Why Most Infrastructure Projects Cost Twice What Anyone Predicted

In 1957, the Sydney Opera House was approved with an estimated construction cost of seven million Australian dollars and a projected completion date of 1963. It opened in 1973 at a final cost of 102 million dollars — fourteen times the original estimate — and was delivered a decade late. The architect, Jørn Utzon, had resigned in acrimony four years before completion, and the building was finished without him, which is why some of its interior spaces are acoustically problematic to this day. The Sydney Opera House is now considered one of the greatest buildings of the twentieth century. It is also, when examined as a project management exercise, one of the most spectacular failures in the history of large-scale construction. And it is perfectly typical.

Bent Flyvbjerg, the Oxford professor who has spent three decades cataloguing megaproject performance data, found that nine out of ten large infrastructure projects come in over budget. The average cost overrun for rail projects is 44.7 percent. For bridges and tunnels it is 33.8 percent. For roads it is 20.4 percent. These are not random errors that sometimes go over and sometimes come in under. They are systematic overruns, clustered on the expensive side. The bias is consistent across countries, decades, and project types. This is not incompetence. It is a system working exactly as designed — a system designed, above all, to get projects approved, not to get them built correctly.

The Approval Problem

The fundamental dynamic of infrastructure cost overruns is simple once you understand it, and it has nothing to do with engineering difficulty. Projects must be approved before they are built. Approval requires demonstrating that a project is worth doing — that its benefits exceed its costs. The people who want projects approved — politicians seeking legacies, contractors seeking contracts, consultants seeking fees, local officials seeking economic development — have strong incentives to make the benefit-cost calculation look favorable. The easiest way to make the calculation look favorable is to underestimate costs and overestimate benefits. The people doing the estimating are almost always employed by or dependent on the people who want approval. The result is entirely predictable.

This phenomenon has a name: strategic misrepresentation. It was identified and documented by the Danish-American planning theorist Albert O. Hirschman in the 1960s, though Hirschman thought it might be useful — that if planners had to be honest about the difficulty of their projects, they might never start any of them, and some projects that ultimately prove worthwhile would never be attempted. Flyvbjerg’s more recent work demolishes this charitable interpretation. He found no significant relationship between cost overrun magnitude and project quality or usefulness. Projects that cost three times what was predicted are not systematically more valuable than projects that cost one-and-a-half times what was predicted. The overruns are not the price of ambition. They are the price of dishonesty.

The incentive structure is particularly perverse for politicians. A politician who champions a project, gets it approved, breaks ground, and then leaves office before it is completed gets credit for the ribbon-cutting and none of the blame for the cost overruns that emerge during construction. The politician who succeeds them inherits a project too far advanced to cancel — canceling a half-built bridge or tunnel typically costs nearly as much as finishing it — and therefore faces a choice between throwing good money after bad and accepting political catastrophe. They almost always finish the project. The original promoter is long gone. The accountability mechanism is broken.

This dynamic is visible in the history of almost every major infrastructure project of the past century. Boston’s Big Dig — the project to reroute Interstate 93 through a tunnel under downtown — was approved in 1982 with an estimated cost of 2.8 billion dollars. It was completed in 2007 at a cost of 14.8 billion dollars. Several of the politicians who championed the original estimates were dead by the time the final bill came in. California’s High Speed Rail project was approved by voters in 2008 with a ballot measure promising a 33-billion-dollar system connecting Los Angeles and San Francisco by 2020. By 2026, no train had run, only a segment of track had been built in the Central Valley, and estimates for the full system had ballooned to well above 100 billion dollars. The original vote was for a project that did not, in any meaningful sense, exist as described.

The Psychology of Optimism

Strategic misrepresentation is the deliberate version of the problem. There is also an inadvertent version, and it may actually be more pervasive. The psychologists Daniel Kahneman and Amos Tversky identified what they called the “planning fallacy” — the tendency for people making predictions about their own future projects to systematically underestimate costs and timelines while overestimating benefits. This is not lying. It is a cognitive bias so reliable and universal that it qualifies as a feature of normal human psychology.

The planning fallacy operates through what Kahneman called the “inside view.” When you plan a project, you think about your specific project: your team, your resources, your plan, your anticipated obstacles. You do not naturally think about the statistical distribution of outcomes for projects like yours. If you thought about the “outside view” — what fraction of similar projects came in on time and on budget, what the average overrun was — you would give a much more pessimistic estimate. But your project feels different from all those failed projects. Your team is better. Your plan is more thorough. The obstacles you’ve anticipated will be managed. This feeling is almost always wrong, but it is entirely sincere.

The reference class forecasting approach that Flyvbjerg advocates addresses the outside view problem directly: before you estimate a project, look at the historical distribution of outcomes for similar projects and anchor your estimate to that distribution. If the average rail project costs 44 percent more than estimated, your initial estimate should be adjusted upward by at least that much. If your specific project has characteristics that make it more difficult than average — complex geology, urban construction in dense areas, novel technology — you should adjust further. This approach consistently produces more accurate forecasts than conventional project estimation methods.

It has been almost universally resisted by the infrastructure industry. The reason is clear: a realistic forecast based on reference class data would show many projects to be uneconomical. Projects that pencil out at estimated cost do not pencil out at 1.5 times estimated cost, let alone double. If forecasters were required to use historically calibrated estimates, fewer projects would be approved. This outcome would be rational but is politically unacceptable to an industry whose entire business model depends on a continuous pipeline of project approvals.

Complexity as Cover

There is a third dimension to the cost overrun problem that is neither strategic misrepresentation nor cognitive bias, but genuine complexity — and the genuine complexity is real enough to be easily exploited as cover for the other two. Large infrastructure projects are extraordinarily complicated systems integrations. A major transit project must coordinate civil engineering, structural engineering, mechanical engineering, electrical systems, software systems, environmental compliance, property acquisition, utility relocation, construction sequencing across multiple contractors, supply chains that span dozens of countries, labor agreements across multiple unions, and regulatory approval processes that can require renegotiation if the project scope changes significantly. The number of ways any of these can go wrong is very large.

The problem is that genuine complexity creates the conditions for plausible deniability when things go wrong. Every cost overrun can be attributed to a geological surprise, a material supply disruption, a regulatory change, a design modification, or a labor dispute — all of which are real phenomena that genuinely affect costs. This makes accountability nearly impossible to assign and nearly impossible to enforce. The contractor blames the client for scope changes. The client blames the contractor for poor execution. The regulator blames both for inadequate planning. Nobody is lying, exactly, and yet the project costs twice what it was supposed to cost, and nothing changes for the next project.

Edinburgh’s tram project, which opened in 2014 after years of delays and disputes, provides a particularly clean case study. The project involved a contract dispute so total that the contractor and the city council stopped speaking to each other and communicated only through lawyers for two years while the partially-constructed tramway sat idle. The final cost was roughly three times the original estimate. The subsequent inquiry produced a detailed report attributing the failure to a procurement model that transferred too little risk to the contractor, a contract that was not fit for purpose, and a client organization that lacked adequate project management expertise. All of those findings were correct. All of them described conditions that anyone familiar with major infrastructure procurement would have predicted in advance. The inquiry documented a failure that was entirely foreseeable and was, in fact, foreseen by critics who were ignored.

The foreseeable failure mode is common enough to constitute a genre. Infrastructure procurement systems in most countries are designed around the fiction that competitive tendering will produce efficient outcomes. In practice, for highly complex projects, the bid that wins is frequently the bid that is unrealistically cheap — because the winning bidder either intends to renegotiate once the contract is signed and the client is locked in, or because they genuinely believe their low bid is achievable and are simply wrong. Either way, the outcome is the same: the client ends up paying more than the winning bid, usually much more, and the competitive tendering process that was supposed to control costs produces exactly the costs it was supposed to prevent.

Why Some Countries Do It Better

The pattern is not universal. Some countries consistently build infrastructure faster, cheaper, and more reliably than others, and the differences are large enough that “it’s just complicated” cannot be the explanation. Spain built its high-speed rail network, the largest in Europe, at costs per kilometer substantially below the European average. South Korea built major transit and highway infrastructure at costs that shamed most developed-world comparators. Japan’s Shinkansen network has been running for sixty years and has never had a fatal accident involving train passengers. These countries did not have access to magic engineering knowledge. They had different institutional structures.

The common factors in countries that execute infrastructure well include: strong technical capacity within government agencies so that the public sector client can actually manage contractors rather than being managed by them; procurement models that align incentives correctly rather than transferring all risk to contractors who will either price it in or fail; planning systems that provide genuine certainty on design before construction begins rather than allowing scope changes during construction; and democratic accountability structures that punish politicians for cost overruns rather than rewarding them for project initiation.

The last factor is probably the most important and the hardest to engineer. In systems where voters punish politicians for failed or expensive projects, politicians have an incentive to be honest about costs before committing to construction. In systems where voters are engaged by ribbon-cuttings and ground-breakings and disengaged by construction schedule reports, politicians have the opposite incentive. The infrastructure problem is, at its root, a democracy problem — a problem of how voters engage with complex, long-duration commitments that outlast any single electoral cycle.

The fix is not complicated to describe: require realistic forecasts grounded in historical data, impose cost overrun consequences on the organizations and individuals who produced the original estimates, build client-side technical capacity that can hold contractors accountable, and — hardest of all — create political cultures in which voters punish dishonest project promoters rather than rewarding them. These changes are straightforward in principle and politically nearly impossible to achieve, because they would reduce the flow of projects approved and therefore the flow of money to the industry that benefits from approvals. That industry is extremely well organized and extremely well connected. The taxpayers who bear the cost are diffuse and often unaware of the specific mechanisms that are picking their pockets. Until that balance of organized interests changes, the Sydney Opera House story will keep replaying, at varying levels of architectural glory and fiscal catastrophe, forever.