Norway fixed half the problem. The other half is buildable.

Norway has been curing the overrun disease for 25 years

Norway has been proving it for a quarter of a century. Since 2000, every state project above one billion Norwegian kroner must pass external quality assurance, known as QA2, before parliament will fund it. When NTNU researchers Morten Welde and Ole Jonny Klakegg followed 111 large government projects through the regime, the average project finished 5 percent under budget. Three of four stayed within their cost frame.

The result is not a fluke of one dataset. An earlier analysis by the same pair in IEEE Transactions on Engineering Management covered 96 projects under the regime: only about one in four overran, and the average project was delivered 4.4 percent below budget. Against an international record where going over budget is the norm, Norway made coming in under budget the expected outcome.

What Norway actually built

The machinery is deliberately simple. After a string of overruns in the 1990s, the Ministry of Finance introduced a state project model with two checkpoints. QA1 tests the concept: is this the right project at all? QA2 sits immediately before parliament's funding decision: external consultants on framework agreements review the cost estimate and the management documentation, and parliament fixes the project's cost frame on the basis of that review. Above one billion kroner, no project gets money without passing through.

The gate exists to answer one question independently: is this estimate realistic? The numbers it has produced:

Costs still grow 40 percent, before the gate

Because the growth happens before the gate. Welde and James Odeck examined Norwegian road projects in Transport Reviews in 2017 and found that estimated costs grew by 40 percent on average during the planning stage, before the funding decision. After the decision, the same projects finished about 2 percent under. The regime disciplines everything downstream of the gate. The escalation lives upstream of it.

Upstream is document work. Alternatives are investigated or quietly dropped, impact assessments are produced, design choices harden, review rounds run their course, and the estimate takes shape along the way. A gate can verify the estimate that emerges from those years. It cannot reach into them. It does not evaluate the alternative nobody had time to draw up, and it does not re-test the assumption that got locked in three years earlier. The pattern matches what we traced across Sweden, the Netherlands and the international record in Cost overruns are born before the first shovel: cost growth concentrates in planning, and the mechanism is lock-in.

This is worth stating precisely. Norway's remaining 40 percent is not a governance hole. Norway already runs the most admired project governance in Europe. It is a tooling hole: the front end runs on documents produced and checked by hand, at human speed, so few alternatives get evaluated and early choices set before anyone can test them.

Whether the Norwegian result travels is unproven

Not proven, and the researchers say so themselves. 111 projects is a modest sample. Norway is small and centrally governed, and the Ministry of Finance runs the scheme directly. Whether the same regime would produce the same numbers in a larger, more decentralised system has not been demonstrated. Sweden is the uncomfortable comparison: there is no equivalent external gate, and Riksrevisionen has shown that the government almost never reconsiders a project once it has entered the national plan, however much its estimate grows.

But notice which way the caveat cuts. Even in a small, tightly governed country, with a regime the rest of Europe studies with envy, 40 percent of the cost escalation happens where the gate cannot see. If that is the residual in the best case, governance alone was never going to be the whole answer.

Bending the rest of the curve

Norway fixed verification at the decision point. What remains is the front end: the two to five years in which alternatives are, or are not, evaluated, and documents are produced and reviewed at human speed. That half is buildable now, because it is document work, and document work no longer has to run at human speed.

Machine speed does not mean less scrutiny or shorter deliberation; overruns are not cured by deciding faster. It means more scrutiny per year: three evaluated alternatives where the fee covered one, and review of every document instead of a sample. The gate tells you whether the estimate is realistic. Machine-speed front-end work changes what the estimate is by the time it reaches the gate. Combine the two and the whole curve bends. Norway proved the first half of the title. The second half is an engineering problem, and engineering problems get solved.