The two calculations
Take signed error for each period as forecast minus actual.
Bias is the average of those signed errors. Positive bias means the forecast runs high. Variance is the spread of the same errors around that average, usually reported as the standard deviation so it stays in dollars.
Errors of +$400K, -$380K, +$420K and -$390K average to +$12.5K, which reads as almost no bias. Their spread is roughly $400K, which is the number that matters. Errors of +$180K, +$200K, +$170K and +$210K average to +$190K with a spread near $16K, and that forecast is the more useful one despite the larger bias.
The four combinations
| Low variance | High variance | |
|---|---|---|
| Low bias | Trustworthy. Plan against it. | Right on average, wrong every quarter. |
| High bias | Predictably off. Calibrate and it becomes usable. | Broken. Fix inputs before touching the lean. |
Bias is the cheaper problem
Bias is arithmetic. Measure the signed average over four to eight quarters, apply the correction, then verify it disappeared. The underlying causes are known and finite. Entry criteria let weak deals into commit, and deal amounts sit above what those deals actually close for.
Variance is structural. It rises when the forecast stops tracking what the business is doing. ORM's read on why forecasts fail is that something in the business or the market changed and the forecast is still built on old assumptions. New pricing pressure pulls deal sizes down, a territory change disrupts execution while pipeline still looks full, and a static model absorbs none of it. The result is not a steady lean. It is a forecast that lands somewhere different every quarter for a different reason.
Reading them together
Report bias and variance on the same page, over the same window, at both company and segment level. Segment-level reads matter because opposing biases cancel in the roll-up. Enterprise running 20% high and SMB running 20% low nets to a company bias of zero and two broken forecasting processes.
Then act on them separately. Correct bias with calibration and correct variance with better inputs and a model that updates as conditions move. Track the pair against your forecast accuracy baseline and fold both into the sales forecasting review cadence.
Frequently Asked Questions
What is the difference between forecast bias and forecast variance?
Bias is the average signed error, so it captures whether the forecast leans high or low. Variance is the spread of those errors around their average, so it captures how erratic the forecast is. A forecast that misses 5% high every quarter has high bias and near zero variance.
Can a forecast have zero bias and still be untrustworthy?
Yes, and this is the most common trap in accuracy reporting. A forecast that lands 15% high one quarter and 15% low the next averages to zero bias while being wrong by 15% every single quarter. The bias number looks perfect and the forecast is unusable.
Which is easier to fix, bias or variance?
Bias, by a wide margin. A consistent lean can be measured from four quarters of history and removed with a calibration factor. Variance comes from inputs and from a model that is not tracking current conditions, and reducing it requires fixing those sources.
How do you measure both at once?
Compute the signed average error for bias and the standard deviation of those signed errors for variance, over the same four to eight periods. Report them side by side, because either number read alone will mislead you about the state of the forecast.
Put these metrics to work
ORM builds custom revenue forecast models that turn concepts like forecast bias vs forecast variance into prescriptive action for your team.
Schedule a Demo