Forecast a distribution, not a point estimate

Represent your forecast as a range of likely outcomes, not a single predicted number.

Why it works

Point estimates communicate false precision and suppress useful information about variance. A project that takes six months on average might have a 10th-percentile outcome of three months and a 90th-percentile outcome of eighteen months — a distribution that should change planning decisions, resource allocation, and contingency budgeting in ways a point estimate conceals. Communicating distributions makes uncertainty legible and improves downstream decisions.

How to do it

1. For any significant forecast, report three numbers: optimistic (10th percentile of class), median, and pessimistic (90th percentile).
2. Weight the distribution asymmetrically if the reference class shows right-skewed distributions (most projects do).
3. Budget and plan for the median, but ensure the plan survives the 75th-percentile outcome.
4. Resist pressure to convert the distribution to a point — the point estimate destroys the planning-relevant information.

Evidence

Research on cost overruns and schedule slippage consistently finds right-skewed distributions — outcomes can be much worse than average but rarely much better. Planning to the median in a right-skewed distribution systematically underprepares. Interval forecasts are more accurate and more useful than point forecasts across multiple domains studied by Tetlock and colleagues. (observational)

Communicating distributions is harder socially — stakeholders prefer confident point estimates. The accuracy gain comes with a communication cost that must be managed explicitly.

Sources

• Flyvbjerg (2002), cost overrun distributions in infrastructure projects
• Tetlock & Gardner (2015), Superforecasting, Chapter 7

Common mistake

Treating the "optimistic" estimate as the plan and the "pessimistic" one as the contingency, when the reference class shows the pessimistic outcome is the modal experience, not the edge case.