cutoff explicit

Select the matching date and value series, then choose Data, Forecast Sheet. Open Options and set Forecast Start to the first comparison month. Microsoft documents that only observations before this start date are used for the prediction. Set Forecast End to include the comparison period and create the worksheet containing historical and predicted values.

The timeline needs consistent intervals. Check seasonality and missing-point or duplicate-timestamp treatment in Options. Manually specified seasonality needs at least two historical cycles. Use complete observations for the initial test.

Compare the same dates and units

For each of the three retained months, place the observed value beside its forecast and calculate forecast minus observed. A positive difference means the forecast is higher; a negative difference means it is lower. Read both the sign and the size before compressing the results into a single score.

As an arithmetic example only, suppose an observed workload is 40 and a forecast is 46. The difference is +6. Another month with 55 observed and 50 forecast has a difference of -5. Adding the two errors gives +1, which conceals that both months missed by several units. These numbers illustrate the comparison, not an Excel forecast produced for a real dataset.

Explain any operational threshold before judging the result. A six-unit difference may matter differently for staffing, stock or another use. Avoid changing the cutoff repeatedly just to select the most attractive comparison.

Record the source period, forecast start, relevant options and the individual differences with the candidate report. A short historical comparison can reveal a poor fit or a misunderstood setup; it does not guarantee future accuracy. Keep any later forecast using the full history clearly identified as a separate run, because its information cutoff has changed.

Sources: Microsoft: Forecast Sheet.