Capital projects are usually judged during construction, when the site is busy and cost reports arrive monthly. The evidence points earlier. By the time a contractor mobilises, the scope, the budget, the schedule, the contract model and the allocation of risk have already been fixed, and most of what later goes wrong can be traced back to one of those choices. Bent Flyvbjerg, whose database of project outcomes spans more than 16,000 projects, put it plainly in a 2023 interview: projects “often don’t just go wrong; they start wrong”.
So why do so many capital projects fail before construction begins, and what should the executives who approve them do differently? The short answer is that the front end is too often run as a route to approval rather than a test of whether, and how, a project should be built. Optimism and advocacy pull cost estimates down and benefit forecasts up. Estimates prepared on a few per cent of design are turned into budgets and opening dates. Savings are banked before anyone has a plan to deliver them. Contracts then move risk to parties who cannot carry it, and it comes back later as claims.
Little of this is visible on the day of the final investment decision, yet all of it is settled there. The remedy is better decision quality before commitment: defining the problem before the solution, forecasting from the record of comparable projects, stating estimates as ranges tied to design maturity, and choosing a delivery model on evidence rather than habit.
In brief
- Overruns are mostly set up early. Nine out of ten megaprojects have cost overruns, and overruns have stayed high over the 70-year period for which comparable data exist (Flyvbjerg, Project Management Journal, 2014).
- Two forces distort early numbers. Optimism bias is self-deception; strategic misrepresentation is deliberate. Both push costs down and benefits up, and the projects that look best on paper tend to perform worst.
- An early estimate is a range, not a budget. Under AACE International’s classification for the process industries, a concept-screening estimate can carry an accuracy range as broad as −50% to +100%.
- Take the outside view. Reference-class forecasting and the UK Treasury’s optimism-bias uplifts correct estimates using the actual outcomes of comparable projects.
- Procurement is a front-end decision. Risk should sit with the party best able to manage and bear it; transferring risk a contractor cannot absorb does not remove it from the owner.
The pattern: most overruns are decided early
In a 2014 review of megaproject research, Flyvbjerg found that nine out of ten megaprojects have cost overruns, that overruns of up to 50% in real terms are common, and that benefit shortfalls of up to 50% are also common, with no sign of improvement over time or across geographies (Flyvbjerg, What You Should Know About Megaprojects and Why, 2014). His later book with Dan Gardner reports that, of more than 16,000 projects in the database, only 8.5% came in on budget and on time, and only 0.5% on budget, on time and with the promised benefits (Flyvbjerg and Gardner, How Big Things Get Done, 2023, as reported in The Independent Review, 2023).
What matters is where the damage originates. Flyvbjerg describes a “break-fix” pattern: projects are approved on optimistic or manipulated estimates, break when reality catches up, and are then often paused, reorganised and sometimes refinanced in an attempt to salvage something. His prescription is to get projects right from the outset through proper front-end management. Independent Project Analysis, a firm that benchmarks capital projects, reports that its research across more than 25,000 capital projects shows the completeness of front-end loading to be the single best predictor of safety, cost, schedule and operability outcomes (IPA, What Is Front-End Loading in Project Management?, 2025).
Our reading is that the front end is where influence is highest and the cost of change lowest. A different concept costs a workshop at the options stage; the same change after contract award costs redesign, claims and months.
Five front-end decisions that set projects up to fail
Choosing the solution before defining the problem
Flyvbjerg lists early “lock-in” among the characteristics of megaprojects: overcommitment to one concept at an early stage, weak or absent analysis of alternatives, and escalating commitment later. Once a tunnel or a plant has been announced, the options work that follows tends to justify the answer rather than test it.
The UK National Audit Office sees the same problem in government. Despite a required strategic case at each decision point, it often finds bodies struggling from the start to keep a clear focus on a programme’s objectives and on how its scope serves them; objectives are not always coherent and are not clearly prioritised when they conflict (NAO, Lessons Learned from Major Programmes, 2020). A scope not anchored to ranked outcomes will grow, and every addition will look reasonable on its own. The test before an option is chosen: can the sponsor explain the problem being solved, how success will be measured, and why a smaller, phased or non-build alternative would not do?
Writing the business case to win approval
Flyvbjerg separates two explanations for inaccurate forecasts. Optimism bias is a cognitive tendency to judge future events more favourably than experience warrants. Strategic misrepresentation is deliberate: forecasters and managers overestimate benefits and underestimate costs to improve the odds that their project, rather than a competitor’s, wins approval and funding (Flyvbjerg, From Nobel Prize to Project Management: Getting Risks Right, 2006). The first dominates where political and organisational pressure is low, the second where it is high. The cures differ: better methods address optimism, but misrepresentation also needs accountability that rewards accurate forecasts, and independent review of the numbers.
The errors do not cancel out. Flyvbjerg’s 2014 review found that for rail projects an average cost overrun of 44.7% combines with an average demand shortfall of 51.4%, and for roads an average cost overrun of 20.4% combines with a fifty–fifty chance that demand is wrong by more than 20%. With forecasting errors of that size, the business cases and cost–benefit analyses built on them are likely to mislead.
The most uncomfortable consequence is what Flyvbjerg calls the “survival of the unfittest”. The projects that look best on paper are those with the largest cost underestimates and benefit overestimates, and therefore the ones most likely to suffer the largest overruns and shortfalls. An approval process that rewards attractive numbers selects for disappointment.
A business case written to win approval is, almost by design, the business case most likely to be wrong.
Turning early estimates into commitments
An estimate is only as good as the definition behind it. AACE International’s recommended practice 18R-97 classifies estimates for engineering, procurement and construction in the process industries by maturity of project definition; the typical ranges below are at an 80% confidence interval after appropriate contingency, and AACE publishes separate practices for other sectors.
| Estimate class | Project definition | Typical purpose | Typical accuracy range |
|---|---|---|---|
| Class 5 | 0% to 2% | Concept screening | −20% to −50% low; +30% to +100% high |
| Class 4 | 1% to 15% | Study or feasibility | −15% to −30% low; +20% to +50% high |
| Class 3 | 10% to 40% | Budget authorisation or control | −10% to −20% low; +10% to +30% high |
| Class 2 | 30% to 75% | Control or bid/tender | −5% to −15% low; +5% to +20% high |
| Class 1 | 65% to 100% | Check estimate or bid/tender | −3% to −10% low; +3% to +15% high |
Source: AACE International, Recommended Practice 18R-97, 2020 revision.
Two caveats deserve board attention. For weak project systems or complex and risky projects, the high end of the range may be two to three times what the table shows. And AACE warns that organisational pressure for a predetermined value can produce a biased estimate.
The NAO describes how this plays out. Concept-stage estimates rest on high-level information, lack supplier input and depend on assumptions, such as ground conditions, that cannot be known until surveys are done. In many programmes it reviewed, government used such estimates to set budgets and completion dates, and forecasts rose as the risks crystallised; early figures used as delivery targets also push delivery bodies and suppliers towards unrealistic expectations (NAO, 2020). The NAO welcomes publishing ranges instead, citing the 2020 estimate for Phase One of High Speed Two of £35 billion to £45 billion, with services starting between 2029 and 2033, but warns that decision-makers must understand what a range represents and what could push outturn beyond it.
Banking savings and schedules with no plan behind them
Affordability pressure produces a particular kind of optimism: savings that are assumed rather than planned. The NAO found that HS2 Ltd included £4.9 billion of savings in its April 2017 estimate, from efficiencies, design and scope changes and price reductions, but had not turned them into a programme of work. The savings were not delivered, and the elements with the most assumed savings saw significant cost increases in the 2020 estimate. More generally, the NAO finds it rare for bodies to realise savings as planned (NAO, 2020).
Schedule assumptions deserve the same scrutiny. HM Treasury’s optimism-bias guidance sets upper-bound adjustments for works duration as well as cost: 39% for non-standard buildings, 25% for non-standard civil engineering and 54% for equipment and development projects (HM Treasury, Green Book supplementary guidance: optimism bias, 2003, republished 2013). A schedule that assumes unprecedented productivity is a cost estimate in disguise, because delay drives cost and defers value. How to measure that during delivery is the subject of our companion piece on the capital project equation of cost, time and value.
Choosing a procurement route that moves risk instead of managing it
The McKinsey Global Institute’s study of construction productivity found mismatches between risk allocation and reward in construction contracts, and inexperienced owners struggling in an opaque market. It also observed that many contractors win work by optimising up-front pricing and then recover the lost margin through change orders and claims (McKinsey Global Institute, Reinventing Construction, 2017). An owner that selects on lowest price while transferring risks a supplier cannot carry has not reduced its exposure; it has deferred it.
The UK government’s Construction Playbook sets out the alternative. Risks should be owned by the parties best able to manage and bear them, taking account of practical capability and financial capacity; inappropriate allocation, it notes, is one of the most frequent issues the NAO raises in audits of government contracts. It asks for a delivery model assessment early enough to inform the first business case, and a should-cost model of whole-life cost during planning. It adds a point private owners also need to hear: reputational risk cannot be transferred to the supply chain (Cabinet Office, The Construction Playbook, version 1.1, 2022).
Testing decision quality before the final investment decision
Take the outside view
Reference-class forecasting replaces the project team’s inside view with the recorded outcomes of comparable projects, in three steps: identify a relevant reference class of past, similar projects; establish the probability distribution of outcomes for that class; and position the new project within that distribution (Flyvbjerg, 2006). In 2004 the UK Department for Transport and HM Treasury adopted the method for large transport projects. Using the resulting distributions, a rail project whose sponsor would accept a 50% risk of cost overrun needed an uplift of 40%; accepting only a 10% risk required 68%. For roads the equivalent uplifts were 15% and 45%.
HM Treasury’s supplementary Green Book guidance applies similar logic across project types. Its upper-bound capital expenditure adjustments are 24% for standard buildings, 51% for non-standard buildings, 44% for standard civil engineering, 66% for non-standard civil engineering and 200% for equipment and development projects. Appraisers start from the upper bound and reduce it only as the causes of optimism are mitigated, on clear evidence that should be independently verified. Where optimism bias remains high, the guidance says, approval should be withheld or given only on a qualified basis (HM Treasury, 2003).
Make every gate ask a real question
The UK Infrastructure and Projects Authority’s project lifecycle, reproduced by the NAO, frames each business case stage as a question; the table adapts it for any capital owner.
| Decision point | The question it must answer | Evidence the board should see |
|---|---|---|
| Strategic outline case | Is there a need, and which options deserve further work? | A defined problem and ranked objectives; genuinely different options; a concept-grade cost range |
| Outline business case | Which option should be invested in? | A reference-class check; an optimism-bias adjustment; a delivery model assessment and should-cost model |
| Full business case or final investment decision | Can we confidently commit to undertaking the project? | Design mature enough for a budget-grade estimate; a risk allocation that suppliers have tested; a funded contingency; an integrated schedule; a named owner for benefits |
The final question is the one most often answered by momentum. A board that has announced a project, appointed a team and spent on design finds it hard to say “not yet”, which is precisely when the question matters most.
What this means for boards and investment committees
For chief executives, chief financial officers and investment committee chairs, the implications are practical. Before approving a capital project, ask:
- What class of estimate is this, and what range does it carry? A single figure with no stated basis should not be approved.
- What reference class was used, and where does our number sit? If the project is cheaper and faster than most comparators, the burden of proof is on the sponsor.
- Which savings and productivity assumptions have a delivery plan? Unplanned savings belong in the risk register, not the budget.
- Who produced the numbers, and who gains if they are low? Separate advocacy from estimating, and hear independent assurance before commitment.
- Why this delivery model? Ask for the evidence behind the contract route and the risk allocation it implies.
- What would make us stop? A project with no stop criteria has no real gate; our analysis of the stop question in portfolio governance sets out how to build them.
Two further disciplines help. Fund in stages, releasing capital as definition and delivery confidence improve, as explored in how organisations connect capital to projects. And treat time spent on definition as an investment: Flyvbjerg and Gardner’s principle, “plan slow, act fast”, captures the trade-off. The baseline set at the final investment decision also becomes the yardstick for every variance reported during delivery. A weak baseline makes even good reporting misleading, which is why estimating and project controls belong in the same conversation from the start.
Why the front end is everyone’s business
Front-end quality is not only a concern for engineers and estimators.
- Portfolio leadership. The survival of the unfittest is a portfolio failure. Flyvbjerg notes that inflated benefit–cost ratios either start projects that are not viable or displace better ones that would have been chosen had true costs and benefits been known.
- Technology and transformation. Flyvbjerg reports that one in six large information and communications technology projects becomes a statistical outlier on cost, with an average overrun for those outliers of 200% in real terms (Flyvbjerg, 2014). The same disciplines apply to technology and transformation investment.
- Infrastructure owners. National pipelines magnify the effect, because an optimistic appraisal process misallocates capital across a whole programme; see our article on investing in infrastructure well.
- Results and value. Benefits promised at approval become the standard for judging the finished asset. If they were overstated, no amount of good delivery will close the gap.
Where the front end sits in the project economy
Project Economy Forum frames the life of a project as Decide → Fund → Deliver → Prove. The front end covers the first two steps, and the others inherit what they settle: delivery works within the scope, budget, schedule and contract fixed at commitment, and proof is measured against the benefits promised at approval. When those commitments are unrealistic, delivery teams are blamed for outcomes that were designed in, and the organisation learns the wrong lesson. Seen this way, the project economy is not only about building more; it is about deciding better what to build and on what terms.
That sequence is the organising idea of the Forum’s Dubai edition on 27–28 January 2027, where the second day follows a single project from first decision to final proof, and where capital projects and CAPEX form one of the eight areas.
Keeping score on your own forecasts
The deepest fix is institutional rather than procedural. Every organisation that builds repeatedly holds its own reference class, in the gap between what it approved and what it got, but few keep it. Recording the estimate, schedule and benefit forecast at each gate, then comparing them with outturn once the asset operates, turns individual disappointments into evidence. Over time that record shows sponsors, estimators and boards how optimistic they have actually been, and makes the next front end harder to game. The most valuable thing a capital programme can build before construction may be a truthful memory of its own past.
Frequently asked questions
What is front-end loading in capital projects?
Front-end loading is the early planning and definition work of a capital project, in which the business case is translated into a defined scope and then into a project ready to execute. Independent Project Analysis describes three stages, business planning, scope development and project definition, each ending in a gate, with the last ending in full funds authorisation.
What is the difference between optimism bias and strategic misrepresentation?
Optimism bias is an unintentional tendency to see future outcomes more favourably than experience justifies. Strategic misrepresentation is the deliberate understatement of costs or overstatement of benefits to win approval. Both inflate business cases; the first is addressed mainly by better forecasting methods, the second also needs accountability for forecast accuracy.
Sources
- Bent Flyvbjerg, What You Should Know About Megaprojects and Why: An Overview, Project Management Journal (2014). arXiv
- Bent Flyvbjerg, From Nobel Prize to Project Management: Getting Risks Right, Project Management Journal (2006). arXiv
- Penguin Random House, How Big Things Get Done by Bent Flyvbjerg and Dan Gardner (2023). Penguin Random House
- The Independent Review, Book review: How Big Things Get Done by Jody W. Lipford (2023). The Independent Review
- Thought Economics, Bent Flyvbjerg on megaprojects, interview (2023). Thought Economics
- Independent Project Analysis, What Is Front-End Loading (FEL) in Project Management? (2025). IPA
- National Audit Office, Lessons Learned from Major Programmes, HC 960 (2020). NAO
- AACE International, Recommended Practice 18R-97: Cost Estimate Classification System, as Applied in Engineering, Procurement and Construction for the Process Industries, sample (2020). AACE International
- HM Treasury, Green Book supplementary guidance: optimism bias (2003; republished on GOV.UK 2013). GOV.UK
- McKinsey Global Institute, Reinventing Construction: A Route to Higher Productivity (2017). McKinsey
- Cabinet Office, The Construction Playbook, version 1.1 (2022). GOV.UK