From one date to a distribution

1. Add uncertainty where it lives

You don't need estimates on every task. Pick the handful whose durations you'd honestly describe as "it depends" — integration, certification, anything waiting on a third party — and give each three numbers: optimistic, most likely, pessimistic. Tasks without estimates simply keep their planned duration in every iteration.

2. Run it

From a scenario, choose Simulate, pick an iteration count (5,000 is plenty to start), and launch. Each iteration draws a duration for every uncertain task, recomputes the whole schedule through the same CPM engine that powers your Gantt, and records the finish. The run's random seed is stored with the results, so any simulation can be reproduced exactly.

3. Read the S-curve

The cumulative curve answers one question: "What's the chance we finish by date X?" Read up from a date to the curve, across to the probability. Two markers matter most:

• P50 — the coin-flip date. Half your simulated futures finish by here.
• P80 — the commitment date. You'd beat it in 80% of futures; this is the number to put in front of a customer.

The gap between your deterministic finish date and the P80 is the size of the optimism built into your plan. Most people find it uncomfortably large the first time — that's the point.

Explaining the chart to your boss

"The single date we used to report assumed every estimate lands exactly. The curve shows what happens when they vary the way they actually vary. We can commit to the P80 with our eyes open, or talk about what it would take to pull it left — that conversation is the scenarios guide."