Data May Be the New Oil Spill: Alexander Budzier on Project Failures and AI's Promise

In the world of project management, the statistics are sobering. Only 1 in 200 projects are delivered on budget, on time, and with the expected benefits. This appalling stat comes from Alexander Budzier, Fellow in Management Practice at the Saïd Business School, University of Oxford, and Director at Oxford Global Projects, who recently joined us on the Project Flux podcast.

Budzier's career has been dedicated to understanding why projects fail and how we can set them up for success. His insights come from decades of research across sectors ranging from IT and banking to construction and infrastructure. His latest book, "Intelligent Change," and his recent Harvard Business Review article on "The Uniqueness Trap" offer valuable lessons for anyone involved in project delivery.

As Budzier succinctly puts it: "A well-set up project is not a guarantee for success, but a badly set up project is a guarantee for failure." This wisdom forms the foundation of our fascinating conversation about project pitfalls and how AI might transform the landscape.

The Uniqueness Trap: Why Projects Spiral Out of Control

One of the most intriguing findings from Budzier's research is what he calls "the uniqueness trap." When project teams believe their project is entirely unique, there's a 40% chance of the budget doubling or experiencing even higher cost overruns.

Why does this happen? Budzier explains that in many organisations, especially with IT projects, system replacements happen infrequently—often with gaps of a decade or more. During this time, staff turnover means institutional knowledge is lost.

"If you think about it, what's the average tenure in a job in IT these days? About three to four years," Budzier notes. "If you're doing a project in an organisation... who is left in the organisation who has implemented last time our CRM system? You go around and can't find anybody anymore. They've all moved on."

This knowledge gap leads teams to believe they're charting entirely new territory, when in reality, they're often facing challenges that have been solved elsewhere. The uniqueness trap becomes a self-fulfilling prophecy, leading to costly overruns and project failures.

Technical Debt: The Oil Spill Nobody Wants to Clean Up

While many in the industry talk about data being "the new oil" in the context of AI, Budzier offers a more sobering perspective: "Data may be the new oil spill, this mess that is really hard to clean up."

He illustrates this with the cautionary tale of Queensland's payroll system. What began as an ambitious project to create a payroll system for all public sector employees in Queensland was eventually scaled down to just the healthcare sector. The project became infamous for creating countless errors—but as Budzier points out, these weren't new errors created by the system, but existing errors in the old system that were simply scaled up.

The result? At its peak, nearly 2,000 people were employed just to reconcile calculations between what the system said people should be paid and what they should actually receive. Ten years later, around 900 people are still performing this function.

"That's kind of the tech debt," Budzier explains. "Everybody knew the old system wasn't working, but nobody really wanted to talk about it because some people were overpaid, and that asks uncomfortable questions."

This technical debt presents a significant challenge for AI implementation. Before organisations can fully leverage AI tools, they may need to address the "oil spill" of messy, incorrect data that has accumulated over decades.

AI's Transformative Potential in Project Management

Despite these challenges, Budzier is optimistic about AI's potential to transform project management. He points to historical precedent: between 1987 and 1995, white-collar productivity in construction doubled with the introduction of computers and planning software like Primavera and Microsoft Project.

Today, AI offers several promising applications in project management:

• Advanced statistical models for forecasting: Traditional earned value management uses linear forecasting, but projects typically follow S-curves. AI can provide more sophisticated, non-linear forecasting models.

  
  

• Finding and organising information: AI excels at discovering concepts and topics across documentation, helping teams navigate complex regulatory requirements and compliance issues.

  
  

• Natural language interfaces: AI can help users access and combine data from multiple systems without needing to know specific database languages or understand complex data structures.

  
  

• Agentic AI for development: Systems like Amazon Q demonstrate how multiple AI agents can work together—one to develop a strategy, another to write code, a third to test it—all orchestrated to solve complex problems.

"AI has some really good statistical models that are advanced statistics that go way beyond Excel," Budzier notes. "If you have to think about a complex analytical task, AI is good at that."

The Human Element: Where Are the People Making Things Better?

Perhaps the most thought-provoking part of our conversation centred on the human element in AI-enhanced projects. Budzier shared a story about a Toyota lean guru visiting BMW's highly automated Mini factory in Oxford. After touring the robotics-filled production line, the visitor asked a simple but profound question: "Where are all the people who are trying to make this better?"

This question cuts to the heart of how we should think about AI in project management. The goal isn't just to automate existing processes but to free up human creativity to improve those processes.

"You might not need the people on the assembly line, but you might need a lot of people standing next to the assembly line, whose job it is to make it a little bit better every single day," Budzier explains.

This perspective is especially important given the current state of project management. If we simply use AI to automate our current approaches—which deliver success only 0.5% of the time—we're missing the opportunity for real transformation.

As Budzier puts it: "Hope is not a strategy. You don't get lucky in this industry." Success requires deliberate effort to learn from past projects, clean up data, build trust, and continuously improve processes.

Looking Ahead: Measurement and Improvement

Budzier's upcoming book, "How to Measure Anything in Project Management," promises to take us deeper into the world of project metrics beyond simple red-amber-green indicators. By making projects more measurable, we can make better decisions and drive continuous improvement.

The future of project management with AI isn't just about automation—it's about augmentation. It's about using AI to handle routine tasks while empowering humans to focus on making processes better. It's about learning from past projects rather than reinventing the wheel. And it's about recognising that while a well-set-up project doesn't guarantee success, it dramatically improves the odds.

You can hear the whole podcast in the podcast area of our website https://www.projectflux.ai/podcast

Want to stay updated on the latest in AI and project management? Subscribe to the Project Flux newsletter for weekly insights, tools, and expert perspectives. And don't miss Alexander Budzier's latest book, "Intelligent Change," available now, and his upcoming "How to Measure Anything in Project Management," coming in October 2025.