The reason to work this way is that a single number hides the thing executives most need to know: how much the outcome can move and what moves it. Two quarters can both forecast $10.4M while one has a $600K spread and the other has a $4M spread because three enterprise deals decide everything. Those are different businesses and they demand different decisions. The point estimate treats them as identical.
Building the distribution
Start at the deal level. Each open opportunity gets a close probability and an expected value, then a simulation draws thousands of quarters by resolving every deal against its probability and summing the winners. The spread of those simulated totals is the distribution.
Two properties make this more honest than a weighted sum. Deal concentration shows up as a fat, lumpy curve, because a quarter carried by four large deals really does have a wide range of outcomes. Correlation shows up too. Deals in the same segment tend to win or lose together when a market moves, and simulating them as correlated widens the range in a way a stage-weighted average never will.
Calibrating the deal probabilities
The simulation is only as good as the probabilities feeding it, and stage-based probabilities are usually wrong. ORM's data makes the scale of the problem clear: of the pipeline carrying close dates inside the quarter on day one, about 20% actually closes in that quarter, which means 80% of the value sitting in the quarter does not land in it. A model that assigns 60% to every late-stage deal will produce a confident distribution centered in the wrong place.
Calibrate against your own closed history instead. Group deals by attributes that predict close timing and take the observed close rate for each group as the probability. ORM's models do this by grouping every opportunity and fitting a close-timing curve per group, with most of the expected closing weight landing before week 12 and very little past week 52. Probabilities built from observed outcomes hold up. Probabilities inherited from a stage-name default do not.
What changes on the forecast call
The conversation moves from one number to the shape of the range. A wide distribution is a concentration problem or a data problem, and both have owners. A distribution whose floor sits below plan tells you the quarter needs more created pipeline now, not a better call in week 11.
It also changes accountability. Committing at a fixed percentile makes the forecast falsifiable across quarters, because you can check whether actuals land inside the stated band as often as the band claims. That is a real test of forecast accuracy, unlike a point estimate that is simply high or low. Read the distribution against pipeline coverage and the two together explain both how much pipeline exists and how reliably it converts.
Frequently Asked Questions
What is the difference between a probabilistic forecast and a weighted pipeline?
A weighted pipeline multiplies each deal by a stage probability and sums the result, producing one number that describes an outcome nobody expects to happen. A probabilistic forecast keeps the deals separate and simulates which combinations close, producing a range. The weighted sum is roughly the mean of that distribution, with all the shape information thrown away.
What do P10, P50 and P90 mean?
They are points on the outcome distribution, stated as exceedance levels. P90 is the level you exceed nine times in ten, so it is the conservative floor, and P10 is the level you reach one time in ten. P50 is the median, where half of outcomes land above and half below. Naming them this way removes the ambiguity in words like commit and best case. Some tools label the same points by percentile and invert the two, so confirm which convention a number uses before comparing.
Does a probabilistic forecast need Monte Carlo simulation?
Simulation is the most common method because it handles deal concentration and correlated outcomes cleanly. You can approximate a distribution analytically when deal sizes are uniform and outcomes are independent, but enterprise pipelines rarely satisfy either condition, and simulation is cheap enough that the shortcut is not worth it.
How do you commit a single number from a distribution?
Pick a conservative percentile, hold it constant across quarters, and carry the gap up to the median as upside. The specific level matters less than never moving it, because a percentile that shifts to fit the desired answer stops being a forecast.
Put these metrics to work
ORM builds custom revenue forecast models that turn concepts like probabilistic forecasting into prescriptive action for your team.
Schedule a Demo