How wrong is your forecast? Put a number on it.

Finance teams spend enormous effort producing forecasts and almost none measuring them. Ask how accurate last year's revenue forecast was and the honest answer, in most businesses, is a shrug and a feeling. The plan is filed, the actuals arrive, and the two are compared for one month before everyone moves on.

This is a strange gap, because forecast accuracy is easy to measure and it changes how much anyone should trust the next one.

Error and bias are two different things

Error is how far off you were, ignoring direction. Bias is whether you are consistently off the same way. A team that misses by eight percent in both directions has an accuracy problem. A team that misses by four percent and is under every single month has a credibility problem, and a much easier one to fix.

Track both, per line, for a rolling three months. Revenue, gross margin, headcount cost and the cash balance are enough to start. If the same line is wrong in the same direction three times running, the assumption behind it is not a forecast, it is a habit.

Decide what good looks like before you measure

Accuracy targets are business specific and they should be written down in advance, otherwise every result is retrospectively acceptable. A subscription business with low churn might reasonably hold revenue within three percent one month out. A project business with lumpy delivery might set ten and be doing well. The number matters less than having agreed it before you saw the answer.

• One month out: how good is the estimate you present to the board.
• One quarter out: how good is the number the business is planning against.
• The cash trough: the one that costs real money to get wrong.

Publish it

The instinct is to keep accuracy scores private, on the basis that they are embarrassing. The opposite is true. A forecast presented with a track record attached is far more persuasive than one presented with confidence and nothing behind it. Saying that this line has been within four percent for six months tells the room how much weight to put on it, and saying that this other line has been all over the place is useful information too.

Nobody expects the forecast to be right. They expect you to know how wrong it usually is.

Then fix the driver, not the number

When a line is consistently off, the fix is upstream. A revenue forecast that runs hot every month usually means the conversion assumption is optimistic, not that revenue needs a haircut at the end. Adjusting the output hides the error. Adjusting the driver removes it, and the next forecast is better without anyone having to remember the fudge.

Dash keeps forecast accuracy alongside the model rather than in a separate tracker, so the comparison happens automatically as actuals sync. The point is not the scorecard. It is that the model gets less wrong over time, and you can prove it.