Step one, compute the period error
``` Forecast error = Forecast - Actual ```
A $5.2M call against $4.8M closed gives positive $400K. The sign carries the meaning. Positive means the forecast ran high, negative means it ran low. Do not take an absolute value yet, because the sign is destroyed permanently once you do.
Step two, summarize twice
| Quarter | Forecast | Actual | Signed error | Absolute error |
|---|---|---|---|---|
| Q1 | $4.6M | $4.5M | +$100K | $100K |
| Q2 | $5.2M | $4.8M | +$400K | $400K |
| Q3 | $5.3M | $5.5M | -$200K | $200K |
| Q4 | $6.0M | $5.6M | +$400K | $400K |
| Average | +$175K | $275K |
The absolute average of $275K is the accuracy, known as mean absolute deviation. Convert it to a percentage by dividing each period's absolute error by that period's actual and averaging, which gives MAPE. For this run, MAPE is 5.3%.
Step three, choose the measurement point
The number changes depending on when you snapshot the forecast, so fix the point before you start. Day one of the quarter grades whether the team understood the quarter before it began. End of month two grades late-stage conviction.
Grading only the final call is the common mistake. A forecast that becomes accurate in the last week of the quarter has no operational value, because by then the quarter has already happened. ORM's position is that the value sits in knowing the shape of the quarter on day one, early enough to change it, and its models hold accuracy from day one to day 90 rather than converging late.
Step four, act on the two numbers separately
Bias and accuracy have different fixes. A persistent lean is a calibration problem you correct from history. A wide spread with no lean is an input problem, usually stale close dates or amounts that do not match what deals close for.
Run both numbers by segment and by rep, not only at the company roll-up, because opposing errors cancel in aggregate and hide where the process is failing. Track the results against your forecast accuracy baseline, and see how to create a sales forecast for the process that generates the calls being graded.
Frequently Asked Questions
What is the basic forecast error formula?
Forecast error for a period is forecast minus actual. A $5.2M forecast against $4.8M actual gives an error of positive $400K, meaning the forecast ran high. Keep the sign at the period level, because you need it for the direction calculation later.
Do you need more than one period to calculate forecast error?
One period gives you one error. It takes at least four periods to separate a lean from noise, since any single quarter can miss in either direction for reasons that will not repeat. Four to eight quarters is the practical window for a stable read.
Should you calculate forecast error against the original forecast or the final one?
Pick a fixed measurement point and hold it. Day one of the quarter and end of month two are the two most useful. Grading only the final submitted call flatters the process, because by the last week of the quarter the outcome is already decided.
How do you calculate forecast error at the rep level?
Same formula, applied to each rep's submitted call against their own actual. Use percentage error for cross-rep comparison so territory size does not decide the ranking, and roll dollar errors up separately for the team number finance will act on.
Put these metrics to work
ORM builds custom revenue forecast models that turn concepts like how do you calculate forecast error? into prescriptive action for your team.
Schedule a Demo