THE DIGITAL BACKBONE

DALONE ARTICLE  ·  INFRASTRUCTURE & AI

The failure patterns in complex infrastructure delivery are not unique to construction. They are arriving in AI programmes right now, and the organisations that recognise them early will be the ones that survive scale.

Rod Terry  ·  The Digital Backbone  ·  ~6 min read

SHORT VERSION

There is a point in every large infrastructure project where the plan starts to drift. Not fail. Not collapse. Just drift. Teams are ready but materials are not there. The work orders go out but cannot be completed. KPIs are set but the system underneath cannot support them.

That gap between planning and field reality is where most delivery failures originate. Not in strategy, but in execution. This is a pattern has been consistent for decades. One study of 800 major projects found that, on average, they ran one year behind schedule and 30 percent over budget. Global labour productivity in construction has averaged only 1 percent annual growth over the past two decades, against 3.6 percent for manufacturing.

The cause is not incompetence; it is the behaviour of complex systems under pressure.

When sequencing assumptions fail, dependencies get missed, and that is when rework starts. When At scale, that does not stay contained. Small delays cascade. And when organisations respond by pushing harder rather than fixing the system, pressure amplifies failure rather than correcting it.

Five patterns show up consistently across the environments I have worked in:

•     You do not fix performance. You fix systems. If one person struggles, it is probably them. If a team struggles, it is the structure.

•     Bad data breaks everything upstream. You cannot outwork incorrect inputs, and it is one of the least controlled parts of most delivery systems.

•     Scale exposes what low volume hid. Constraints that appeared manageable in a pilot do not stay manageable when volume doubles.

•     Workforce is the real constraint. It takes roughly two years to make someone effective in a complex infrastructure environment, not trained, effective. Planning still assumes labour is scalable. It is not.

•     Cost does not come from spend. It comes from flow. Reducing teams creates bottlenecks. Adding the right support roles often produces more output, not less.

The organisations that succeed will not be the ones with the best strategy. They will be the ones that built systems capable of holding under pressure.

Here is what matters for AI: we are watching the same pattern repeat.

RAND Corporation research shows more than 80 percent of AI projects fail to deliver their intended business value. That is roughly twice the failure rate of comparable IT projects without AI.

S&P Global found that 42 percent of companies abandoned most of their AI initiatives in 2025, up from 17 percent the year before.

MIT Project NANDA found that 95 percent of organisations deploying generative AI saw zero measurable return.

These are not technology failures. They are system failures. The model works. The infrastructure, the data quality, the integration, the workforce capability, the operational controls, those are where projects collapse. Exactly the same mechanisms that have been killing major infrastructure delivery for decades.

The first wave of AI at scale will not fail in development. It will fail when it tries to operate; under utility constraints, workforce shortages, integration gaps, and coordination breakdowns that no pilot ever exposed. The organisations that survive it will be the ones that built operational foundations before they scaled, not after.

Stay Connected

If you're interested in how AI is actually delivered; across infrastructure, energy, networks, materials, and supply chains, please subscribe to The Digital Backbone:

https://digitalbackbone-be8806.beehiiv.com/

Disclaimer: The views expressed in this article are the author's own and are intended for general information and discussion purposes only. They do not represent the views of any employer, organisation, or client.

© Rod Terry / The Digital Backbone. All rights reserved.

EXTENDED VERSION

There is a point in every large infrastructure project where the plan starts to drift. Not fail. Not collapse. Just drift.

Teams are ready but materials are not there. Work orders go out but cannot be completed. KPIs are set but the system underneath cannot support them.

That is where leadership actually shows up; not in strategy, in execution. And this is not an opinion. One study of 800 major projects found they ran, on average, one year behind schedule and 30 percent over budget (McKinsey, The Art of Project Leadership, 2017). Globally, labour productivity growth in construction has averaged only 1 percent per year over the past two decades, against 2.8 percent for the total world economy and 3.6 percent in manufacturing (McKinsey Global Institute, Reinventing Construction, 2017).

The cause is not incompetence. It is the behaviour of complex systems under load.

Most environments do not collapse overnight. They degrade. You start to see duplication, rework, teams working outside process, unclear priorities, systems slowing people down.

I have walked into environments running at 30 percent of their KPI target. Not because people did not care. Because the processes were outdated, structure did not exist, data was poor, and roles were misaligned. The response to that “push harder” made it worse, not better. When systems are inefficient, pressure amplifies failure rather than correcting it.

It almost always breaks in the same place: the gap between planning and field. That is where sequencing assumptions fail, dependencies get missed, and rework starts. At scale, this does not stay contained. Small delays cascade across systems, what starts as a minor issue propagates until it becomes structural.

There is also a consistent planning bias that makes it worse. Research into megaproject delivery identifies optimism bias and strategic misrepresentation of costs and benefits as primary drivers of overruns, meaning the failure is often locked in before work even begins (Flyvbjerg, Bruzelius & Rothengatter, Megaprojects and Risk: An Anatomy of Ambition, 2003).

Starting before structure is built, process is defined, and teams are aligned does not accelerate delivery. It guarantees rework. And rework is where cost hides.

After more than 20 years in complex infrastructure delivery, the following patterns show up in every environment I have worked in, regardless of sector, size, or geography.

1. You do not fix performance. You fix systems.

If one person struggles, it is probably them. If a team struggles, it is the structure around them. Every major turnaround I have seen came from fixing process flow, removing duplication, aligning roles, and clearing blockers. Not from performance management. System inefficiencies, not individual failure, are the dominant cause of delivery breakdown at scale.

2. Bad data breaks everything upstream.

Poor data leads to incorrect builds, rework, delays, and wasted labour. You cannot outwork bad inputs. It is one of the least controlled parts of most delivery systems and one of the least visible until the damage is already done.

3. Scale exposes what low volume hid.

Systems that appear stable at low volume often fail when scaled because of constraints in supply chains, infrastructure capacity, and coordination dependencies that no pilot ever tested. I have seen environments where workforce was in control, and it was material supply and infrastructure capacity that broke instead. Demand outpaced system readiness.

4. Workforce is the real constraint.

It takes roughly two years to make someone effective in a complex infrastructure environment. Not trained; effective. Planning still assumes labour is scalable on short timelines. It is not. McKinsey's 2025 infrastructure report found that more than half of construction firms report project delays due to worker shortages, and that workforce development timelines are among the hardest constraints to accelerate in infrastructure delivery (McKinsey, The Infrastructure Moment, 2025).

5. Cost does not come from spend. It comes from flow.

Most cost cutting efforts target the wrong lever. Reducing teams creates bottlenecks, delays, and rework. One of the most significant productivity improvements I have seen came from adding support roles, because it freed production teams to produce. Productivity improvements, not headcount reduction, drive sustainable cost reduction in infrastructure delivery.

AI programmes are walking straight into these same failure patterns. The names are different. The mechanisms are identical.

The evidence on AI delivery failure is significant.

RAND Corporation's 2024 research found that more than 80 percent of AI projects fail to deliver their intended business value, that is roughly twice the failure rate of comparable IT projects.

Gartner found that only 48 percent of AI pilots reach production deployment.

S&P Global's 2025 survey of over 1,000 enterprises found that 42 percent of companies had abandoned most of their AI initiatives that year, up from 17 percent in 2024.

MIT Project NANDA, covering 300-plus AI initiatives, found that 95 percent of organisations deploying generative AI saw zero measurable return on the income statement.

The diagnosis that follows these numbers almost always points to the same causes: data not ready for production, infrastructure and integration gaps, workflow failures, workforce capability shortfalls, and the absence of operational control structures. Not the model. The system around the model.

This is not a new problem in a new domain. It is the “planning to field” gap operating in a new context. The optimism bias that drives infrastructure cost overruns is the same bias behind AI business cases that assume pilots will scale cleanly. The workforce constraint that delays infrastructure delivery is the same constraint now limiting AI operationalisation, data engineers, ML engineers, and operational AI specialists are not scalable on 90-day timelines. And the cascade failure pattern that turns small construction delays into systemic overruns is the same pattern that turns a failing AI integration into queue growth, repeat tickets, and customer churn.

Most AI programmes do not fail in development. They fail when they try to operate; under constraints that no pilot ever exposed.

The organisations that navigate this well will not be the ones with the most sophisticated models. They will be the ones that treated AI deployment as an infrastructure problem from the start, with building operational foundations before scaling, investing in data quality before deploying, and respecting workforce development timelines rather than planning around them.

That means accepting that AI at scale behaves like any other complex system: it degrades before it collapses, it exposes what low volume hid, and it amplifies failure when you push harder instead of fixing the structure.

Leadership in this context is not about having the right strategy. It is about building systems that hold when the strategy meets reality.

References

McKinsey & Company — The Art of Project Leadership: Delivering the World's Largest Projects (2017)

McKinsey Global Institute — Reinventing Construction: A Route to Higher Productivity (2017)

McKinsey & Company — The Infrastructure Moment: Investing in the Expanding Foundations of Modern Society (2025)

Flyvbjerg, B., Bruzelius, N., & Rothengatter, W. — Megaprojects and Risk: An Anatomy of Ambition. Cambridge University Press (2003)

RAND Corporation — The Root Causes of Failure for Artificial Intelligence Projects (2024)

S&P Global Market Intelligence — AI Initiatives Survey (2025)

Gartner — AI Hype Cycle and Project Failure Analysis (2024–2025)

MIT Project NANDA — The GenAI Divide: State of AI in Business 2025 (July 2025)

 Stay Connected: If you're interested in how AI is actually delivered; across infrastructure, energy, networks, materials, and supply chains, please subscribe to The Digital Backbone:

https://digitalbackbone-be8806.beehiiv.com/

Disclaimer:The views expressed in this article are the author's own and are intended for general information and discussion purposes only. They do not represent the views of any employer, organisation, or client.

© Rod Terry / The Digital Backbone. All rights reserved.