A single number conceals the uncertainty behind the estimate.
Confidence-aware planning asks for three numbers in place of one: best case, most likely case, worst case. They produce an expected estimate and keep the range and its assumptions visible where targets, commitments, funding, and capacity are decided.
- Type
- Methodology
- Method
- Three-point estimation / Trivariate method.
- Inputs
- Best case · most likely case · worst case, per work item.
- Status
- The bounded browser instrument follows a paper review of the calculation model. This page documents the method.
A single-point estimate hides three different quantities inside one figure: what happens if things go well, what usually happens, and what happens when the work meets material risk. Delivery work can be asymmetric: late-emerging dependencies, review cycles, rework, and shifting availability can extend the upper end of the range. A single number cannot express that uncertainty.
Every planning layer then treats that number as firmer than the layer below intended. Capacity plans are built on it, commitments are made against it, and sequencing locks around it. As the number hardens across planning layers, the portfolio can commit more capacity than the estimate supports, leaving delivery teams to absorb the difference.
Best, most likely, and worst each serve a distinct purpose. The worst case names the material downside. The best case names the conditions required for a favorable outcome. The spread records the range of uncertainty currently being expressed, and that range is information an executive can act on.
The conventional three-point treatment combines the inputs into an expected estimate, weighting the most likely case while incorporating both ends of the range. A companion approximation treats the spread as a rough standard deviation, a coarse confidence signal.
The expected-value formula and the uncertainty divisor are separate concepts: one locates the estimate, the other approximates its spread.
(best + 4 × most likely + worst) / 6
(worst − best) / 6
The default divisor of 6 reflects the conventional assumption that best and worst approximate a near-total range.
An organization can substitute its own calibrated divisor once its delivery history supports one. No alternative value is recommended here, and no internal calibration is published.
An estimate is analysis: the range the evidence supports today. A target is business intent: where leadership wants to land. A commitment is a promise an accountable leader makes with the range in view. Control is what happens after: actuals tracked against the promise, with variance driving correction while options still exist.
Planning goes wrong when the four collapse into one number. A target quietly becomes the estimate, the estimate becomes the commitment, and the uncertainty surfaces late, in delivery. Keeping the four distinct is what lets a leader commit confidently with the range in view.
The estimate, target, commitment distinction follows Steve McConnell’s Software Estimation: Demystifying the Black Art, the standard reference on calibrated software estimation.
- Range under uncertainty
- Commitment, human-owned
- Control: actuals vs promise
Capacity
A portfolio of wide ranges cannot be packed like a portfolio of point estimates. Ranges force capacity plans to hold explicit buffer before overtime discovers it.
Funding
Funding a wide-range initiative is a different decision from funding a narrow one. Sometimes the right first spend is the discovery work that narrows the range.
Sequencing
High-uncertainty work scheduled early can increase risk across dependent commitments. Ranges make that sequencing exposure visible before dates are finalized.
Commitments
A point estimate can conceal the confidence level behind an external commitment. Making that confidence explicit gives executives a clearer basis for the commitment.
The same link between long-range priorities and near-term funding and capacity ran at portfolio scale in delivered work.
A wide range is a prompt for evidence. Each phase of work should shrink the spread: a spike resolves a technical unknown, a dependency map removes a surprise, a first increment calibrates the team’s actual pace.
This is where confidence-aware planning changes executive decisions. “Six to fourteen weeks, most likely nine, wide because the integration surface is unmapped” invites a real choice: fund the discovery that narrows the range, commit at the wide end, or defer. A single “eight weeks” invites a yes and stores up a surprise.
A bounded, browser-based planning instrument will live on this page: enter best, most likely, and worst case; see the range, the expected estimate, the spread, and the capacity implication. It ships after the calculation model and decision-use language pass a paper review.
This methodology is stage 3 of the intake-to-decision system, the point where capacity realism enters the portfolio.