Estimates built from how comparable past projects actually turned out hold up in the finance review, and rival plans built step by step from their own assumptions look exposed beside them.
An example
Two teams at a mid-sized airline were competing to replace the crew-scheduling system. The vendor-backed team presented a fourteen-month plan built task by task, each estimate signed off by its owner. The director of operations technology took another route. She listed every major operational system the airline had replaced in the past fifteen years, nine in all, and showed how each had done against its original plan: a median overrun of 70 percent on time and 45 percent on cost. Her own proposal, a phased replacement, carried a 23-month schedule and a contingency sized to that history. At the capital review, the chief financial officer asked the vendor team which of the nine projects theirs most resembled, and why it would beat all of them. The team had no answer. The phased plan was approved at the larger budget.
Why it works
Most plans are built from the inside: the team lists its own steps and estimates each one, and the total leaves out the delays nobody can foresee individually but almost every project meets. Anchoring the estimate on how a class of similar projects actually turned out gives the proposal numbers that survive scrutiny, so it looks better. The rival plan looks worse and costlier, since its optimism now shows against a record. Two conditions carry it: a fair set of comparable cases, and a reviewer who cares more about accuracy than about the lowest number. A sponsor who picks only the worst comparisons undoes it.
The thinking behind it
Daniel Kahneman Kahneman and Lovallo's Timid Choices and Bold Forecasts: A Cognitive Perspective on Risk Taking (1993) argues that decision-makers treat each problem as unique and ignore statistics from past cases. This inside view, which anchors forecasts on the specific plan and its scenarios, produces bold, optimistic forecasts. A plan built task by task, each estimate signed off by its owner, is the inside view in its purest form, which is why it looks exposed beside the record of similar projects.
Dan Lovallo The same paper contrasts the inside view with an outside view that judges a plan by the results of a class of similar past cases. A reference class forecast is that outside view put to work in a capital review: the sponsor picks the class, shows how its members actually turned out, and places the plan within that record, which gives reviewers a test they can apply to every plan in the room.
When it fits
Use it for large, one-off investments where estimates are routinely missed: system replacements, plant builds, acquisitions and integrations, new product launches. You need records of past projects, your own company's or an industry's, and a reviewer such as a finance committee that has been burned before. The decider must be willing to accept a larger, credible number over a smaller one.
How to use it well
Choose the reference class before you look at results, so the choice of cases is not shaped by what you want to show. Gather the actual outcomes against original plans for every case that fits, including the ones that went well.
- Show the distribution, with the median and the spread, alongside your plan.
- Explain where your project sits in that distribution, and what would move it.
- Size the contingency to the history, and say so.
- Invite finance to check the case list.
Present the method once, plainly, and let the reviewers apply it to every plan in the room. Leave the rival team alone: their estimates are honest inside-view work, and the comparison speaks for itself.
What it costs and where it backfires
Your number will be larger, and a larger number can lose to a cheaper plan in a room that wants to believe. If your project then runs well under the forecast, you look like you padded it. Building the reference class takes real work, and critics will argue the comparisons do not fit. The clean line is the case list: once you leave out the comparable projects that went well, the forecast has been rigged and is no longer an outside view.
When it is used on you
Look hard at the list of cases when your plan is put beside a reference class: how they were chosen, whether they truly resemble yours, and whether good outcomes are missing. If the comparison is fair, adopt it and re-plan, since a credible larger number beats a small one that breaks. If it is skewed, offer a fairer class.