What 64 Years of Olympic Cost Overruns Can Teach US
A new Flyvbjerg paper, and what it means for our work
Happy Tuesday, Transformation Friends. Another week, another opportunity to go Beyond the Status Quo.
Many of us have read post-mortems, audits, or the “lessons-learned” compendium that follow initiatives, usually when things have gone wrong. The work behind those documents is real, and sometimes it’s excellent. Capturing what went wrong, identifying the contributing factors, and circulating the findings is a genuine attempt to ensure the next program does better.
But, then, why do we keep making the same mistakes?
It’s a question that Bent Flyvbjerg and colleagues try to answer in a new study, “'Do projects learn across space and time? Evidence from the Olympics'“ published in Public Management Review (Ansar, Flyvbjerg, and Budzier 2026).
(Yes, I have a Google Scholar alert set up for any new Flyvbjerg publication. Anyway...)
The study is nominally about the Olympic Games, which are interesting because they repeat every four years and remain mostly consistent in scope. But the authors are clear that what they’re studying is a public management problem: how large, dispersed, repeating initiatives learn, and why most of them don’t. What struck me reading their findings is how directly the diagnosis maps onto how we deliver complex work in the public sector.
Today, we’ll walk through what the authors found, why their diagnosis is broadly applicable to our work, the four strategies they offer for breaking the pattern, and the precondition these strategies depend on.
Grab your morning coffee, and let’s get started.
Sixty-four years, twenty-three Games, no improvement
Here’s a high-level overview of the paper's argument.
The Olympics are, on paper, an ideal setting for organizational learning. They happen every four years, the events are largely consistent, and the technical, logistical, and operational requirements repeat from one Games to the next. By any reasonable theory, costs should come down, schedules should tighten, and risks should shrink as host cities and the International Olympic Committee accumulate experience.
The data show none of this. Across all 23 Games from 1960 to 2024 with reliable cost figures, the authors find no sustained reduction in the magnitude, frequency, or variance of cost overruns over 64 years. The average overrun runs at about 159 percent in real terms. Across 23 major project types studied in Flyvbjerg’s broader research program, only nuclear waste storage performs worse in terms of cost.
Reform efforts haven’t moved the underlying numbers. The IOC’s Agenda 2020 was meant to bring discipline. Paris 2024 was sold as a “lean Olympics” with 95 percent of venues already in place, yet it still overran by 115 percent.
The authors are emphatic that the people delivering each Games are skilled, and that tactical learning is everywhere within individual Games. The puzzle is why none of it aggregates into something durable. That’s where their diagnosis gets interesting.
Why the Olympic structure is built to forget
The authors view the starting point as a classic concept in organizational theory: the myopia of learning (Levinthal and March 1993). Organizations are systematically shortsighted in three ways. Spatial myopia: they overweight nearby experience and underweight what they could learn from elsewhere. Temporal myopia: they overweight the recent and underweight longer historical patterns. Failure myopia: they overweight successes and under-record failures, building organizational histories that flatter the institution and erase the evidence it most needs.
The authors add that in some delivery models, these three myopias compound. They call the resulting condition spatiotemporality: geographic dispersion, multi-year gaps, and the temporary form of the delivery organization combine to make higher-level learning nearly impossible.
The Olympics are an extreme case. Each Games is staged in a different city, four years apart, by a host committee that assembles, delivers, and disbands. The authors describe this as a franchise model: the IOC is a permanent franchisor that sets the rules and coordinates knowledge transfer; each host committee is a temporary franchisee.
Here’s how it compounds. In a permanent organization, distance and time are challenges, but manageable ones: people from the last project are still there for the next, tacit knowledge sits in the same building, and institutional memory persists. In a temporary delivery organization, those compensating structures are absent by design. Knowledge has to be transferred across geography, across years, and across teams that have never met. The barriers stop being additive and become multiplicative.
The pattern shows up vividly in something one of the authors heard from the CEO of a large real-estate firm: despite his organization having built projects continuously since 1973, every new project felt as though they had never built anything before. Flyvbjerg and colleagues call this the eternal beginner syndrome (Flyvbjerg, Budzier, and Lunn 2021). Public delivery models with the three features, work spread across distance, multi-year gaps between iterations, and teams that come together for one program and then dissolve, are vulnerable to the same dynamic. Many of the biggest ones have all three.
When tactical learning gets mistaken for the strategic kind
A useful distinction from the organizational learning literature, first proposed by Fiol and Lyles (1985), separates lower-level from higher-level learning. Lower-level learning is the tactical kind: procedural improvements within existing routines, better checklists, sharper handoffs, smarter logistics. Higher-level learning involves shifts in the underlying assumptions, delivery models, and norms of the organization itself.
The Olympics show abundant lower-level learning. Successive Games have produced real improvements in crowd management, venue accessibility, anti-doping protocols, and security after the tragic events of Munich 1972 and Atlanta 1996. None of that shows up in the cost data over 64 years.
The authors are critical of this. They critique what they call “learning legacies”, the post-event summaries produced after Games like London 2012, as fragmented tactical learnings celebrated as if they were strategic. The risk, they argue, is that learning legacies become political rhetoric without substantial impact.
For me, this is the most uncomfortable part of the paper. Our gate reviews, audit reports, closeouts, and lessons-learned exercises are the same genre. Some are excellent on tactics. Many produce documents that the next program team reads, but then fail to change the underlying delivery model.
A question worth asking: when we point to a “lessons-learned” report as evidence we’ve learned, what specifically did the next initiative do differently because of it? Anyone I’ve worked with knows this is a point I make all the time: we can’t call something a lesson learned until we actually learn it. Maybe my hard line on this distinction comes from when I cut my teeth as a public servant at DND, which has a robust framework and makes the explicit distinction between a lesson identified (we noticed something) and a lesson learned (we changed how we operate). Most “lessons-learned” reports I’ve seen are doing the first while claiming the second, and celebrated at that.
Four strategies, three of them radical
The authors identify four archetypal strategies that permanent organizations have used to overcome spatiotemporal barriers, each illustrated with a company that has used it well. The four are worth carrying into our own work as a menu of structural choices.
The first is incremental (Ferrari, Holiday Inn, IKEA, McDonald’s): continuous improvement within a stable structure, with clear accountabilities, codified routines, and repeatable formats prior to full modularity. In the public sector, this is the world we live in: refreshed project management methodologies, gateway and stage-gate reviews, maturity models, central-agency guidance changes, mandatory training, and “stronger” governance bolted onto each new initiative. It’s the strategy many of our delivery systems have defaulted into. The IOC has been pursuing the same approach for years through guidelines, roving expert teams, and reform packages like Agenda 2020. The 64-year data show it has not produced higher-level learning at the Olympics.
The second is centralizing (Apple): consolidate delivery in one or a few permanent locations to accumulate expertise, infrastructure, and memory. For the Olympics, the authors mean a permanent host city. The public-sector parallel is standing up a permanent delivery agency for a class of work, a major IT modernization shop, an enduring infrastructure delivery body, and a permanent benefits transformation team that builds and retains capability across successive programs. We’ve seen recent examples of this here in Canada: Build Canada Homes and the Major Projects Office come to mind. What they have in common is continuity of staff, tooling, and method that persists across successive projects and accumulates real institutional memory and expertise. (A thought: why not a major IT/transformation equivalent?)
The third is decentralizing (Toyota): distribute work across specialized permanent nodes that each become world-class at one component, with deliberate knowledge transfer between them. The Olympic version would be permanent global homes for individual events. The public-sector version would be a network of specialized permanent delivery institutions, one for digital services, one for major procurement, one for benefits delivery, one for grants and contributions, each accountable for depth in its component and each feeding learning into the network. The capability lives in the institutions and persists across initiatives.
The fourth is real options (Merck, Netflix): build flexibility in through sequential, staged commitments that can be expanded, paused, or abandoned as evidence emerges, in place of locked-in big-bang plans. In the public sector, this means structuring a major programme as a sequence of authorized stages: a six-month proof of concept, then a twelve-month pilot, then a phased rollout, with each stage funded and approved only on the evidence produced by the one before it. What cabinet and central agencies approve at the outset is the staging discipline itself, with full data feeding go-or-stop decisions at each gate. This was the idea behind the Deputy Minister Committee on Core Services (DMCore). The implementation didn’t quite meet the intention.
The authors’ verdict travels well: incrementalism alone is the one option the 64-year evidence shows does not produce higher-level learning. The other three require structural reform, with different governance, funding, and accountability arrangements. None is automatically right for any given initiative. The point is that the choice should be deliberate, because drifting into incrementalism by default produces the outcomes documented by 64 years of evidence.
The harder part is letting ourselves see it
There’s a precondition for any of these strategies to work. Levinthal and March’s third form of myopia, the failure kind, is the one most relevant to public administration. We over-remember our successes and under-record our failures because political incentives push that way. Acknowledging that a delivery model has been producing predictable failures for a long time is professionally and politically costly.
The authors invoke Rayner’s (2012) concept of “uncomfortable knowledge” and call for what they describe as radical candour: a willingness to confront challenging evidence directly, including evidence about what our own delivery model is actually producing. Without that, the four strategies are unread papers on a shelf.
This is the part of the diagnosis that’s hardest to sit with. Structural reform is difficult. The willingness to name what our structures are producing, in language straightforward enough for the people who’d have to fund and authorize a different model, is harder still. The structures themselves are downstream of whether we’re willing to look at the data with clear eyes.
Wrap up
This study offers us three key takeaways: a structural diagnosis for why dispersed major programs keep failing in the same ways; the distinction between tactical post-mortems and the higher-level learning that actually changes the next program; and four options for strategic responses, three of which require more than tinkering. The pattern in our biggest delivery failures is structural. That’s the bad news, because tactical fixes won’t reach it. It’s also good news because the structure is ours, which means we can redesign it.
Three questions to sit with this week:
Does your delivery model have a “franchise structure”, a permanent overseer paired with a temporary delivery team that disbands at completion, and what does that team take with it when it goes?
The “lessons learned” or post-mortem report you are most proud of: what did it specifically change about how the next program was delivered? If you can’t name something concrete, what does that tell you?
If you were honest about which of the four strategies your current model is closest to, which one would it be? And is it the one you meant to pick?
Until next time, stay curious and I’ll see you Beyond the Status Quo.
References
Ansar, A., Flyvbjerg, B., and Budzier, A. (2026) ‘Do projects learn across space and time? Evidence from the Olympics’, Public Management Review. https://doi.org/10.1080/14719037.2026.2650426
Fiol, C. M. and Lyles, M. A. (1985) ‘Organizational learning’, Academy of Management Review, 10(4), pp. 803–813.
Flyvbjerg, B., Budzier, A., and Lunn, D. (2021) ‘Regression to the tail: why the Olympics blow up’, Environment and Planning A: Economy and Space, 53(2), pp. 233–260.
Levinthal, D. A. and March, J. G. (1993) ‘The myopia of learning’, Strategic Management Journal, 14(S2), pp. 95–112.
Rayner, S. (2012) ‘Uncomfortable knowledge: the social construction of ignorance in science and environmental policy discourses’, Economy and Society, 41(1), pp. 107–125.