What the Research Says About Decision Latency on Megaprojects

Every megaproject post-mortem eventually arrives at the same uncomfortable finding: the concrete was fine, the steel was fine — the schedule slipped and the budget blew out because someone took too long to decide something, or decided it too early with too little information and couldn’t reverse course. Researchers have documented this pattern for decades, across sectors and continents, with enough consistency that it no longer reads as bad luck; it reads as a structural feature of how large projects are governed. This article synthesizes what published evidence — academic studies, national audit reports, and industry surveys — actually says about decision latency on megaprojects: where the numbers agree, where they diverge, and what genuinely reduces it.

This is deliberately a review, not a manifesto. Megaproject research is one of the more rigorously quantified corners of management studies, thanks largely to Bent Flyvbjerg’s multi-decade dataset work at Oxford’s Saïd Business School and Oxford Global Projects, and to audit bodies — the UK’s National Audit Office (NAO) and US Government Accountability Office (GAO) — that examine major public programs line by line. Where the data is consistent, this piece says so. Where it varies by study or sector, it says that too, because flattening disagreement into one tidy statistic stops being useful the moment someone applies it.

The Evidence Base: What We Actually Know About Megaproject Performance

The most-cited data point in this field comes from Bent Flyvbjerg’s long-running analysis of large infrastructure projects: roughly nine out of ten megaprojects experience cost overruns, with overruns of up to 50% common, occurring regularly rather than as rare outliers. Performance has stayed poor and roughly constant across the 70-year period for which comparable data exists — not recent decline, but a stable pattern across generations of delivery.

Schedule performance tells a similar story. Dam projects have shown average schedule delays near 45% against plan — a project budgeted for ten years typically taking closer to fourteen and a half. Rail projects in Flyvbjerg’s dataset averaged cost overruns near 44.7% versus roughly 20.4% for roads, and rail demand shortfalls have averaged over 50% — the benefit side of the ledger is often as unreliable as the cost side. A separate peer-reviewed synthesis (Hu et al., 2023, Sustainability) cites Flyvbjerg’s broader dataset of 258 transportation projects across 20 countries: 90% overrun incidence, 28% average overrun — lower than the sector-specific figures above, a reminder that “the” overrun rate depends heavily on which projects and years sit in the sample.

Flyvbjerg’s estimate of true megaproject success — hitting cost, schedule, and benefit targets simultaneously — lands at roughly one project in a thousand; hitting any single target individually happens in only about one in ten. Whatever “success” means on a megaproject, it is the exception, not the baseline — precisely why treating overruns as isolated management failures, rather than a systemic pattern, has been so unproductive.

Recent portfolio audits confirm the pattern holds in 2025–2026 data too. The GAO’s latest assessment of NASA’s major projects found four of eighteen active projects had cost growth in a single fiscal year, collectively exceeding $500 million, with the interdependent Artemis program responsible for nearly half of all overruns tracked across 53 projects since 2009 — illustrating how delay in one decision node cascades through a linked program. PMI’s Pulse of the Profession 2026 survey found 97% of practitioners managed at least one complex project in the past year, that complex projects fail at roughly 33% versus 13% overall, and that managing complexity well raises success odds roughly fivefold.

What the Data Actually Shows: A Source-by-Source Summary

Popular writing on megaproject failure often compresses this evidence into one headline statistic. It’s worth laying out what each major source actually measured, side by side.

Why Decisions Stall: The Root Causes the Research Identifies

Optimism Bias and Strategic Misrepresentation

Flyvbjerg’s work distinguishes two mechanisms often conflated. Optimism bias is a genuine cognitive tendency — planners underestimate costs and durations and overestimate benefits without intending to deceive anyone. Strategic misrepresentation is different: the deliberate understatement of costs by sponsors competing for approval, described via what PMI’s summary calls the “Four Sublimes” — technological, political, economic, and aesthetic motivations that push decision-makers toward green-lighting projects whose real costs are knowable but inconvenient to surface. Both produce the same symptom: decisions get made on numbers everyone has reason to know are optimistic, and the correction, when it arrives, is late and expensive.

Diffuse and Shifting Accountability

The NAO’s 2025 review is unusually direct here. On HS2, successive senior decision-makers were “no longer aligned to a central purpose,” eroding political support over time. On Crossrail, governance staffed predominantly by civil engineers failed to surface operational-integration needs early enough, contributing to cost increases and delay once those gaps became unavoidable. The NAO’s broader critique: the standard UK model — a single government department as sole sponsor — doesn’t match mega-projects spanning multiple sectors and electoral cycles. When no one holds end-to-end accountability, decisions default to whoever is in the room, and that default changes as staff and ministers rotate.

Information Gaps and the Data-Decision Lag

Khahro et al.’s 2023 survey of 91 construction professionals found incomplete documentation and lack of technical competence among the top drivers of decision delay — decision-makers often can’t decide quickly because the information they need hasn’t reached them in usable form. It’s the same failure mode explored around the data discipline that keeps that model trustworthy: a dashboard that looks authoritative but lags real site conditions doesn’t speed up decisions, it just makes slow ones feel confident. Hu et al.’s 2023 systems-dynamics study found information-management technology adoption had a stronger effect on outcomes than expert participation rates alone — the bottleneck is often how fast good information reaches the decision, not whether good judgment exists somewhere in the organization.

Governance Built for Ordinary Projects, Not Mega Ones

A recurring theme across the NAO’s case studies, echoed in PMI’s complexity data, is that most organizations run megaprojects through governance designed for ordinary capital projects, scaled up rather than redesigned. Gateway reviews exist, but the NAO found them frequently satisfied before feasibility and options analysis are genuinely complete — projects get announced with budgets attached before appraisal work is finished, locking in commitments the organization can’t later walk back without political cost. This connects to the mechanics of cost overruns and how to prevent them: the same premature lock-in driving overruns is often the direct product of a governance structure that never had the authority, or the information, to say “not yet.”

What the Evidence Says Reduces Decision Latency

Reference Class Forecasting

Flyvbjerg’s proposed remedy for optimism bias, developed with Oxford Global Projects, is reference class forecasting: instead of building an estimate bottom-up from one project’s assumptions, planners compare it against a reference class of similar completed projects and adjust for that class’s historical base-rate overrun. It doesn’t eliminate uncertainty, but replaces optimism-driven point estimates with an outside view grounded in what similar projects actually cost — shortening the gap between the number a decision is based on and what reality delivers.

Clearer, Earlier-Locked Governance Authority

The NAO’s recommendations to UK Treasury read as a general governance template: define a distinct category for mega-projects triggering a different governance approach; strengthen gateway approvals so they provide genuine assurance before formal announcement, not after; give central bodies a formal board seat rather than an advisory role; and preserve the ability to stop a project if early work shows it’s too risky, rather than treating cancellation as unthinkable once announced. The NAO’s positive counterexample: London 2012’s Olympic delivery ran through a Cabinet Committee chaired by the Prime Minister, giving cross-government coordination genuine executive authority. A companion piece on this site develops this argument further around why firms need a standing decision governance board rather than a PMO alone.

Faster, More Reliable Information Flows

Khahro et al. and Hu et al.’s studies point to the same fix from different angles: standardized documentation and stronger information-management technology adoption measurably improve decision timeliness. This is consistent with the broader conversation around generative AI’s role in construction scheduling — a companion piece publishing alongside this one examines where those tools produce real gains versus where claims outrun the evidence — but the consensus here is narrower: technology helps when it closes the gap between information existing and reaching the person empowered to act on it, not simply because it exists.

Matching Leadership Structure to Complexity, Not Just Scale

PMI’s finding that managing complexity well improves success odds roughly fivefold reinforces what the NAO’s case studies show: Crossrail’s early governance, staffed predominantly by civil engineers, missed operational-integration risks that only became visible once transit operators joined the project board in 2020. The fix wasn’t more reporting; it was putting decision authority with people equipped to see the risk that mattered, while it was still addressable.

Connecting the Evidence to Practice

None of this evidence base was assembled with any single leadership model in mind — Flyvbjerg’s dataset, the NAO’s audits, and PMI’s surveys are independent lines of inquiry that happen to converge on the same diagnosis. That convergence is what makes it useful as a foundation for practice, not just academic curiosity. Mohammad Habibur Rahaman’s AI-GCLM (Global Construction Leadership Model) — built around strategic visioning, data-driven decision-making, lean and agile execution, human-centred leadership, technology and AI integration, and the newer pillars of autonomous intelligence leadership and regenerative sustainability stewardship — is best understood as one applied response to this evidence, developed through 16 years overseeing 300+ projects, not as the source of the underlying research. Where the research says latency stems from diffuse accountability, optimism-driven estimating, and slow information flow, a framework built around data-driven decision-making is, in effect, a working answer to that diagnosis. A companion piece publishing alongside this one, The AI-GCLM Framework Explained, walks through how those pillars operationalize this evidence; this piece’s purpose is narrower and prior to that — establishing, from third-party research alone, that the problem it answers is real and consistent across sectors and decades.

Key Takeaways

• Roughly 90% of megaprojects run over budget, a pattern Flyvbjerg’s research shows has held for 70 years — though average overrun magnitude varies by sector and dataset (28% in one 258-project cross-country sample; higher for rail and dams specifically).
• True megaproject success — hitting cost, schedule, and benefit targets together — occurs in only about one in a thousand projects; hitting any single target individually happens in roughly one in ten.
• National audit bodies (UK NAO, US GAO) independently find decision-making structure, not engineering difficulty, is the recurring failure point: diffuse accountability, premature commitment before appraisal is complete, and governance built for ordinary projects rather than mega-scale ones.
• Peer-reviewed survey research (Khahro et al., 2023) confirms decision delay is measurable and traceable to fixable factors — incomplete documentation, technical competence gaps, weak leadership — rather than an unavoidable feature of project size.
• The strongest-evidenced interventions are structural, not technological alone: reference class forecasting, dedicated mega-project governance with real stop authority, and leadership seats matched to the risks that matter at each phase.
• Complexity, not scale or effort, is the sharpest predictor of failure in recent PMI data — managing it well is associated with roughly five times better odds of success.

Frequently Asked Questions

Why do infrastructure megaprojects get delayed so often?

Research points to a combination of factors: optimism bias in early estimates, diffuse or shifting accountability as projects span years and change leadership, slow information flow to decision-makers, and governance built for routine capital projects rather than mega-scale ones. The UK NAO’s 2025 review found these patterns recurring across HS2, Crossrail, and Hinkley Point C — suggesting the causes are structural rather than project-specific.

What is the “iron law of megaprojects”?

Bent Flyvbjerg’s summary of megaproject performance data: projects tend to run “over budget, over time, over and over,” a pattern his research shows has stayed statistically stable for roughly 70 years. It describes an observed pattern, not an inevitability — his own subsequent work on reference class forecasting is framed as a way to break it.

How much do megaprojects typically go over budget?

It depends on the sector and dataset. Flyvbjerg’s broader work finds overruns above 50% common; a cross-country sample of 258 transportation projects cited in Hu et al. (2023) found 90% overrun incidence with a 28% average; rail projects have averaged closer to 44.7%, roads closer to 20.4%. Quoting a single overrun percentage as universal simplifies a more nuanced evidence base.

What is decision latency in construction and infrastructure project management?

The time gap between a project having the information needed to make a sound call — on scope, budget, risk, or design — and that decision actually getting made and acted on. It isn’t a standardized metric the way “cost overrun percentage” is, but the phenomenon is documented under related framings: decision-making delay factors (Khahro et al., 2023), governance and decision-making failures (UK NAO, 2025), and information-processing lag (Hu et al., 2023).

Can technology or AI actually fix megaproject decision delays?

Technology helps when it closes the gap between information existing and reaching the decision-maker — Hu et al.’s 2023 study found information-management technology adoption had a stronger effect on outcomes than several other factors tested. But the NAO’s case studies show governance failures that better dashboards alone would not have fixed. A companion piece on generative AI in construction scheduling examines this real-gains-versus-hype question further.

What actually reduces decision delays on large projects, according to the research?

Four interventions have the strongest support: reference class forecasting to counter optimism bias; dedicated governance for mega-scale projects with real authority to pause or stop before sunk costs make that politically impossible; faster, standardized information flow to decision-makers; and matching leadership composition to the risks that matter at each phase, rather than staffing boards purely by technical discipline.

Conclusion

The research on megaproject performance is unusually well-documented for a management topic — decades of Flyvbjerg’s dataset work, independent national audits from the UK and US, industry-wide surveys, and a growing body of peer-reviewed studies specifically on decision-making delay. What they converge on is not a message about better engineering or tighter budgeting alone; it’s a message about decision architecture — who has the authority to decide, how fast good information reaches them, and whether the organization can still say “no” once a project has momentum. For firms and public bodies serious about closing that gap between what the evidence shows and how projects are actually governed, that’s a research agenda worth engaging directly. Explore the broader research work behind this analysis at mrahaman.com/research, or if your organization is evaluating its own decision governance against this evidence base, get in touch to discuss what applying it would look like on your program.