PDF Summary:How Big Things Get Done, by Bent Flyvbjerg and Dan Gardner

Large projects frequently fail to meet their budgets, timelines, and promised benefits—in fact, only 0.5% of big projects succeed across all three measures. In How Big Things Get Done, Bent Flyvbjerg and Dan Gardner examine why major projects so often go wrong and offer practical strategies for improving outcomes.

The authors explain how psychological biases like optimism and poor time estimates, combined with strategic misrepresentation, lead to unrealistic project plans. They also describe how small delays compound into major failures. To counter these patterns, Flyvbjerg and Gardner present methods for better planning and execution, including reference-class forecasting to improve predictions, iterative testing to reduce risk, and modular design to enable learning through repetition. Their approach emphasizes thorough preparation, experienced teams, and flexible strategies that adapt as projects progress.

(continued)...

The authors explain that fat-tailed distributions aren't well-represented by the mean, making it a poor forecast estimator. For the fattest tails, there isn't a steady average to predict results clustering around, since an even more extreme result can and will emerge and drive the mean further away. Therefore, relying on the average and expecting your result to be similar to it is a risky mistake.

Defining Fat-Tailed Distributions

The authors’ claim that 80% of results are in the body and 20% in the tail of a fat-tailed distribution is not a standard definition in statistics. Fat-tailed distributions are characterized by their slow decay, meaning that extreme values are more probable than in thin-tailed distributions. Newman, a physicist and mathematician, explains that in a power-law distribution, the probability of observing a value x decreases as a power of x, rather than exponentially. This means that the probability of observing a value 10 times larger than the mean is only 10^α times less likely, where α is a constant.

Compounding Risks and Project Drift

Flyvbjerg and Gardner assert that early delays can lead to a series of additional setbacks. The majority of project leaders see initial delays as unimportant, believing they have time to make up for lost ground. However, these delays can have a domino effect on the delivery timeline.

(Shortform note: In Normal Accidents, Charles Perrow argues that in loosely coupled systems, components and subsystems are buffered from one another by slack, redundancies, and temporal and spatial separation, so that most disturbances remain localized and can be absorbed or corrected before they propagate. Loose coupling, by inserting these buffers and delays, prevents many small failures from combining into system-wide breakdowns.)

Furthermore, the authors point out that the cumulative impact of typical dangers is often what undoes projects. Projects rarely fail for just one reason; instead, they fail because of the cumulative impact of risks typically listed in the risk register.

(Shortform note: This claim may be less applicable in environments where “black swan” events are common. In The Black Swan, Nassim Nicholas Taleb argues that the most significant events in history are often unpredictable and unprecedented.)

Strategies for Success in Big Projects

The authors assert that successful large-scale endeavors require meticulous preparation and a cohesive group. While it's vital to plan in order for a project to run smoothly, the most effective plan will still fail without a strong team to execute it. The most successful teams have experience and have worked together before. They deeply understand the undertaking and trust each other.

(Shortform note: To create a cohesive group, try to work with the same people on multiple projects. After each project, hold a debriefing session to discuss what went well and what could be improved. This will help your team learn from each experience and build trust. Even short, 10-minute debriefs can make a big difference in how well your team works together.)

Next, Flyvbjerg and Gardner discuss core strategies for planning and forecasting, as well as execution and delivery.

Core Strategies: Planning & Forecasting

Flyvbjerg and Gardner emphasize the importance of starting with a clear understanding of your goal and working backwards. The objective is fundamental to a project’s success. Without a clear goal, you can become caught in the details and lose track of your objectives. To begin with a clear goal, ask yourself what you want to achieve and why. Then, work backwards to figure out what needs to happen to reach that goal.

(Shortform note: While starting with a clear goal and working backwards is a common approach to project management, it may not always be the best strategy. In situations where the problem is novel or highly uncertain, this approach can lead to premature assumptions and limit the potential for discovery. In Effectuation, Saras D. Sarasvathy argues that starting with a fixed goal can constrain creativity and adaptability, especially in uncertain environments.)

They also stress that planning involves active, iterative, and learning processes. It involves experimenting, evaluating effectiveness, and then testing something else based on your insights. This cycle, along with thorough testing, generates a strategy that boosts the likelihood of delivering quickly and without problems.

(Shortform note: The idea of planning as an active, iterative, and learning process has a long history in quality management. W. Edwards Deming, a pioneer in quality control, developed the Plan–Do–Check–Act (PDCA) cycle in the 1950s. This cycle emphasizes planning as a process of experimentation and learning.)

Next, Flyvbjerg and Gardner discuss strategies for accurate forecasting and de-risking plans and technologies.

Forecasting & Reference Classes

The authors explain that reference-class forecasting (RCF) improves the accuracy of project forecasts. This method uses data from similar past projects to estimate the outcomes of a current project. By using real data, RCF helps counteract biases and account for unexpected problems. It’s simple to use and has demonstrated that it produces more precise projections than traditional methods.

To use RCF, identify a set of similar projects that have been completed previously to the one you’re planning. Gather data on factors like their costs and timelines. Compute the mean results from these projects to form a baseline estimate for your own project. Adjust this estimate if specific reasons exist to believe the endeavor will differ significantly from the typical outcome.

The Potential for Bias in Reference-Class Forecasting

While RCF can improve forecasting accuracy, it may also reinforce structural biases present in the data from past projects. In Weapons of Math Destruction, Cathy O’Neil explains that algorithms trained on historical data can perpetuate existing inequalities. For example, if a city’s past infrastructure projects have consistently underfunded certain neighborhoods, RCF might predict lower budgets for similar projects in those areas, perpetuating the cycle of underinvestment. O’Neil argues that while data-driven methods like RCF can be powerful, they require human oversight to ensure they don’t entrench existing disparities.

Flyvbjerg and Gardner assert that RCF addresses unpredictable events in project planning. The data from the reference class accounts for every unexpected event that happened in those projects. Even without knowing the specifics of those events or their significance, the numbers reflect their frequency and impact.

(Shortform note: RCF may not always account for unpredictable events. In An Engine, Not a Camera, Donald MacKenzie argues that financial models can change the behavior of the markets they’re meant to predict. When people use a model to make decisions, they change the incentives of the people they interact with, which can make the model’s predictions less accurate.)

De-risking Plans & Technologies

The authors suggest using a cyclical method to identify and address potential problems early. These processes allow you to experiment and test every part of your plan, which saves both time and funds. They also force you to explain your plan, helping you identify any gaps in your understanding.

(Shortform note: The authors’ cyclical method of probing and refining a plan has a long history. In 1984, David Kolb published his experiential learning theory, which describes how people learn and adapt through a cyclical process. Kolb’s theory suggests that effective action comes from moving through four stages: concrete experience, reflective observation, abstract conceptualization, and active experimentation. This approach emphasizes continuous learning and adaptation rather than relying on a single, static plan.)

Additionally, the authors recommend thoroughly testing ideas and technologies before implementation. This is critical for initiatives that need to accomplish what's unprecedented. These projects lack experience from the outset, and to realize their vision within schedule and financial constraints, that gap must be closed with consistent application of experience.

To test ideas, optimize experimentation with a very repetitive approach. Evaluate every aspect, from overarching concepts to minor specifics. Use effective evaluation methods to mitigate the dangers of failure, make strategic risk assessments, and explore novel concepts. If something's unproven, you must test it more. Retain whatever functions successfully. Discard what's ineffective. Attempt, learn, re-attempt, and allow the strategy to change.

The Pitfalls of Excessive Testing

While the authors advocate for rigorous validation, this approach can inadvertently stifle innovation. In The Innovator’s Dilemma, Clayton Christensen argues that established companies often fail to innovate because they focus on refining existing products through incremental improvements. He explains that these companies prioritize projects that promise reliable returns and meet established performance metrics, which can lead them to overlook disruptive innovations that initially perform poorly in tests but have the potential to transform industries. This suggests that a relentless focus on testing and refinement may cause organizations to abandon unconventional ideas that could succeed in real-world applications, even if they don’t excel in controlled experiments.

Core Strategies: Execution & Delivery

Flyvbjerg and Gardner recommend using modularity to improve efficiency and reduce costs. Modularity is the practice of creating large structures from small parts. It's a question of degree, and repetition is central to modularity.

The authors explain that modularity is the best method to reduce risk because it enables experimentation and learning. When a method proves successful, you retain it in your strategy. If not, you experience rapid failure and modify the approach. Repetition propels you quickly up the learning curve, improving each subsequent iteration and making them simpler, less expensive, and faster.

The Risks of Modularity

While modularity can reduce risk, it can also create risk. In Design Rules, Volume 1, Carliss Y. Baldwin and Kim B. Clark explain that modularity can hide interdependencies between modules. When these interdependencies are not fully understood, they can lead to unexpected failures and costly integration problems. For example, in the development of the Boeing 787 Dreamliner, extensive modularity led to significant delays and cost overruns when the integration of different modules revealed unforeseen technical challenges.