The forecast says next year's revenue will be €4.2 million. It is a confident, presentable number — and it is quietly ambiguous. Is it the average of what might happen? The single most likely outcome? The figure someone needs to hit to make budget? Each reading implies a different probability of actually reaching it, and the slide does not say which one it is.
That ambiguity is not an accident of sloppy analysts; it is what lets point forecasts survive. A number with no stated probability cannot be held to account: if the year comes in at €3.6M, the forecast wasn't exactly wrong — it was just a number. The hidden figure — the one the forecast never shows — is the probability that the number happens at all.
Often much lower than anyone in the room assumes. When a forecast is assembled the usual way — most-likely values for sales volumes, prices and costs, added and multiplied through the spreadsheet — the result inherits a systematic optimism, because business outcomes are skewed: there are more ways to sell less than planned than to sell dramatically more, more ways for costs to overrun than underrun. The peak of a skewed distribution is not its average, and combining peaks drifts the total away from what actually happens on average.
A Monte Carlo simulation of the same spreadsheet makes the hidden number visible: replace the uncertain cells with realistic ranges, recalculate thousands of times, and count the fraction of outcomes at or above the forecast. Below, a simulated NPV distribution with its cumulative curve — the markers read off the probability of any outcome level directly, including the ones below zero that the point forecast never mentioned:
Because the arithmetic betrays them. Most-likely values do not add: the sum of the most likely value of every line is neither the most likely total nor the average total. The same trap, in mirror image, is why cost estimates need calculated contingency rather than summed conservative values — percentiles and modes are properties of whole distributions, and they refuse to pass through a plus sign.
Incentives finish the job. A forecast is often a negotiation — between what the analyst believes, what the manager needs, and what the board wants to hear — and a single number gives the negotiation nowhere visible to happen. A distribution is harder to lean on: you can move a target, but you cannot quietly move the probability of hitting it without the chart changing shape in front of everyone.
Conversations change first. "The forecast is €4.2M" becomes "there is a 35% chance of reaching €4.2M, and a 90% chance of at least €3.4M" — and suddenly the room is discussing which commitment to make at which confidence, which is the discussion that was always needed. Targets can be set deliberately ambitious knowing they are ambitious; banking covenants and cash planning can use the conservative end; nobody is surprised by an outcome the distribution plainly contained.
Priorities change second. The simulation's sensitivity ranking shows which uncertainties actually move the outcome — so effort goes into firming up the two assumptions that matter instead of polishing the twenty that don't. And when someone proposes an action to improve the year, re-running the simulation shows what it does to the probability, not just to the headline number.
The mechanics are not the hard part: an existing Excel forecast becomes a simulation model by replacing uncertain cells with distributions — ModelRisk does this in the workbook you already have, at most €1,550 per user per year, and the free business cashflow model shows the pattern on a full example. The hard part is only the decision to look.
Usually nothing precise — it could be a mean, a most-likely value, or an aspiration, and each implies a different probability of being reached. Until the probability is stated, the number cannot be held to account.
Replace the forecast's uncertain inputs with realistic ranges, simulate with Monte Carlo, and read off the fraction of outcomes that reach the target. That fraction is the number the point forecast was hiding.
They answer different questions, and for skewed outcomes they differ materially. The honest answer is to report the distribution — or at minimum a central value together with its probability of being achieved.
Because most-likely values do not add: summing the peaks of skewed distributions gives a total that is neither the most likely total nor its mean. Only the whole distributions produce an honest total.
A Monte Carlo add-in. ModelRisk turns an existing Excel forecast into a simulation model — uncertain cells become distributions, and the results report the probability of any target directly, together with the inputs that drive it.
Find the number your forecast is hiding: put ranges on the cells you're unsure about, simulate, and read the real probability of your target — in the 15-day free trial.