i

Description

Gravure ancienne du Trésor de Whitehall à Londres : une longue colonnade, des calèches et des promeneurs.

Source

Thomas Hosmer Shepherd · original page

Licence

PD Public domain

Original photo, unretouched. The background is a version of it stylised by the site.

Unrealistic optimism

Everyone knows projects go off track; everyone plans their own as if it should escape it.

In brief

• What : we judge our chances of success as better, and our risks as lower, than those of others in the same situation.
• Why : we picture our plan, believe we're more competent than average, and remember good news better than bad news.
• Detect : “it should be fine,” a single scenario, no buffer, a risk that's never quantified.
• Avoid : reference class, explicit buffer, pre-mortem, commitment to a range.

On this page

1. Definition
2. Mechanism
3. Where you encounter it
   1. In project management
   2. In management
4. Famous cases
   1. The UK Treasury institutionalizes an “optimism adjustment” 2002
   2. Weinstein (1980): “accidents happen to other people” 1980
   3. Crédit Lyonnais: losses repeatedly underestimated (France) 1988-2000
   4. The Flamanville EPR: €3.3 billion announced, €23.7 billion estimated (France) 2006-2025
   5. The price shield: protection designed for one winter (France) 2021-2024
5. Detection signals
6. Avoidance reflexes
7. With AI: detect, fix, mitigate
   1. Detect
   2. Fix
   3. Mitigate
   4. Prompts to integrate into your practices
   5. How to integrate them
   6. Precautions
8. Limits
9. Sources and verification
10. See also
    1. Combines with, or reinforces
    2. Related biases
    3. Counterweights and related reflexes

Definition#

Fictitious illustration

Unrealistic optimism (optimism bias, or unrealistic optimism) is the tendency to believe that we are more likely than others to experience favorable events, and less likely to experience unfavorable events. Neil Weinstein described it in 1980 regarding everyday life events; the idea was later applied to project forecasting.

It should not be confused with an optimistic temperament, which is often useful for action. The bias starts when optimism gets into the calculation: a duration, a cost, a probability of success are estimated more favorably than the data justify. It is close to The Planning Fallacy, which is its most measurable manifestation, but it is broader: it also concerns risks, expected benefits, and user adoption.

Mechanism#

Several drivers combine:

• Comparison to oneself : we think about the project we're going to run, with our skills and our team, not the average of similar projects. Statistics apply to “other people.”
• Illusion of control : we overestimate what we can influence, so we underestimate what comes from outside (see The illusion of control).
• Asymmetric updating : according to work by Tali Sharot and colleagues, we revise our beliefs more in response to good news than to bad news. So the announced risk is only partially integrated.
• Interest in believing it : a project first has to be approved; the flattering plan is the one that gets through.

Where you encounter it#

In project management#

• Planning and sprint commitment : the sprint is filled to 100% of theoretical capacity, as if no one will be absent, interrupted, or blocked.
• Single scenario : the date relies on a chain where every task finishes on time, every third-party delivery arrives, every test passes on the first try.
• Risk register : probabilities are set to “low” by default; risks that depend on the team itself are judged almost zero.
• Expected benefits : user adoption, productivity gains, and payback period from the initial business case are rarely revised downward.
• Client communication : a “likely” date is presented as certain, with no range, because any caveat would look like a lack of confidence.

In management#

• Recruitment and skill ramp-up : we assume a new hire will be autonomous in two weeks, that training “will happen in parallel.”
• Goals : goals are set assuming everything will go like in the best quarter, without accounting for departures, vacations, or emergencies.
• Change management : “teams will naturally buy in”; resistance is treated as a special case rather than the rule.
• Self-assessment : everyone thinks their own projects will be less affected than those of other departments.

Famous cases#

Detection signals#

• “It should be fine.” “Normally, it’ll be fine.” “At worst, one extra day.”
• The plan includes only one scenario: no pessimistic alternative.
• The risk register contains few risks, all “low,” and none related to the team itself.
• Every past overrun is presented as a special case, never as a trend.
• Sprint capacity is calculated at 100%, with no vacation, no support, no bug fixes.
• Key assumptions (third-party delivery, expert availability, access to an environment) are neither listed nor dated.

Avoidance reflexes#

1. Take the outside view. Look at how many comparable projects met their deadlines and budgets; start from that rate, then explain why yours would do better.
2. Apply an optimism uplift. On your past projects, calculate the actual/estimated ratio and apply it to the raw estimate, as the UK Treasury does.
3. Run a premortem. “It’s six months from now, the project has failed: what happened?” This framing allows people to say what they were thinking but not saying.
4. Commit to a range. Announce a “one-in-two chance” value and a “four-in-five chance” value, not just a single number.
5. Assign a devil’s advocate for every launch decision or milestone gate, tasked with defending the unfavorable scenario.
6. Date and track assumptions. An assumption with no owner and no review date is a wish.
7. Reward bad news raised early. If reporting a risk costs more than hiding it, the team will learn to be optimistic only.

With AI: detect, fix, mitigate#

Prompts to integrate into your practices#

1. Guided premortem. To run before validating a schedule or committing in front of a client.

It is [six months] from now and project [name, scope in three lines] has failed: it is more than 50% late and the client is unhappy.
Write the most plausible story of this failure, as ten causes ranked from most likely to least likely. For each: what happened, the signal that should have alerted us, and one preventive action we can decide today.
Do not write any generic cause: base your analysis on the context below.

Context: [team, technologies, dependencies, timeline, constraints]

2. Hunt for optimistic assumptions. On a schedule, estimate, or decision file.

Here is a schedule (or estimate): [text].
1. List all implicit assumptions it relies on (availability, third-party deliveries, no defects, fast decision-making).
2. For each one, assess how fragile it is (low / medium / high) and explain what happens to the timeline if it is wrong.
3. Identify optimistic wording (“should,” “normally,” “just”).
4. Propose a three-point range (optimistic, likely, pessimistic) and justify each gap.
Do not validate the schedule: look for what could make it fail.

How to integrate them#

• During estimating: prompt 2 is used for systematic review, before sending a date to the client.
• At decision milestones: prompt 1 feeds committee discussion, to be read before the meeting.
• In retrospectives: comparing what AI had flagged with what actually happened improves both future use of AI and team judgment.

Precautions#

• Give AI the raw plan, without commentary: an enthusiastic presentation makes it enthusiastic too.
• The causes it proposes are hypotheses to test, not facts; references or figures it cites must be verified.
• It does not know your past gaps: provide your history, otherwise it will only produce generalities.
• Do not send confidential project data to an unauthorized tool.

Limits#

• Optimism is not always a flaw: it supports motivation and commitment, and without it many useful projects would never be launched. The issue is not believing in success, it is budgeting as if success were already achieved.
• The size of the bias varies by person, culture, and situation: it is stronger when people feel responsible for the outcome and when risk is seen as avoidable.
• UK Treasury uplifts are historical averages from large public projects; they cannot be directly transferred to a small software project.
• The concept covers distinct phenomena (comparison with others, underestimation of one’s own risks, asymmetric updating) that research increasingly treats separately.

Sources and verification#

• Weinstein (1980), “Unrealistic optimism about future life events,” Journal of Personality and Social Psychology
• Sharot (2011), “The optimism bias,” Current Biology — synthesis, including the share of people affected.
• Lovallo & Kahneman (2003), “Delusions of Success: How Optimism Undermines Executives' Decisions,” Harvard Business Review
• Buehler, Griffin & Ross (1994), “Exploring the planning fallacy,” Journal of Personality and Social Psychology
• Flyvbjerg (2014), “What You Should Know About Megaprojects and Why: An Overview,” Project Management Journal
• Mott MacDonald (2002), Review of Large Public Procurement in the UK, study for HM Treasury; HM Treasury, The Green Book and its supplementary guidance on optimism (Supplementary Green Book Guidance: Optimism Bias) — uplifts by project type.
• Jones & Steinhardt (2022), “Capturing Failures of Large Language Models via Human Cognitive Biases” — cognitive biases in language models.
• French National Assembly (1994), “Report of the inquiry commission on Crédit Lyonnais” (No. 1480, rapporteur François d'Aubert, chair Philippe Séguin) — details expansion, 1993 losses, real estate asset transfers, and inadequate controls.
• Court of Auditors (2000), “State intervention in the financial sector crisis,” report to the President of the Republic — establishes the optimistic nature of workout assumptions and the operation’s cost.
• Jean-Martin Folz (October 2019), “The construction of the Flamanville EPR,” report to EDF’s Chairman and CEO — states the initial estimate, describes it as unrealistic “at best,” and estimates at about €2.3B the share due to volume and scope.
• Court of Auditors (July 9, 2020), “The EPR sector,” thematic public report — establishes a construction cost multiplied by 3.3 (€12.4B2015) based on “unrealistic initial estimates.”
• Court of Auditors (January 14, 2025), “The EPR sector: new momentum, persistent risks,” summary — estimates Flamanville 3 total cost at about €23.7B2023, including financing costs during construction.
• Court of Auditors (2024), “Exceptional measures to fight rising energy prices,” thematic public report — establishes the initially temporary nature of the shields, their successive extensions, the net cost of €36B over 2021-2024, and weak targeting.

See also#