Why deterministic plans always lie
A capital project plan that says "this will cost $200M and take 24 months" is asserting something that almost never happens. The cost is a distribution. The schedule is a distribution. The deterministic numbers in the plan are the project team's central estimate — usually optimistic, usually missing the long-tail risks that historical data say will eventually bite.
Probabilistic project planning fixes this by making the uncertainty explicit. Instead of "$200M and 24 months," the plan says "P50 cost is $215M, P80 is $260M, with the dominant variance drivers being long-lead vendor delivery and engineering productivity." That's the data the capital committee actually needs.
How probabilistic planning actually works
Capital Project AI combines three established methodologies — Monte Carlo simulation, PERT-style three-point estimation, and reference-class forecasting — and runs them at the speed of a live planning meeting rather than as a once-per-stage-gate exercise.
Monte Carlo simulation
For each cost line and schedule activity, the platform runs thousands of randomized scenarios drawing from the input distributions. The output is the full probability distribution of project outcomes — not a single number, but P10, P50, P80, P95 and the variance drivers behind each.
PERT three-point estimation
Each input gets an optimistic, most-likely, and pessimistic value rather than a single point estimate. PERT-style distributions capture the asymmetry that real activities have — most things take a bit longer than planned, occasionally a lot longer, almost never less.
Reference-class forecasting
The platform finds projects in your historical record (or the broader reference dataset) that are structurally similar to the one you're planning, and uses their actual outcome distributions as a sanity check on the bottom-up estimate. When the bottom-up estimate disagrees with the reference class, that's a signal the planning team is missing something.
Sensitivity and driver analysis
Knowing the P50 cost isn't enough. The platform decomposes the variance into drivers: which inputs explain most of the spread between P50 and P90? Long-lead deliveries? Engineering productivity? Permit timing? You then know exactly which risks are worth buying down.
Why Capital Project AI
- Probabilistic, not just stochastic. Most schedule-risk tools run Monte Carlo on guessed distributions. We calibrate the inputs against reference-class data and your historical performance.
- Fast enough to use live. Run a probabilistic plan in under a minute. Re-run with revised assumptions during the planning meeting itself.
- Driver decomposition built in. Every output is paired with the variance drivers behind it — so you know which risks to spend management attention on.
- Built by an ex-Shell capital owner. Founded by an engineer who has lived through the gap between deterministic plans and actual outcomes at $800M megaproject scale.
Build a probabilistic plan in minutes
Upload your schedule and cost estimate — get the full outcome distribution and the driver decomposition in under a minute.
Open the Dashboard →What it looks like in practice
A power utility is planning a $400M transmission build. The bottom-up estimate from the project team is $410M with a 32-month schedule. The board wants to know how much contingency to fund and at what cost.
Capital Project AI runs the probabilistic plan. Monte Carlo across the schedule produces a P50 of $435M and a P80 of $498M. The reference-class comparison against 14 similar transmission builds in the historical dataset shows a slight optimism bias in the engineering productivity assumption — closing that gap moves the P50 to $445M. The variance decomposition identifies three dominant drivers: right-of-way acquisition timing (drives 31% of the variance), structural steel delivery (24%), and substation commissioning (18%). The recommendation: sanction at $475M (P70), with $30M of management reserve held against the three identified drivers, and an early-action plan for accelerating the right-of-way acquisition that buys down 12% of the total variance for $4M of incremental spend.
The same engine drives AI capital project risk management and the EPC bid analysis at AI for EPC contractors. For owners optimizing across the portfolio, see project portfolio optimization with AI.
Frequently asked questions
What is probabilistic project planning?
It replaces single-point estimates of cost and schedule with distributions that reflect uncertainty. Instead of "this project will cost $200M and take 24 months," you get the full P10 / P50 / P90 outcome range — and the drivers behind each.
How is this different from traditional schedule risk analysis?
Traditional SRA runs a Monte Carlo over a CPM schedule, but the input distributions are usually guesses. Capital Project AI calibrates the input distributions against reference-class data and your historical productivity — so the output is a forecast, not just a math exercise.
What is reference-class forecasting?
Reference-class forecasting predicts a project's outcome by comparing it to a class of similar past projects rather than relying on the project team's own bottom-up estimate. It's the technique that consistently produces the most accurate megaproject forecasts in the academic literature.