Why Smart Teams Make Bad Decisions on Large Capital Projects

Across every region and sector, large capital projects are entrusted to some of the most capable people an organisation can assemble. Experienced engineers, seasoned estimators, financial analysts, lawyers and executives spend months or years preparing the case for an investment that may shape a company, a city or a national grid for decades. These teams are rarely short of intelligence, data or diligence. Yet the record of large capital projects is marked by estimates that proved far too low, schedules that proved far too short, benefits that never fully arrived and warning signs that were visible to someone long before they were visible to the people who decided. The uncomfortable conclusion is that bad decisions on large capital projects are seldom made by careless people. They are made by capable people working inside a decision system that quietly bends their judgment in the same direction, at the same time, and then rewards them for not noticing.

This comprehensive blog will explore why smart teams make bad decisions on large capital projects, and what leaders can do to design better decisions rather than simply hoping for better decision makers. It is written for sponsors, project directors, project managers, owner representatives, estimators, project controls professionals and board members who prepare, review or approve major investment decisions. We will examine error as a system operating at three layers: the individual mind, the group in the room and the organisation that funds and rewards the work. We will look at how the mind turns thin evidence into a confident story, the core biases that follow from that habit, and why expertise does not automatically protect against them. We will then examine how groups amplify individual error, how organisational incentives can reward it, and how pressure, framing and hindsight make the problem harder to see. Finally, we will set out practical decision hygiene and a checklist that leaders can apply at each decision gate.

1. Capable People, Poor Outcomes

The scale of the problem is not in doubt. In one large international database of projects, only 8.5 percent hit both their cost and schedule targets, only 0.5 percent hit cost, schedule and benefits together, and 91.5 percent were over budget, over schedule or both. Most project types in that database show fat-tailed outcomes, meaning that a small share of projects overrun by extreme amounts rather than clustering politely around a modest average. Another long-running body of research on industrial projects reports that success rates for the largest projects improved from under 20 percent in the first years of this century to over 40 percent in the most recent period, a meaningful gain but still well short of what established practices can achieve.

Failure on large capital projects is not a rare accident caused by an unusually weak team. It is the normal pattern, and it occurs in organisations that employ excellent people, use sophisticated tools and follow formal gate processes.

The research on behavioural bias in projects offers a clear explanation for this paradox. The drivers of poor outcomes are described as universal, rooted in psychology and in power, and present at every scale of project. Psychological drivers include optimism, the planning fallacy and the belief that a project is unique. Power-related drivers include strategic misrepresentation and political lock-in. Neither set of drivers depends on a lack of intelligence.

2. The System View: Three Layers of Error

A useful way to understand decision error on large capital projects is to treat it as a system with three interacting layers.

The first layer is the individual mind. Every estimator, engineer and executive relies on fast, intuitive judgment to make sense of incomplete information. That intuition is efficient and usually helpful, but it carries systematic tendencies: anchoring on the first number seen, overweighting what is easy to recall, trusting a coherent story more than its evidence deserves, and expressing more confidence than the evidence supports. A widely used project management practice guide lists optimism bias and the planning fallacy, anchoring, the framing effect, loss aversion and the sunk cost effect, resistance and misrepresentation as individual-behaviour causes of complexity in projects.

The second layer is the group. Major decisions are made in meetings, review boards and steering committees, where individual tendencies interact. The same guide names groupthink as a group-level cause of complexity, alongside tribal mindsets and shifts in group risk appetite.

The third layer is the organisation. Large capital projects sit inside long chains of principals and agents: taxpayers or shareholders, governments or boards, sponsoring agencies, owner organisations, contractors and subcontractors. Research on governing large projects describes this as a complex, multi-tier principal-agent situation shaped by political power, self-interest, asymmetric information and accountability, and different time horizons and perceptions of risk. In such a chain, the people who hold the best information are often not the people who carry the consequences, and the incentives at each link can reward an optimistic number more than an accurate one.

These layers reinforce one another. An individual's optimistic estimate becomes a group's shared assumption, and the organisation's funding process then selects and rewards the most optimistic proposals. This is why bias on large capital projects should be treated as a property of the decision system rather than a personal failing.

Figure 1: Three layers of decision error. Individual bias, group dynamics and organisational incentives reinforce one another.

3. How the Mind Fills the Gaps

At the root of the individual layer is a simple mental habit. The fast, intuitive mode of thinking works with whatever information is currently active in the mind. Information that is not retrieved, even unconsciously, might as well not exist. Psychologists summarise this with the phrase "what you see is all there is": the mind builds the most coherent story it can from the evidence at hand and does not naturally ask what evidence is missing.

The most important consequence for large capital projects concerns confidence. Subjective confidence depends on the quality of the story, not on the quantity or quality of the evidence behind it. People often fail to ask what important evidence is still missing, and the cleaner the story, the less likely they are to ask.

This habit links many later errors. Overconfidence, framing effects, predictions that ignore regression towards typical outcomes, the inside view of a project and the tendency to focus on whatever is currently in view can all be understood as variations of the same theme: judging from what is visible and treating it as complete.

A related tendency is the belief that small samples are representative. Intuition treats the law of large numbers as if it applied to small numbers too. A non-specialist does not feel much difference between a survey of 150 responses and one of 3,000. On a project, the same habit appears when a team generalises from two or three previous projects, a single site visit or one supplier's quotation. Keeping doubt alive requires deliberate effort; the fast mind prefers certainty and spreads the story as if it were true unless it is immediately rejected.

Figure 2: What you see is all there is. Confidence tracks the coherence of the story, not the evidence behind it.

4. The Core Biases on a Project

Several well-documented biases follow from this way of thinking. On large capital projects they rarely act alone.

Anchoring. In a classic experiment, people watched a rigged wheel stop on either 10 or 65 and were then asked to estimate the percentage of African countries in the United Nations. Those who saw 10 estimated 25 percent on average; those who saw 65 estimated 45 percent. The wheel carried no information, yet it moved the answer. Anchoring works both through insufficient adjustment from a starting point and through priming that makes compatible answers come to mind more easily. On projects, planners anchor on best-case plans, on the proposer's early optimistic figure, or on their own limited experience with a different type of asset. Research on governing large projects describes a plan that starts from a positive anchor, the proposal's expected performance, with later adjustments that remain slightly optimistic. The recommended remedy is not to have no anchor at all, but to anchor deliberately on a relevant base rate drawn from similar completed projects.

Availability. People judge how frequent or likely something is by how easily examples come to mind. Recency, novelty and emotional intensity all make examples easier to recall, so they distort judgments of likelihood. The research notes that people in power can be more influenced by the ease of retrieval than by the content retrieved. A related bias, sometimes called localism, treats the local as global because local experience is the most available.

Overconfidence. Overconfidence is the tendency to have excessive confidence in one's own answers. People overestimate how much they understand and underestimate the role of chance and ignorance, which leads them to underestimate how variable events can be. In one widely retold study, senior finance executives gave forecasts of market returns with 80 percent confidence intervals. In practice, 67 percent of actual outcomes fell outside those intervals, more than three times the 20 percent that the executives had allowed.

Uniqueness and confirmation. Uniqueness bias is the tendency to see one's project as more singular than it really is. Many project definitions emphasise that each project creates something unique, which primes teams to believe they have little to learn from others. Research cited in the literature found that managers who see their projects as unique perform significantly worse. A project can be unique in its place and time and still belong to a wider class of similar projects. Confirmation bias completes the pattern: in multi-tier projects, agents can provide principals mainly with information that confirms the positive aspects of a proposal, while funding bodies sometimes push for confirmatory information themselves.

The important point is how these biases combine. An anchor sets the starting number, availability supplies a few vivid supporting examples, uniqueness dismisses inconvenient comparisons, confirmation filters the evidence, and overconfidence narrows the range of outcomes considered. The cumulative effect is a forecast that is precise, persuasive and systematically wrong.

5. Optimism and the Planning Fallacy in Brief

Optimism deserves a brief mention because it is the most familiar bias in project discussions. Most people are highly optimistic most of the time, and executive overoptimism has been traced to both cognitive biases and organisational pressures. Studies comparing the outcomes of capital projects, acquisitions and market entries with the original expectations show a strong tendency towards overoptimism. One study of chemical plants found that actual construction costs were more than double initial estimates, and about half of the plants produced below 75 percent of design capacity after their first year.

The planning fallacy is the related tendency to build forecasts around the most likely scenario. Imagination cannot foresee every way a plan might fail, and treating the "most likely" path as the most likely outcome understates the combined probability of many small-chance unfavourable paths.

Two further points matter for this blog. First, optimism is not something to eliminate. It supports enthusiasm, resilience and motivation, and people in delivery roles often benefit from it once a commitment has been made. The problem arises when optimism decides whether to commit. Second, optimistic error is different from deliberate misrepresentation. Delusion arises from honest belief and the inside view; deception arises when estimates are shaped to win approval or funding. Both can operate on the same project at the same time. The distinction between honest optimism and strategic misrepresentation is important enough to warrant its own treatment, and it will be examined in a forthcoming ProTrain blog.

6. Why Expertise Does Not Fix It

A natural response to these findings is to assume that experts are less vulnerable. Unfortunately, the evidence suggests otherwise. Overconfidence has been found among laypeople and experts alike. The experts most in demand may be the most overconfident of all, because people tend to pay for confidence more than for expertise.

Expertise can also be embedded in tools that look objective. Risk models such as Monte Carlo simulations are often built on assumptions of thin-tailed risk, even when the historical distribution of comparable projects is fat-tailed. The error then lies in the inputs rather than in the simulator, but the polished output lends the result a precision it has not earned.

A third limit concerns which estimate is judged. Estimates naturally become more accurate as scope firms up during design. If an organisation evaluates its forecasting performance against later estimates or control budgets, it can conclude that its forecasting is good while overlooking the estimate that actually informed the decision to build. Research on cost overruns therefore insists on measuring performance against the estimate at the time of the decision. Government appraisal guidance makes a related point: adjusting forecasts for optimism is useful, but it does not replace sound project and risk management.

The underlying problem is the inside view. The remedy of taking an outside view, by anchoring on the distribution of outcomes from a reference class of completed projects, is the subject of another forthcoming ProTrain blog. For this blog, the essential lesson is that expertise improves the quality of the inside view without automatically correcting its blind spots.

7. When the Group Amplifies the Error

Decisions on large capital projects are rarely made by individuals alone.

Groupthink is defined as a strong concurrence-seeking tendency that interferes with effective group decision making. Its recognised symptoms include apparent unanimity, illusions of invulnerability and moral correctness, biased perceptions of outsiders, interpersonal pressure on dissenters and self-censorship.

Silence is one of the most damaging outcomes. Research on psychological safety shows that people constantly run a tacit calculation of interpersonal risk. Where speaking up feels unsafe, people avoid seeking help, discussing errors and raising concerns, which are exactly the behaviours that support good decisions. A well-known space programme investigation found that contractors withheld troubling test information while under relentless criticism and pressure, and the failure was later reframed as a problem of social context rather than simply technical leadership. Practitioners note that the reluctance of people to voice reservations during planning is one of the reasons projects fail, and that teams which fear conflict avoid controversial topics that are critical to success.

Status adds a further distortion. In high-reliability organisations, deference to expertise means that in a crisis the person with the most knowledge of the situation may not be the most senior person present, and hierarchy is de-emphasised so that threats can be learned about quickly. Research on team learning has found that psychological safety enables lower-status members to speak up across the hierarchy. On a large capital project, the site engineer, the junior estimator or the interface coordinator may hold the decisive piece of information, but the decision may be made by people who never hear it.

Figure 3: How a group turns private doubt into public certainty, and the counters that keep dissent alive.

8. When the Organisation Rewards the Error

The third layer is the organisation and the wider institutional chain that funds a project.

Competition for capital is a powerful selector. When many proposals compete for limited funds, the process tends to select the most optimistic proposals, because they show the best returns on paper. Stretch targets and compensation linked to approvals can push in the same direction. In multi-tier chains, the agent closest to the work often knows more about site risks, budget accuracy and technical difficulty than the principal funding it. In underground construction, for example, builders may understand obstacles better than the client, bid low to win, and then allow variations to grow when the obstacles appear. Research on governing large projects also cites audit findings in which procurement was manipulated to favour suppliers who provided the information that was requested.

Lock-in compounds the problem. The commitment fallacy describes a premature, irreversible commitment to a project concept after shallow analysis, the pattern of thinking fast and then acting slowly that leads to a cycle of breakdowns and fixes. Once an organisation has publicly committed, reversing course becomes politically and personally costly.

Two related mechanisms deserve brief mention because each will be treated in its own forthcoming ProTrain blog. The first is escalation of commitment, sometimes called the sunk cost fallacy: continuing to invest because of what has already been spent despite new evidence that the decision may be wrong. Raising commitment to a failing project can serve the executive who owns it, because cancellation stains that person's record, even when continuing conflicts with the organisation's interest. Loss aversion, the finding that losses loom larger than equivalent gains, sits alongside this pattern, with losses typically weighted around 1.5 to 2.5 times as heavily as gains in experiments. The second is the normalisation of deviance, in which repeated departures from standards become accepted because nothing has yet gone wrong, so that mistakes can become socially organised and systematically produced.

The organisational remedies proposed in the research focus on incentives and transparency. They include sharing financial responsibility so that local sponsors carry a real stake, applying optimism adjustments to forecasts, ensuring that private parties carry a meaningful share of risk capital rather than relying on full public guarantees, rewarding estimates that are well sourced, independent peer review of forecasts, publication of forecasts and benchmarks, and in extreme cases professional or legal penalties for deliberate deception. The principle is stated simply: reward honesty and punish deception.

9. Pressure, Framing and Hindsight

Three further forces make decision errors harder to detect.

Pressure. Large capital project decisions are often made under deadline, financial and political pressure. Research reviews find that time pressure tends to push people towards faster heuristics and can lead to poorer choice of strategy, while more open time windows support better use of information. These are single studies and should be read with caution. Experimental research has also linked higher stress hormone levels with lower decision quality, with the largest deficits appearing when acute stress coincides with time pressure.

Framing. The same facts can produce different choices depending on how they are described. In a well-known experiment, most people preferred a programme that would save 200 of 600 people with certainty, but when the identical outcome was described as 400 people dying, many shifted to a risky alternative. Preferences attach to the framed problem, not to an underlying frame-free reality. On projects, a cost increase described as "protecting the investment" can feel very different from the same increase described as "a loss on the original budget". A further distinction is between narrow framing, considering each risky decision in isolation, and broad framing, considering many decisions together. The importance of setting the right question before choosing among options is examined in the ProTrain blog Decide the Question First.

Hindsight. After an outcome is known, people struggle to reconstruct what they believed beforehand. Studies have found that people later remember assigning higher probabilities to events that occurred and lower probabilities to events that did not. Hindsight feeds illusions of certainty and therefore overconfidence. It also distorts learning: decision makers who took reckless risks and were lucky are treated as skilled, while careful decision makers who were unlucky are blamed. Judging decisions only by their outcomes, sometimes called resulting, prevents organisations from learning whether their process was sound.

10. Decision Hygiene: Designing Better Decisions

If decision error is systemic, the response must also be systemic. Decision hygiene refers to routine practices that reduce the opportunity for predictable error. Several practices are well supported in the sources reviewed for this blog.

Independent views before discussion. Before a meeting discusses a forecast, risk or option, each participant writes a short independent view. This preserves the diversity of judgment that a group is supposed to provide and prevents the first confident speaker from setting the anchor for everyone else.

The premortem. Before a major commitment, the team imagines that the project has failed and each member writes down the reasons for the failure. This technique uses prospective hindsight, imagining that an event has already occurred, which has been reported to increase the ability to identify reasons for future outcomes by around 30 percent. Its greater value is social: it makes dissent legitimate and gives knowledgeable sceptics a safe way to raise concerns. Research on governing large projects recommends listing the reasons a forecast could be wrong as part of good practice.

The outside view. Every major forecast should be compared with the distribution of outcomes from a reference class of similar completed projects. Where such data are missing, the gap should be acknowledged rather than replaced with a generic uplift presented as if it were reference-class evidence.

A decision journal. At each consequential decision, the decision maker records what is expected to happen, why, and with what probability, using precise rather than vague language. Recording expectations at the time of the decision protects against hindsight and allows the organisation to separate the quality of the process from the quality of the outcome.

A bias audit at the gate. A short set of questions can be asked before each major approval:

• Could this estimate reflect honest optimism, deliberate positioning, or both, and who benefits if it stays low until the next gate?
• Which earlier optimistic plan is this estimate still tied to?
• Is there a named reference class and a distribution, or only a bottom-up build-up?
• Which variance or waiver has been repeatedly accepted because nothing has yet failed?
• Who is personally committed to the current answer, and are future costs being judged against remaining benefits or against money already spent?
• Is the apparent agreement in the room real, or is it self-censorship?
• What fact, if it arrived, would stop or reshape the project, and which assumptions remain unowned?

Structural protections. Further safeguards include independent peer review of forecasts, collective forecasting methods with several independent rounds, and deliberate deference to expertise so that the people closest to the risk are heard regardless of rank.

Figure 4: Decision hygiene before the commitment: independent views, premortem, outside view, bias audit and decision journal.

11. Conclusion and Checklist for Leaders

Bad decisions on large capital projects are rarely the product of incompetence. They emerge from a system in which normal human judgment, group dynamics and organisational incentives align in the same optimistic direction.

Because the causes are systemic, the remedies must be built into the decision process rather than left to individual vigilance. Leaders of large capital projects can use the following checklist at each decision gate:

1. Has every key participant recorded an independent view before discussion began?
2. Is the forecast anchored on a relevant base rate, or on the first plan proposed?
3. Has a premortem been completed, and have its findings been addressed?
4. Are the most senior voices speaking last, and has the person closest to the risk been heard?
5. Do incentives reward accurate, well-sourced estimates rather than attractive ones?
6. Is performance being measured against the estimate that informed the decision?
7. Has the decision, its expected outcome and its probability been recorded in a decision journal?
8. Are future costs being compared with remaining benefits rather than with money already spent?

Smart teams will always be vulnerable to bias, because bias is part of how capable minds work. The organisations that achieve better outcomes on large capital projects are not those that claim to be free of bias. They are those that design their decisions on the assumption that bias will be present, and build the safeguards in before the commitment is made.

References

1. American Psychological Association. (2023). APA Dictionary of Psychology: Groupthink. https://dictionary.apa.org/groupthink
2. Caviola, S., Carey, E., Mammarella, I. C., & Szucs, D. (2017). Stress, time pressure, strategy selection (open review). Frontiers in Psychology. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2017.01488/full
3. Doroc, K., Yadav, N., & Murawski, C. (2025). Acute stress impairs decision-making. Communications Psychology. https://www.nature.com/articles/s44271-025-00355-x
4. Edmondson, A. C. (2002). Managing the risk of learning: Psychological safety in work teams.
5. Farnam Street. (2013). Creating a decision journal. https://fs.blog/decision-journal/
6. Flyvbjerg, B. (2021). Top-ten behavioral biases in project management: An overview.
7. Flyvbjerg, B., Garbuio, M., & Lovallo, D. (2009). Delusion and deception in large infrastructure projects. https://arxiv.org/pdf/1303.7403
8. Flyvbjerg, B., & Gardner, D. (2023). How Big Things Get Done.
9. Flyvbjerg, B., Holm, M. S., & Buhl, S. (2002). Underestimating costs in public works projects: Error or lie? https://arxiv.org/pdf/1303.6604
10. HM Treasury. (2013). Green Book supplementary guidance: Optimism bias. https://www.gov.uk/government/publications/green-book-supplementary-guidance-optimism-bias
11. Hond, P. (2025). Vaughan on normalization of deviance (Challenger), interview with Diane Vaughan. Columbia Magazine. https://magazine.columbia.edu/article/challenger-disaster-normalization-deviance
12. Kahneman, D. (2011). Thinking, Fast and Slow.
13. Klein, G. (2007). Performing a project premortem. Harvard Business Review, September 2007. https://hbr.org/2007/09/performing-a-project-premortem
14. Lencioni, P. (2002). The Five Dysfunctions of a Team.
15. Lovallo, D., Cristofaro, M., & Flyvbjerg, B. (2023). Governing large projects: A three-stage process to get it right.
16. Lovallo, D., & Kahneman, D. (2003). Delusions of Success.
17. Merrow, E. W. (2024). Industrial Megaprojects.
18. Pellerin, C. J., Jr. (2009). How NASA Builds Teams.
19. Project Management Institute. (2014). Navigating Complexity.
20. Staw, B. M. (1976). Knee-deep in the big muddy (escalating commitment). Organizational Behavior and Human Performance, 16, 27–44.
21. UC Davis PSNet Editorial Team / Agency for Healthcare Research and Quality. (2019). High reliability (primer). https://psnet.ahrq.gov/primer/high-reliability