A weather forecast used to be a single predicted outcome, tomorrow will be 18 degrees and dry. Meteorologists largely abandoned that approach decades ago, not because it was too hard to calculate, but because it implied a certainty the atmosphere doesn't actually have. Small, unavoidable errors in the starting measurements compound quickly in a chaotic system, so a single deterministic run can be confidently, precisely wrong.
Most B2B revenue forecasting still works the way weather forecasting used to. A single number goes to the board, "we'll close £2.4 million this quarter," built from assumptions that are individually reasonable but collectively just as capable of compounding into a confident, precise miss.
The problem with a single confident number
A single-point revenue forecast collapses several genuinely uncertain inputs, win rate, average deal size, the timing of a handful of large deals, into one figure. Each input on its own has a real range of plausible outcomes. Forcing them all into a single midpoint estimate doesn't remove that uncertainty, it just hides it from whoever reads the final number.
A single-point forecast isn't more accurate than a range, it's just more confident-sounding. The uncertainty was always there. A range simply shows it instead of hiding it.
How ensemble forecasting actually works in weather prediction
An ensemble forecast doesn't run the model once. It runs the same model, or several different models, dozens of times, each with slightly varied starting conditions that are all individually plausible given what was actually measured. The result isn't one predicted outcome, it's a spread of outcomes, and the width of that spread is itself useful information: a tight cluster means high confidence, a wide spread means genuine uncertainty that shouldn't be hidden behind a single number.
Building the same approach for B2B revenue
The same logic transfers directly, because the same structural problem exists: a handful of inputs, each individually uncertain, get combined into a forecast that's typically reported as a single number.
A single enterprise deal slipping from this quarter to next can single-handedly move a forecast by a significant margin. Treating its close date as a fixed input, rather than a range spanning several weeks either side, is exactly the kind of false precision an ensemble approach corrects for.
A quarter with 20 qualified opportunities and a "typical" 25% win rate could plausibly close anywhere from 3 to 8 deals purely from normal sampling variation, before accounting for anything unusual happening in the market. A single win rate figure hides that spread entirely.
Deals cluster around fiscal year-ends and budget cycles in ways a flat, evenly-distributed forecast doesn't capture. An ensemble approach that varies the timing assumption directly reflects this instead of averaging it away.
Presenting a range without losing the board's confidence
The obstacle to adopting this isn't usually technical, it's that a range can feel like hedging to a board used to a single confident number. The fix is in how it's framed, not whether to do it.
- Vary the two or three inputs with the most genuine uncertainty: Usually win rate, average deal size, and the timing of the largest few deals. Run the pipeline math across plausible combinations rather than a single best guess for each.
- Report low, expected, and high cases, not just a range: A named "expected case" gives the board the single figure they're used to, with the low and high cases showing the genuine spread around it rather than replacing it entirely.
- Track how often actuals land inside the stated range: A well-calibrated range should contain the actual result most of the time. If actuals consistently fall outside it, the range itself needs recalibrating, not just the midpoint.
- Explain the range as rigour, not uncertainty about the process: Frame it the way meteorologists frame a hurricane cone, this reflects a more honest model of the world, not a weaker one. A range that's well calibrated is more trustworthy than a single number that's precisely wrong.
- A single-point revenue forecast hides genuine uncertainty rather than removing it, exactly the problem weather forecasting solved with ensembles decades ago
- Win rate on a small sample, large deal timing, and seasonal effects are the biggest sources of hidden uncertainty in most B2B forecasts
- Vary the two or three most uncertain inputs and report low, expected, and high cases instead of one figure
- Track how often actuals land inside the stated range to check whether the range itself is genuinely well calibrated
Our attribution modelling work includes pipeline forecasting built on ranges, not single numbers, so a quarter that comes in at the edge of the range looks like a well-understood outcome rather than a missed forecast.
Read The Bullwhip Effect →