The Planning Fallacy

In the AI Story

The fallacy was first named by Daniel Kahneman and Amos Tversky in 1979 as a cognitive bias observed in individuals estimating task durations. Flyvbjerg's contribution was to operationalize the concept at the scale of institutions and to demonstrate, through the world's largest database of megaproject outcomes, that the bias produces a consistent and economically catastrophic pattern across human enterprise. The heuristics and biases tradition that Tversky and Kahneman founded supplied the cognitive vocabulary; Flyvbjerg supplied the institutional evidence.

The explanatory architecture identifies two forces, both necessary. The first is cognitive: the planner sincerely believes that this project is different, that the base rate does not apply, that the team, the technology, or the circumstances exempt this case from the pattern. The second is political: the planner knows the optimistic forecast is wrong but produces it anyway because the institutional incentive structure rewards optimism and punishes realism. The honest project does not get funded. The optimistic one does. The planner, operating rationally within the incentive structure, produces the numbers the structure demands.

The fallacy maps onto the AI transition with unsettling precision. The cognitive version — genuine overestimation of what current AI systems can accomplish — saturates the discourse of engineers and researchers too close to the technology to maintain perspective on its limitations. The political version — deliberate inflation of AI claims to secure investment, market share, and regulatory latitude — fills the earnings calls and keynote addresses of companies whose trillion-dollar valuations depend on the narrative that artificial general intelligence is imminent or that current systems constitute a reliable path toward it.

AI introduces a new and dangerous variable: speed. In traditional megaproject management, the timeline between plan and completion creates an involuntary feedback loop — costs that were underestimated become visible when invoices arrive, benefits that were overestimated become apparent when usage falls short. AI compresses this feedback loop for technical implementation to near-zero, creating the planning fallacy at machine speed: the same cognitive and political distortions operating at a velocity at which corrective mechanisms cannot form in time to intervene.

Origin

Kahneman and Tversky introduced the term in their 1979 paper on intuitive prediction. Flyvbjerg's empirical expansion began with his fifteen-year study of planning in Aalborg, Denmark, and culminated in the database work that produced Megaprojects and Risk (2003) and How Big Things Get Done (2023).

Key Ideas

Monotonous pattern. The fallacy operates consistently across decades, countries, political systems, and infrastructure categories — it is not a local or culture-specific failure.

Two mechanisms. Cognitive optimism bias and strategic political misrepresentation together produce the forecast distortion; either alone would be correctable.

Uniqueness bias. The planner's conviction that this case is exempt from the base rate prevents the comparison with prior cases that would expose the error.

Incentive compatible. The structure rewards optimism and punishes realism, making strategic misrepresentation individually rational even when collectively destructive.

Reference class forecasting as correction. Forcing the planner to identify comparable completed projects and calibrate against their actual outcomes produces the most reliable known improvement in forecast accuracy.

Debates & Critiques

Some critics have argued that Flyvbjerg's database overweights high-profile failures and that the true base rate of megaproject performance is less catastrophic than the headline numbers suggest. Flyvbjerg's response has been empirical expansion: the database has grown to include routine projects, not just notorious ones, and the pattern persists. Others argue that the planning fallacy is better understood through organizational rather than cognitive frameworks. Flyvbjerg's position is that both operate simultaneously and reinforcingly — the cognitive bias produces the optimism that the political incentive rewards.