Representations of Uncertainty

Risk technique · See it on the map

Guide only — A distribution-shape picker could serve this.

Describing an uncertain amount as a range instead of one falsely precise number.

When to use it

Whenever a single-point estimate for cost, duration, or probability is being presented as if it were exact, and it isn't — which is most estimates. Useful input wherever a technique downstream (sensitivity analysis, decision tree analysis) needs a range rather than a point to work with.

When to avoid it

When the extra precision of a full distribution (shape, spread, tails) buys nothing over a plain three-point range, because nobody downstream is going to do anything with the difference. Also avoid dressing up a genuine guess in distribution language — a triangular distribution built from three numbers someone made up in a meeting is still three numbers someone made up in a meeting.

Steps

1. For the quantity in question, gather optimistic, most likely, and pessimistic estimates from whoever is closest to the work.
2. Choose a shape that matches how the uncertainty actually behaves — triangular for a simple three-point estimate, a wider or skewed shape if you have reason to think bad outcomes are more likely than good ones by a large margin.
3. Write the range and shape down next to the estimate, not just the single number that would otherwise get quoted in a status report.
4. Use the range as the actual input wherever the estimate feeds a decision — don't collapse it back to a midpoint the moment it leaves the estimator's hands.

What it produces

• An optimistic/most-likely/pessimistic range (or a described distribution shape) for the quantity, instead of one number.
• A record of who supplied the estimate and on what basis, so it can be revisited.

Common pitfalls

• Reporting the range once, then reverting to quoting only the most-likely figure everywhere else because it's simpler to write down.
• Confusing a wide range with a careful one — width should reflect genuine uncertainty, not a padding reflex applied to look cautious.
• Building a distribution nobody actually samples from. Driftless has no field for a distribution on a risk or estimate today; if you need this to feed an actual simulation, that simulation has to happen outside the tool — the only Monte Carlo Driftless runs is a schedule forecast over past sprint velocity, unrelated to any risk range you define here.

Worked example

An engineering team estimating a custom sensor housing gives one number, '6 weeks,' in the first plan. Pushed for a range, the lead engineer gives 4 weeks optimistic (parts arrive on time, first print works), 6 most likely, and 11 pessimistic (a redesign is needed after the first fit test). The 11-week tail, not the 6-week midpoint, is what changes the delivery commitment the PM makes to the client.

Source

• PMBOK-6 §11.4.2.4

Where it comes from: this technique is named by the PMBOK Guide, 6th edition.