Three-Point Estimation: What Good Looks Like, and What Goes Wrong
Ask a team how long a task will take and you usually get one number. That number is almost always the answer to a quieter question the estimator asked themselves: how long will this take if nothing goes wrong? On a capital project that runs for months, optimism compounds, and a schedule built on best-case numbers drifts late before the first month closes. Three-point estimation is the discipline of refusing to commit to a single figure when you do not yet have one.
The method is simple. For each work item you capture three durations: an optimistic estimate (O) if conditions are favourable, a most likely estimate (M) for the realistic normal case, and a pessimistic estimate (P) for when known risks materialize. Two common formulas turn those three numbers into an expected value: the triangular mean, (O + M + P) / 3, and the PERT weighted mean, (O + 4M + P) / 6, which leans toward the most likely figure. The gap between O and P is itself useful information — it tells you how uncertain the estimator really is.
What good looks like
A healthy three-point estimate is built by the people who will do the work, not handed down from a planner who has never touched the task. The three numbers come from distinct reasoning, not one guess nudged up and down by ten percent. And the pessimistic figure is grounded in named risks the team can point to.
Each point reflects a different scenario, with the assumptions written down beside it.
The pessimistic case names the specific risks — a late permit, a single-source supplier, a key person on leave — rather than padding a round number.
Wide O-to-P spreads are flagged for attention, not averaged away and forgotten.
Estimates are revisited as the work clarifies, instead of being frozen at the moment of the original guess.
What goes wrong
Bad three-point estimation tends to fail in recognizable ways. The most common is theatre: a team produces three numbers because the template demands them, but O, M and P are really the same estimate plus or minus an arbitrary buffer. Nothing is learned, and the resulting expected value carries false precision.
Anchoring on the most likely. The estimator writes M first, then sets O and P symmetrically around it. Real tasks are rarely symmetric — the downside is usually much larger than the upside, so honest distributions are skewed.
Hidden double-counting. The pessimistic figures already include risk, and then a separate contingency reserve is added on top, inflating the plan and eroding credibility when the buffer goes unused.
Summing pessimistic cases. Adding every task's worst case to get a project worst case assumes everything goes wrong at once. It rarely does; this overstates the tail and invites the schedule to be ignored.
One-and-done. The estimate is produced in a kickoff workshop and never touched again, even as scope and conditions change underneath it.
There is a deeper point here. The value of three-point estimation is not the tidy expected number it produces. It is the conversation it forces — about what could go wrong, how confident the team really is, and where the schedule is fragile. A plan that says a task will take twelve days tells you nothing about risk. A plan that says eight days if the permit clears, twelve in the normal case, and twenty-two if the long-lead equipment slips tells a delivery manager exactly where to put their attention.
In the recovery period many teams have just lived through, that fragility is not theoretical. Supply disruption and a partly remote workforce mean the pessimistic case is closer to the centre of the distribution than it used to be. Estimators who still treat the optimistic number as the plan are, in effect, betting that the disruption of the last two years has fully passed. The honest move is to widen the spread and let the expected value reflect it.
If you want estimates that hold up to scrutiny and schedules that survive contact with reality, XNM's program & project delivery advisory can help your team build estimation discipline into how it plans.