6 min. read

The cure for optimism isn’t pessimism. It’s data from projects that already finished.

The first three parts of this series diagnosed the disease. Whenever several estimates compete and the most extreme one wins, the winner is the one who was most wrong. Cursed bids, cursed business cases, cursed deadlines.

This part is the treatment. Five tools, in the order you use them.

Read the previous articles of the Winner’s Curse series:

What is reference class forecasting?

It’s Kahneman’s outside view made operational. Instead of estimating your project from its own plan, you find a class of similar completed projects, look at how their actual outcomes compared to their original estimates, and position your project inside that distribution.

Bent Flyvbjerg developed the method for large infrastructure projects in the early 2000s, and it has since been adopted by public bodies in the UK and Denmark for exactly the reason this series exists: the estimates that won approval kept being the ones that were most wrong.

The method has four steps:

Define the reference class. Projects similar in type, scale, and delivery model. Broad enough to have data, narrow enough to be comparable. Ten to twenty projects beats none.

Get the distribution. For each project in the class, actual cost or duration divided by the original estimate. Plot it. Notice how little of it sits below 1.0.

Locate your project. Decide how confident you need to be. If you want an 80% chance of finishing within budget, find the uplift at which 80% of the class came in under.

Apply the uplift and document why. The uplift isn’t padding. It’s the observed behavior of projects like yours.

Why it defeats the curse: the reference class doesn’t know your project is special. Everyone in the class thought theirs was special too, and the distribution is what happened to them anyway.

Your reference class doesn’t know your project is special. Neither did anyone else’s.

Three-point estimating done right

You know the formulas. Triangular: (O + M + P) ÷ 3. Beta, or PERT: (O + 4M + P) ÷ 6. The math is exam-basic. The curse hides in the inputs.

Two failure modes account for most cursed three-point estimates:

Your entry could not be saved. Please try again.

We sent links to your email! You should have received an email from us already. If you did not receive, make sure you check your spam folders and add masterofproject.com to safe senders list to receive our emails.

M drifts toward O. “Most likely” quietly becomes “what we’re hoping for.” Fix: anchor M to the median of your reference class, not to how the plan feels.

P is “what if it rains.” The pessimistic case describes a mild inconvenience, not a real failure. Fix: define P as the outcome if the two biggest risks in your register both occur. If that number doesn’t make you uncomfortable, it isn’t P.

In Part 3 the PM offered a range instead of a point. This is how you make that range defensible.

Hybrid (Agile + Waterfall) Project Estimation vs. Pure Agile Project Estimation: Which Approach Best Fits Your Project Needs? (Part 2)

Contingency reserve vs. management reserve

Contingency reserve covers known unknowns: the risks you identified and priced. It sits inside the cost baseline and the project manager controls it.

Management reserve covers unknown unknowns: the things nobody identified. It sits outside the baseline, the sponsor controls it, and using it changes the baseline.

The curse kills contingency first, because on a slide it looks like padding. Defend it by tying every dollar of contingency to a named risk in the register. Padding is unnamed. Contingency has a receipt.

Risk to Reward Ratio: How to Develop and Execute Projects for 5x Returns

Bid shading, translated

In auction theory, the fix for the winner’s curse is to shade your bid below your own estimate, because you know that if you win, you probably overestimated. Disciplined bidders lose more auctions and make more money.

The project equivalent is one question, asked before any estimate leaves your hands: “If this estimate is the one that gets selected, what does that tell me about it?” If the honest answer is “that it’s the most optimistic in the room,” adjust it before it wins. Part 5 takes this idea to its conclusion: sometimes the right adjustment is not to bid at all.

Are you a Strategic Project Manager? 7 Strategic Moves Every Project Manager Should Master to Drive Results

The premortem

Gary Klein’s technique, and the fastest tool on this list. Before you commit, gather the team and tell them the project has failed. It’s a year from now and it went badly. Everyone writes down, silently and independently, why.

It takes ten minutes. It surfaces the risks people knew about and hadn’t said, because saying them in advance sounded disloyal. And it gives your pessimistic estimate real content, which is exactly what “what if it rains” was missing.

Embracing Failure: How Project Managers Can Transform Setbacks into Stepping Stones

Before any estimate leaves your hands

Reference class identified and its distribution checked

Three-point estimate built, with P tied to named risks

Every dollar of contingency traced to the risk register

Management reserve requested separately and explicitly

Assumptions written down, not implied

Premortem run and its output folded into the risks

“If this wins, what does it mean?” asked and answered

Frequently asked questions

What’s the difference between reference class forecasting and analogous estimating?

Analogous estimating uses one or two similar past projects as a starting point. Reference class forecasting uses a whole class of them and looks at the distribution of estimate-to-actual ratios, which is what exposes the systematic optimism a single analogy can’t.

How much contingency should a project carry?

There’s no fixed percentage. Contingency should equal the priced exposure of the identified risks in your register. Anything not tied to a named risk is padding, and padding gets cut. For the unknowns, request management reserve separately.

Triangular or PERT: which three-point formula should I use?

PERT weights the most-likely value four times, so it suits an M that’s well grounded, ideally in a reference class. Triangular treats all three points equally, which is safer when you have little confidence in M. Either way, the formula matters less than whether P describes a real failure.

WHERE THE NEW PMP® EXAM TESTS THIS

Estimating techniques and reserve types are Process domain staples (41% of the exam). What changed on July 9, 2026 is the format: case-study sets and graphic-interpretation questions now hand you a table of past project outcomes or a distribution and ask you to choose an action.

Reading a forecast is a tested skill now, not a nice-to-have.

Reference class forecasting runs on data. Renewal is a good time to learn to work it.

Pulling historical project outcomes, building a distribution, modeling the uplift: that’s Excel and Power BI work. If you hold the PMP® and your three-year renewal is coming up, the Data & Analytics 60 PDU Renewal Pack covers Excel for Business Analysts, Data Analytics in Excel, Power BI, and Financial Forecasting & Modeling. Seven courses, 62 PDUs, your whole renewal cycle spent on the skills this post depends on.

See the 60 PDU Renewal Pack →

NEXT IN THE SERIES

Part 5 — When to Walk Away From a Project: The Bid/No-Bid Decision

Reference Class Forecasting: How to Estimate Without Cursing Yourself