Doitify
Pricing Enterprise Download Contact Us
Doitify › Project Management Questions

“Why Did Project Management Fail?”

Updated on November 12, 2026 https://doitify.com/project-management-questions/why-did-project-management-fail/
Share
Summary

Why project management fails, with real data from Flyvbjerg and PMI: root causes, a failure breakdown by cost driver, and the fixes that hold up.

Project failure is systematically under-reported because organisations measure overruns, not outcomes, and rarely publish either. The strongest evidence points to optimism bias and strategic misrepresentation at approval as the dominant cause, not poor execution.

Project team reviewing a failing project dashboard during a steering committee meeting

How often do projects actually fail?

why project management fails is a key topic in modern project management and teamwork. Often enough that failure should be treated as the base case, not the exception. The difficulty is that “failure” has no standard definition: a project can be late, over budget, short of benefits, cancelled, or technically delivered and commercially useless. Each definition produces a different rate, which is why quoted statistics vary so widely.

The most defensible numbers come from large samples with a consistent measure. They are summarised below.

Source Sample Reported finding
Flyvbjerg, megaproject dataset Large capital projects across 20+ countries Nine in ten over budget; overruns up to 50% common, above 50% not uncommon
Flyvbjerg and Budzier, HBR 2011 1,471 IT projects Average cost overrun 27%; one in six a “black swan” at 200% cost and roughly 70% schedule overrun
McKinsey with Oxford, 2012 5,400 IT projects above 15M dollars Average 45% over budget, 7% over time, 56% less value than predicted
PMI Pulse of the Profession, 2020 Global practitioner survey 11.4% of investment wasted; 114M dollars lost per 1B spent
Standish CHAOS, 2004 IT projects Average 43% cost overrun; 71% over budget, late, or scope-reduced

Two things stand out. First, the pattern is remarkably stable across decades and countries. Flyvbjerg’s finding that overruns have stayed high and constant for the seventy-year period with comparable data is the strongest single argument that this is structural, not a skills gap that training will close. Second, the distribution is fat-tailed: the average overrun is bad, and the tail is catastrophic. A portfolio that plans for the average will still be destroyed by the one-in-six case.

What do these failure statistics actually measure?

They measure cost and schedule performance, which is not the same as failure. A project can be delivered on time and on budget and still fail, because nobody uses the system or the benefits never materialise. McKinsey’s finding that large IT projects deliver 56% less value than predicted is the more important number, and it is the one least often reported.

There is also a measurement trap in how overruns are calculated. The same outcome can be described as a 50% overrun or a 33% overrun depending on whether the denominator is the original budget or the final cost. Flyvbjerg’s research documents this ambiguity explicitly, and it matters when a supplier, a client, and an auditor are arguing about the same project.

Finally, published samples skew toward projects that were visible enough to study: megaprojects, government programmes, large IT initiatives. Small internal projects fail constantly and leave no dataset behind. The true failure rate is almost certainly higher than any table can show, because the failures that matter most to a mid-sized company are the ones nobody writes a paper about.

Join Doitify Today

Move projects forward without the chaos: all your tasks, progress, and team reports in one unified workspace. Built for companies, startups, and remote teams — with a quick setup and a free trial.

What are the root causes of project failure?

Root causes sit upstream of the symptoms everyone discusses. The table below maps five causal layers to the evidence, the symptom each produces, and the fix that addresses the layer rather than the symptom.

Causal layer Mechanism How it appears Structural fix
Approval Optimism bias and strategic misrepresentation inflate benefits, deflate costs Business case that only works if everything goes right Reference class forecasting; independent estimate review
Scope No enforceable boundary; additions accepted informally “Small” changes accumulating into a different project Frozen baseline with a written change gate and cost threshold
Estimation Point estimates presented as commitments; no range, no contingency Approval at the low end of a wide range Ranged estimates, explicit contingency, management reserve
Governance Gates that cannot stop, only delay Project continues after the case has failed Kill criteria agreed at approval; funded decision authority
Execution Undersized teams, unresolved dependencies, optimistic sequencing Late integration, rework, key-person dependency Capacity planning, dependency mapping, early integration

Read the table from top to bottom and notice the direction of causation. Most organisations invest in the bottom row — better stand-ups, better tools, more training — because it is visible and within a manager’s control. The evidence says the top rows dominate. PMI’s 2020 data found that organisations undervaluing project management as a strategic competency report 67% more projects failing outright, which is a governance finding, not a tooling one.

Why is “poor communication” usually a symptom rather than a cause?

Because communication failures are how upstream problems become visible. When a scope decision is made in a corridor and never recorded, the resulting confusion looks like a communication problem. When a sponsor will not hear bad news, the missing escalation looks like a communication problem. In both cases, the communication is broken because the decision rights and the reporting incentives are broken.

The same applies to “unclear requirements” and “insufficient resources,” two of the most commonly cited causes in survey data. Unclear requirements usually mean nobody had the authority to say what “done” was, or the person with that authority changed their mind without a change process. Insufficient resources usually means the plan was approved with a staffing assumption that delivery never had the power to enforce.

Treating symptoms as causes produces the classic improvement cycle: a post-mortem lists communication, requirements, and resourcing; the organisation runs a workshop on communication; the next project fails the same way. Naming the decision that was never made is what breaks the loop.

How do you diagnose a failure without stopping at blame?

Work backwards from the outcome to the decisions, and quantify each contribution. A worked example makes the method concrete.

An ERP replacement was approved at 4.1 million dollars over 18 months. It finished at 6.3 million over 32 months, a 54% cost overrun and a 78% schedule overrun. The post-mortem initially concluded “poor vendor management.” Decomposing the gap tells a different story:

  • Approved baseline: 4.1M dollars, 18 months.
  • Scope added without re-baselining: 1.15M dollars of additional reporting and integration work, approved verbally across eleven months.
  • Rework: 0.55M dollars, caused by testing a data model that had been designed before two major requirements were confirmed.
  • Extended delivery cost: 0.48M dollars, roughly eight extra months of a core team at 60,000 dollars a month.
  • Total: 4.1 + 1.15 + 0.55 + 0.48 = 6.28M dollars, against 6.3M reported.

Now the causes are addressable. The 1.15M was a governance failure: changes were authorized by someone without budget authority. The 0.55M was a sequencing failure, visible in the plan as a design freeze scheduled after build. The 0.48M was a consequence of both. “Poor vendor management” explains none of it and fixes nothing.

The same decomposition works on schedule. Thirty-two months against eighteen is fourteen months of slip; attributing them to scope (seven), rework (four), and dependency delays (three) gives three different corrective actions rather than one vague lesson.

What would have prevented it?

Four structural fixes, in the order they would have helped.

Reference class forecasting. Instead of building the estimate from the plan alone, compare it with how similar projects actually performed. If the organisation’s last four ERP replacements overran by an average of 27%, then a 4.1M estimate should be presented as 5.2M with a range, and approved at a figure that reflects the historical distribution rather than the plan’s optimism. A 2.0M estimate with a 27% reference class uplift becomes 2.54M, and using an 80th percentile uplift of about 45% gives 2.9M. That is not pessimism; it is the empirical distribution of comparable work.

A change gate with teeth. Any addition above a stated threshold — say 2% of budget or five working days — requires a written decision by the budget holder, with the corresponding adjustment to cost or date. This single control would have removed 1.15M dollars of the overrun above.

Funded risk. Risks that are logged but unfunded are decoration. Contingency inside the baseline, management reserve above it, and explicit release criteria turn risk management from a register into a budget line.

Forecasts that are safe to deliver. If a rising estimate at completion triggers blame, the next forecast will be optimistic and the organisation will manage on fiction. The fix is procedural: forecasts are reviewed for drivers, not for blame, and a bad forecast is treated as a finding that protects the business.

Which fixes have evidence behind them?

Reference class forecasting has the strongest empirical support, because it directly addresses the bias that dominates the data. It is standard practice in UK government major project appraisal for exactly that reason, and it is cheap: it requires a database of comparable projects, not a new methodology.

Governance maturity has survey-level support. PMI’s Pulse research links organisational valuation of project management to fewer outright failures, and its waste figure — 11.4% — gives a sense of scale. On a 50 million-dollar annual portfolio, that is 5.7 million dollars a year. Spending 200,000 dollars on estimation discipline and gate reviews is a rounding error against that number.

Tooling has the weakest evidence as a failure fix, despite being the easiest to buy. A better board does not fix an unfunded contingency or a sponsor who will not hear bad news. Tools help most after the structural fixes are in place, by making the data visible and the history auditable.

When is failure the right outcome?

Sometimes stopping is the success. A project that is cancelled in month four after a genuine test of the business case has saved the organisation the remaining budget and freed the team for work that matters.

The problem is that most organisations have no mechanism for a dignified stop. Cancellation is treated as a career event, so projects continue on sunk-cost momentum long after the case has failed. Flyvbjerg’s work on escalation of commitment documents this pattern in detail: the Scottish Parliament project is a canonical case of good money following bad.

The practical fix is to agree kill criteria at approval, when nobody is defensive: “if integration is not proven by month six, we stop and re-plan.” Criteria agreed in advance convert a political decision into a technical trigger, which is the only version most teams can actually execute.

What does project failure look like at portfolio level?

A single project is a gamble; a portfolio is a distribution, and the distribution is what a leadership team can actually manage. This is why the failure conversation changes when it moves from the project review to the portfolio review. One project overrunning is bad luck or bad planning. A pattern of overruns across a portfolio is an estimation system producing the same error repeatedly, and that is fixable.

The arithmetic explains why the portfolio view matters more than any individual variance. If 11.4 percent of project investment is wasted, a 50 million dollar annual portfolio loses about 5.7 million dollars a year. That is roughly the cost of fifteen mid-level hires, or the entire contingency on three large programmes. Very few organisations would approve that line item knowingly, yet most approve it implicitly by never measuring it.

Failures also concentrate. In most portfolios a small minority of projects produce the majority of the damage, which is exactly what the fat tail in the Flyvbjerg and Budzier sample implies: the average overrun is uncomfortable, and the tail is existential. Two practical consequences follow.

  • Rank by exposure, not by count. A portfolio review that gives equal time to twelve small projects and one 15 million dollar programme is optimising for fairness rather than risk.
  • Review the tail explicitly. Ask which three projects would threaten the organisation’s year if they overran by 100 percent, then apply the hardest scrutiny, the most independent estimates, and the earliest kill criteria to those three.

There is a governance dividend too. When a portfolio review routinely names the projects most likely to fail, cancelling one stops being a career risk and becomes evidence that the review works. That single cultural shift is worth more than any reporting tool, because it converts the organisation’s failure data from a post-mortem archive into a forward-looking control.

How does Doitify reduce the risk of failure?

Doitify is our product, so treat this as disclosure rather than review. It addresses the layer where most failures are visible first: the gap between the plan and what is actually happening. Projects, tasks, sub-tasks, dependencies, budgets, owners, and due dates live in one workspace, with Kanban, Gantt, timeline, calendar, and workload views on the same data, plus work and performance reports that show planned against actual.

The failure-relevant part is not the views; it is that scope, cost, and completion are recorded on the same objects. When a change is added, the cost and date impact are visible in the same place the work lives, which is exactly the information that was missing in the corridor decisions described above. Reporting history also makes the post-mortem an evidence exercise rather than a memory contest.

What Doitify cannot do is refuse a bad business case or override a sponsor who does not want to hear the forecast. Those remain governance decisions, and no tool substitutes for them.

FAQ

Optimism bias at approval, often reinforced by strategic misrepresentation. Estimates are built from the plan rather than from how comparable projects actually performed, and benefits are modelled on the best case. Execution problems follow from those two numbers.

Reported overruns are larger and more variable in IT, with a fatter tail: Flyvbjerg and Budzier found one in six IT projects at a 200% cost overrun. Construction projects overrun more consistently but with a narrower distribution. Both are dominated by approval-stage bias.

Watch three signals: integration or acceptance work slipping while task completion stays high, a forecast that has not changed for several periods while risks accumulate, and change requests being accepted without corresponding budget or date adjustments. Each indicates that the plan and the reality have separated.

It reduces the cost of discovering failure, which is different. Short iterations surface bad assumptions earlier and cheaper, but agile does not fix an unfunded contingency, a sponsor who suppresses bad news, or a business case built on optimism. Those remain governance problems.

A method that forecasts a project by looking at the actual outcomes of similar past projects rather than by building up from the current plan. It corrects for optimism bias by using the empirical distribution of overruns in the reference class, and it is used in major public project appraisal for that reason.

Enough to cover the identified risks with a stated probability, typically 10% to 20% on well-defined work and considerably more on novel or early-stage work. The number should come from a risk assessment or a reference class, not from a round figure chosen to make the total look acceptable.

No. Some projects should be re-planned rather than stopped: the case may still hold once the scope and estimate are corrected. The test is whether the remaining investment is justified by the remaining benefit, ignoring what has already been spent.

Conclusion

Project management fails because projects are approved on numbers that were built to be approved, scoped without a boundary that anyone enforces, and governed by gates that cannot say no. The execution failures that fill post-mortems are downstream of those three decisions.

The data is consistent enough to plan against: nine in ten large projects overrun, one in six IT projects blows up, and roughly a tenth of project investment is wasted. Those are not reasons for fatalism. They are reasons to build the estimate from history, freeze the scope with a real change gate, fund the risks, and make the forecast a document that tells the truth.

If this post on why project management fails was helpful, you might also enjoy How Much Should Project Management Cost? and Does Project Management Have A Future?.

Join Doitify Today

Move projects forward without the chaos: all your tasks, progress, and team reports in one unified workspace. Built for companies, startups, and remote teams — with a quick setup and a free trial.

0 0 votes
Article Rating
Share
Subscribe
Notify of
guest
0 Comments
Oldest
Newest Most Voted
Table of Contents

Ready to do more with Doitify?

Bring your projects, team, and goals together in one AI-powered workspace.

Get Started
Table of Contents