Cost Risk — Estimate Ranging with Monte Carlo

What Monte Carlo simulation is

The name comes from the casino: like spinning a roulette wheel over and over, the computer plays out the project thousands of times, each time drawing a slightly different value for every uncertain cost. One run might hit the high end on three items and the low on two; the next, the reverse. Collect all the totals and you get not a single number but the full range of what could happen — and how likely each outcome is.

How a Monte Carlo run works

• Range each uncertain item — Give every critical cost item a distribution (Lesson 11) — typically a triangular low / most-likely / high.
• Sample once from each — The computer draws a random value from each item's distribution — one plausible "version" of the project.
• Add up the total, and repeat — Sum the sampled items into one total cost. Then do it again — thousands of times (5,000+ is common).
• Read the distribution — The collected totals form a histogram and an S-curve. Pick your confidence level and read off the cost; contingency is the gap above the base.

Monte Carlo simulator

Three uncertain cost items, each with a low / likely / high range. This runs a real 5,000-iteration simulation in your browser and builds the distribution. Adjust the ranges and watch the percentiles move:

Reading a Monte Carlo result

The output isn't one number, it's a curve — so the key decision is which confidence level to fund to. P50 means you'll overrun half the time; many organizations fund cost contingency to P70–P80 for a sensible safety margin. Choose deliberately, document it, and report contingency as "P80 = $X" rather than a bare percentage. And remember the result is only as good as the ranges you fed in — garbage ranges produce a beautifully precise wrong curve.

Ten things to remember

1. Monte Carlo samples each uncertain input thousands of times.
2. It builds the full distribution of total cost — not a single number.
3. It needs ranged inputs (Lesson 10) with distributions (Lesson 11).
4. Thousands of runs converge to a stable, dependable distribution.
5. Output is a histogram and an S-curve of cumulative probability.
6. Default: base $750k, mean ~$833k, P80 $889k → $139k contingency.
7. Choose a confidence level to fund to — often P70–P80 for cost.
8. Report "P80 = $X," not a bare percentage.
9. Right-skew shows overruns are likelier than underruns.
10. Precision isn't accuracy — the curve is only as good as the inputs.