Three-point estimates: setting ranges people can defend
A three-point estimate gives an activity's duration or cost as a minimum, a most likely value and a maximum. It is the input every schedule and cost risk analysis depends on, and the place most analyses go wrong: ranges are set too narrow, all the same, or without any record of why. Here is how to set them well.
The three points
- Minimum: the shortest the work could reasonably take if things go well. Not a miracle.
- Most likely: the single most probable value, usually but not always the planned duration.
- Maximum: the longest it could reasonably take from the uncertainty in the work itself, before discrete risk events, which belong in the risk register.
Keeping uncertainty (the normal spread of the work) separate from risk events (things that may or may not happen) avoids counting the same problem twice.
Distribution shapes
| Shape | Use it when |
|---|---|
| Triangle | The three points are all you know. It gives the extremes real weight, so it is the more cautious choice. |
| BetaPERT | You have a confident most likely value. It concentrates probability near it and thins the tails. |
| Uniform | Any value in the range is as likely as any other, for example a delivery window with no better information. |
Setting ranges well
- Ask the people who do the work, one activity or group at a time, and ask for the maximum first: it reduces anchoring on the plan.
- Use history: how did similar work actually run against its plan? Past projects are the best antidote to optimism.
- Use rules for the rest: ranges by work type, by estimate class or by maturity of definition, applied consistently and refined by exception.
- Record the basis: who set the range, from what (a quote, records, judgment), and when. A range nobody can explain will not survive a review.
- Review and accept: ranges are inputs to a decision; they deserve the same review as the estimate.
The bias to watch for
Ranges are almost always too narrow and too symmetric. Work overruns far more than it underruns, so maximums should sit further from the most likely value than minimums do. A uniform ±10% on every activity is a sign the analysis has not been done, and produces a P80 barely later than the plan.
In DeltaSys
The DeltaSys Planner keeps each activity's range with its basis: its source (quick risk, a rule, expert judgment, a quote, history, an analogy), the reasoning, and its review status from draft to reviewed to accepted. Company range rules by estimate class apply to activities without their own range, and editing a range sends it back to draft for review. The risk analysis report states how well founded its inputs were.
Questions
Should every activity get a range?
Every activity that can drive the finish should. Small activities far from the critical path can share a rule-based range; the effort belongs on the activities with high criticality.
Primavera and P6 are trademarks of Oracle Corporation; Microsoft Project is a trademark of Microsoft Corporation; PMBOK is a trademark of the Project Management Institute; AACE is a trademark of AACE International; Last Planner is a registered trademark of the Lean Construction Institute. DeltaSys is not affiliated with or endorsed by them. Standards are described in our own words; see the publications themselves for their text.