Forecast Accuracy vs Forecast Bias

What is the difference between forecast accuracy and forecast bias?

Forecast accuracy measures how far your forecast lands from what actually happened, and forecast bias measures which direction it leans when it misses. Accuracy is about the size of the gap. Bias is about the sign of the gap. They are separate measurements, and a forecast can score well on one while failing the other. If you only track accuracy, you can miss a systematic lean that quietly costs you every quarter. If you only track bias, you can call a forecast healthy while it swings wide in both directions.

Both come from the same raw material, forecast error, the difference between what you predicted and what closed. Accuracy summarizes the magnitude of that error. Bias summarizes its direction across many forecasts. Set them side by side and the difference is clean.

What does forecast accuracy measure?

Forecast accuracy tells you how close your prediction landed to the actual result, expressed as the average size of your error. The common form is 100% minus the mean absolute percentage error, or MAPE. If your forecasts are off by an average of 8% in either direction, you are running about 92% accurate. Direction does not enter the calculation. A quarter you called 10% high and a quarter you called 10% low both count as 10% of error.

Accuracy answers a confidence question. When a forecast says $4.2M and your model runs 95% accurate, you can plan on that number landing inside a tight band. When accuracy drops to 80%, the same $4.2M carries a wider spread and you hold more buffer against it. Forecast accuracy is the number you quote when someone asks how much weight the forecast can bear.

On new and expansion business, a carefully built forecast usually lands around 90% accurate, and reaching that by hand is slow and drifts out of date as conditions move. At ORM the target is 95%, held without manual adjustment from day one of the quarter through day 90, because the model re-reads the quarter as it progresses instead of waiting for someone to rebuild the spreadsheet.

What does forecast bias measure?

Forecast bias tells you whether your forecast leans in a consistent direction, over-calling or under-calling results across many periods. Compute it as the average of your signed errors, forecast minus actual, over several forecasts. A number near zero means your misses cancel out. A persistently positive number means you over-forecast, the optimistic lean of a commit that always looks better than it closes. A persistently negative number means you under-forecast, the sandbagging lean of a team that likes to beat its own call.

Bias is a direction, so a single period cannot prove it. You read it as a trend. A tracking signal, the running sum of your errors divided by the average error size, is the standard way to watch it. When that signal drifts past a threshold and stays there, the forecast is leaning, and it will keep costing you in the same direction until you fix the cause.

Bias deserves its own metric because it hides inside a decent-looking average. Sales forecasting reviews often praise a forecast that came in close for the year while every quarter ran 10% hot, saved only by errors canceling on the annual line. That is not a forecast working. That is two problems disguised as none.

Can a forecast be unbiased and still be wrong?

Yes, a forecast can carry zero bias and still be badly inaccurate, which is exactly why you need both numbers. Take two teams forecasting a business that closes $1.0M every quarter. Measured quarter by quarter, both teams are 80% accurate. Each one misses by $200K every quarter, a 20% error every time. Their bias could not be more different. Team Blue is off by +$200K in the same direction all year, a +20% bias, a textbook optimistic lean. Team Gold's errors cancel to zero bias, so on the annual line Gold looks perfect while it was 20% wrong in each of the four quarters.

The two problems call for different fixes. Team Blue's lean is correctable. Shave the known 20% off the commit and the forecast snaps into place, because the error is systematic. Team Gold has no lean to shave. Its problem is noise, and noise only comes down with better inputs and a tighter model. Bias told you Blue's story. Accuracy told you Gold's. Neither number alone would have.

What causes forecast bias?

Most forecast bias comes from a model built on assumptions that no longer match the market, plus the human habits of the people entering the numbers. The most common reason a forecast misses is that something in the business or the market changed and the forecast is still running on old assumptions. A new competitor pressures pricing and average deal size falls. Interest rates climb, so buyers slow down and win rates slide. The model keeps forecasting the old world, so it leans the same wrong way every week until someone resets it.

Behavior adds a second, steadier lean. Reps who sandbag pull the forecast low on purpose, and commit inflation pushes it high. Unmodeled seasonality adds a lean of its own, since the third month of a quarter usually closes stronger than the first two, so a model that ignores that shape under-calls early and scrambles late. Pipeline coverage that looks healthy can mask all of this, because coverage counts dollars without judging whether they close at the value and in the direction the forecast assumes.

When should you focus on accuracy or bias?

Fix bias first, then chase accuracy, because a directional lean is both cheaper to correct and more damaging to leave alone. Bias is systematic. Once you know a forecast runs 15% hot, you can remove the lean immediately and stop the same error from repeating next quarter. Accuracy is harder won, since closing the gap on a noisy but centered forecast means better data and a tighter model, not a single correction.

Read bias when you are auditing a process, like comparing reps or checking whether a change in territory or win rate has thrown your assumptions off. Read accuracy when you are sizing a single decision, like how much to spend or how much pipeline to hold against the number. A forecast you can trust is centered first and tight second. If you want the full method, how to build a sales forecast walks through the inputs that keep both numbers honest. At ORM we build the models that catch a lean the moment conditions move, so the forecast stays centered before the quarter gets away from you.