How to forecast revenue when deal sizes vary widely
When deal sizes vary widely, weighted pipeline stops working — not slightly, but completely. The forecast converges on a number the quarter can barely produce, because a single large deal makes the outcome roughly binary and an average sits in the gap between the two things that can actually happen. The fix is to stop reporting a point and report the two outcomes.
Two pipelines, the same forecast, nothing alike
Both pipelines below hold eleven deals, all at 50%, and both produce an identical weighted pipeline. One is eleven deals of 50.000 €. The other is one deal of 400.000 € and ten of 15.000 €.
Both against a 275.000 € target. Computed with the same engine The Pipe Review runs.
The weighted number is the same for both and describes neither. The even pipeline clusters tightly around it, which is what an average is for. The lumpy one spans 430.000 € — it is really two quarters wearing a trench coat, one where the big deal lands and one where it does not, and the middle is the least likely place to end up.
An average of eleven comparable things is informative. An average of one thing and ten rounding errors is a number that sits between the only two outcomes that can occur.
Why value is squared
The uncertainty a deal contributes to a quarter is p × (1 − p) × value². The squared term is the part that surprises people: a deal twice the size does not carry twice the doubt, it carries four times. Ten times the size is a hundred times the doubt.
That is why one large deal dominates so completely, and why coverage ratios mislead in lumpy pipelines. In the second pipeline above, the big deal alone carries 99% of all the uncertainty in the quarter. The other ten deals together carry the rest.
What to do instead
- Report the two cases, not the average. "With the big one we land near 490.000 €; without it, near 60.000 €" is a sentence your CEO can plan around. The weighted figure is not, because no quarter ends there.
- Forecast the large deal separately. Below some size, deals genuinely do average out and weighted pipeline is fine for them. Above it, each deal deserves its own conversation. Pick a threshold — a useful rule is any deal worth more than a fifth of the target — and treat those individually.
- Stop rounding big deals to 50%. On a small deal a lazy 50% costs almost nothing. On the deal carrying 99% of your uncertainty it is the whole forecast, and it deserves a real answer rather than a default.
- Check how concentrated you are before deciding how much to trust any single number. Sum p × (1 − p) × value² across open deals and look at the top three as a share. Under about half, averages are reasonable. Over it, you are forecasting a coin flip and should say so.
None of this requires new software. A spreadsheet column for p × (1 − p) × value², sorted descending, tells you in thirty seconds whether your quarter is an average or a bet.