Why your sales forecast is always wrong, and how to fix it

A sales forecast is usually wrong for three reasons, and none of them is that your reps are lying. The probabilities come from the stage rather than the deal, the arithmetic averages too few deals to mean anything, and nobody writes down what was forecast so the error is never measured. Fix the third one first: it is the cheapest, and it tells you how much the other two are costing you.

That is the short answer. The rest of this is why each one happens and what to do about it, with the arithmetic shown rather than asserted.

Reason one: the probability is about the stage, not the deal

In most CRMs, a deal's probability is attached to its stage. Everything in Negotiation is 70%. Everything in Proposal is 40%. That number is either the historical close rate of deals that have passed through that stage, or — more often than anyone likes to admit — a figure an administrator typed in during implementation and nobody has revisited since.

Either way it is a statement about your company's history in aggregate, applied flat to an opportunity it knows nothing about. It does not know the champion left last month, that procurement has not been engaged, or that the last four deals of this shape took two quarters instead of one.

A stage probability answers "what usually happens to deals here". A forecast needs an answer to "what will happen to this deal". Those are different questions, and only one of them has a person who could actually answer it.

The fix: make the rep put their own number on each deal, and then score them on it later. Not to catch anyone out. So that after six months you can say that one rep's 70% closes about 70% of the time and another's closes 40%, and weight their calls accordingly. Nobody argues with their own track record.

Reason two: you are averaging too few things

Weighted pipeline — value times probability, summed — is an average. Averages are reliable when you have many comparable items and no single one moves the total. Insurance works this way. A quarter with eight open deals, one of them worth more than the three smallest combined, does not.

The single number is not wrong. It is a reasonable estimate of the middle. But the quarter can land anywhere across a 530.000 € span, and a forecast that reports the middle while hiding the span is reporting the least useful half of what it knows.

The fix: report a range and the odds against target, not a point. "We are at 46% to make the number" is a sentence a CEO can plan around. "We are at 451.000 €" is not, because it conceals whether that is comfortable or desperate.

Reason three: nobody writes down what was forecast

This is the one that makes the other two permanent. Almost no company records what it forecast at the start of a month and compares it to what closed. Without that record there is no error to analyse, no direction to the error, and no way to tell a bad quarter from a bad forecasting process.

And the error usually has a direction. Optimism is not random; it is a bias, and a bias repeats. Here is the same pipeline as stated by a team running 25% hot — every open probability inflated by a quarter:

A quarter of optimism per deal becomes 93.500 € of phantom pipeline and turns the odds of making the number from 46% into a confident-sounding 66%. Nobody lied. Everyone rounded up slightly, in the same direction, as people do.

A bias you can measure is a bias you can subtract. That is the entire reason to keep the record: not to prove anyone wrong, but to turn a vague sense that you run hot into a number you can correct by.

How to fix it, in order of cost

• Keep the record. One row per month: what you forecast, what closed. Twelve rows will tell you more about your forecasting than any new report. This costs nothing and is almost never done.
• Replace stage probability with rep judgement, then score the rep rather than the stage. The number gets worse before it gets better, and then it gets much better.
• Report a range and the odds, and name the two or three deals carrying most of the uncertainty. Those are the only deals whose movement changes the answer.
• Correct for your measured bias. If you have run 20% hot for six months, the honest forecast is this month's number minus 20%, and you should say so out loud rather than hoping this month is different.

I ran this on my own numbers for a year. My forecast error went from roughly 50% to about 14%, and not because I got better at reading deals. I kept a record of how I had guessed, found I was high by a similar margin nearly every month, and subtracted it.