The single biggest reason QSRA and QCRA outputs are dismissed by senior management isn't the simulation — it's the three-point estimates feeding them.

Min, most likely, max. Three numbers per activity or risk. Asked in a workshop. Submitted in 20 seconds. And almost always wrong — in a predictable, systematic way.

What three-point estimates are (and why we use them)

A three-point estimate captures uncertainty in a duration or cost using three numbers:

• Min (optimistic): the best plausible outcome — typically P10.
• Most likely: the mode of the distribution.
• Max (pessimistic): the worst plausible outcome — typically P90.

These feed PERT, triangular, or beta distributions in Monte Carlo tools (Safran Risk, Argo, @Risk, Crystal Ball). They are the input that makes everything else possible — and the input that breaks everything when wrong.

The four ways three-point estimates go wrong

1. Anchoring on the baseline

When the baseline duration is 10 weeks, workshop participants instinctively offer 9 / 10 / 12 — a 30% range when reality might be 7 / 11 / 18 (a 100%+ range). The baseline becomes the anchor; the range collapses around it.

2. Optimism bias

Engineers want to look competent. Estimating “16 weeks” for something the planner has baselined at 10 feels like admitting failure. So the max gets shaved down.

3. Confusing “most likely” with “mean”

For right-skewed activities (most construction work), the mode and the mean are different. Participants pick a number near the mean and call it “most likely” — flattening the model's ability to capture tail risk.

4. Range = aspiration, not data

Without historical reference data, the “max” is whatever the most pessimistic person in the room dares to say. In an aspirational team culture, that's not very pessimistic.

What RDE™ does differently

The Risk Data Engine™ (RDE™) is IQRM's empirical alternative to the workshop guess. Instead of asking people in a room, RDE™ asks the data:

• Ingest historical data. Completed projects, packages, or activities — durations and costs vs. baselines.
• Cluster by activity type and project profile. Concrete works on GCC summer projects behave differently from UK winter ones.
• Fit empirical distributions. Lognormal, beta, or non-parametric — whatever the data supports.
• Output fitted (min / most likely / max + distribution shape) for each activity class.
• Feed directly into QSRA / QCRA / JCL models — replacing the workshop sticky-notes.

The result: Monte Carlo inputs that reflect how your projects actually behave, calibrated to your data, defensible to the auditor.

When workshop estimates are still OK

RDE™ doesn't replace workshops — it strengthens them. Workshops remain essential for:

• First-of-a-kind activities with no historical analogue.
• Discrete risk events (probability × impact) — these are inherently judgmental.
• Validation — sense-checking RDE™ outputs against the team's experience and challenging anomalies.

Frequently asked questions

What distribution should I use for three-point estimates?

PERT is the safe default — it down-weights the extremes vs triangular. Use lognormal when you have evidence of strong right-skew.

How wide should my three-point range be?

If you've never overrun on this activity type by more than 5%, narrow ranges (±10%) are reasonable. If you've overrun by 80% before — your “max” needs to be at least 80%.

Why do all my workshops produce similar ranges across activities?

Because participants mentally apply a fixed % spread to every line. This is a classic anchoring failure. Use RDE™ outputs as the starting reference.

Related: How to Write a Risk Statement · What is QCRA?