Why 70% of AI Initiatives Fail: You're Running a Tech Project, Not a Transformation

Adding change management to a tech project doesn't make it a transformation. It makes it a tech project with better PowerPoint.

Your AI initiative is probably going to fail.

Not because the technology is wrong. Not because your team isn't smart. Not because you picked the wrong vendor or the wrong use case.

It's going to fail because you're running a tech project and calling it a transformation. And those are two fundamentally different things.

McKinsey's research shows that 70% of transformation initiatives fail to achieve their goals. Not 70% of bad ones. 70% of all of them. Including the ones led by the very firms publishing the research. For digital transformations specifically, some research puts the number as high as 87%.

I've spent 20 years watching this pattern up close. As an auditor at a Big Four firm, I evaluated the controls that were supposed to make transformations work. I led transformation across 80+ locations at a major food company. I ran a finance strategy team driving initiatives across 100+ countries at a Fortune 500.

None of that is to sound impressive. You can't even change how people file their expense reports in an environment like that without an army of people working for months. That's table stakes in enterprise. The point is that I've had a front-row seat to what works, what fails, and why. And the organizations that succeeded and the ones that failed looked nothing alike in their approach, even when the technology was identical.

The difference is structural. And I'm starting to think it might be unfixable with the current playbook.

That's a hard thing to say after spending a career in transformation work. But the pattern is too consistent to explain away. The problem isn't that organizations lack good intentions or smart people or adequate budgets. The problem is where they start.

The Six Layers Nobody Talks About

Every successful transformation I've been part of touched six layers. In this order.

1. Leadership

Can your leadership team articulate what the organization looks like on the other side of this initiative? Not "we'll use AI." That's a tool choice. What does the business actually become? What decisions get made differently? What capabilities exist that don't exist today?

When I led transformation across 80+ locations, it didn't start with technology. It started with leadership defining what the culture needed to become. What accountability looked like. What "good" meant in concrete, measurable terms. Without that clarity, every location interpreted the transformation differently. Some moved fast. Some waited it out. Most hedged.

I've seen the same pattern everywhere since. When leadership can describe the destination in specific terms, teams navigate toward it even when the plan breaks down. When leadership says "we're doing AI," teams build what they already understand, with a new label on it.

2. Governance

Who makes decisions about this initiative? Not the project manager. The person with authority to resolve real conflicts.

Because real conflicts are coming. The VP of Sales is going to say "my team isn't changing how they work." IT and Operations are going to disagree about data ownership. Legal is going to raise concerns that slow everything down. Two departments are going to want the same data governed by different rules. Someone's territory is going to get threatened.

Most AI initiatives have a project plan. They don't have governance. There's no escalation path. There's no clear authority when the initiative threatens someone's power base. There's no structure for the inevitable moment when organizational physics pushes back against the change.

At the Fortune 500 level, the initiatives that worked had governance structures that could absorb political conflict. The ones that didn't had project managers trying to negotiate between VPs with no mandate to resolve anything. I've written about what happens when organizational politics overpower execution. Governance is how you prevent that from killing your initiative.

3. Workforce

Which roles change? Which new skills are needed? Who gets displaced, retrained, or reorganized?

This is where most "change management plans" live. And here's the uncomfortable truth: a good change management plan operating at this layer is addressing maybe one-sixth of the problem.

Change management typically means: communicate the change, train people on the new system, manage resistance. That's necessary work. But it assumes the two layers above it are solid. It assumes leadership has defined a clear destination. It assumes governance can absorb the conflicts that role changes create.

When those layers aren't solid, change management becomes the scapegoat. "We did everything right. We communicated. We trained. We had champions. It still failed." Of course it did. You built layer three on layers one and two that didn't exist.

4. Data

Is your data actually ready for what you're trying to do?

Not "we have data." Is it clean, structured, accessible, and governed in a way that AI can actually use?

Gartner predicts that through 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data. That's not a technology problem. That's an organizational readiness problem masquerading as a technology problem.

I can't count how many AI initiatives I've seen discover their data problem in month four. By then, you've spent budget, made promises, and set timelines. Now you're trying to retrofit data quality into a project that assumed the data was ready. The project manager is scrambling. The timeline is slipping. And someone is giving a status update that makes it sound like they're "on track with minor adjustments."

The data is never ready unless someone explicitly made it ready. If nobody owns data quality as their job, your AI initiative owns it by default. That's not where you want to discover it.

5. Technology

This is where most organizations start. Which platform? Which vendor? What models? How do we integrate with our existing stack?

Technology is layer five of six. Not layer one.

When you start here (and most organizations do), you're working backwards. You pick a tool, then try to find problems it solves, then try to get people to use it, then discover governance doesn't exist for the conflicts it creates, and leadership never defined what success looks like beyond "we implemented AI."

That's not transformation. That's a vendor demo with a budget attached.

I get why organizations start here. Technology is tangible. You can demo it. You can put it in a slide deck. "We're implementing GPT-4 across customer service" sounds like progress. It feels like movement. And it is movement. Just not necessarily in the right direction.

By the time you reach Technology in a real transformation, the choices are almost obvious. You know what the organization needs to become. You know who decides what. You know which people need which skills. You know what your data can support. Now pick the technology that serves all of that.

6. ROI

This is the layer everyone thinks they understand. Almost nobody uses it correctly.

ROI isn't just measurement. It's a steering mechanism.

Here's what I mean. If you're measuring cost savings and time savings from your AI initiative, you're measuring whether you're doing the old thing faster. That's optimization. It confirms you're heading deeper into the direction you were already going.

Transformation ROI asks a different question: are we heading toward a new direction? Are we building capabilities that didn't exist? Are we making decisions we couldn't make before? Is the organization becoming something fundamentally different, or just something more efficient?

The transformations that worked at that scale didn't just save money. They changed what was possible. The measurement wasn't "we did this 30% faster." It was "we can now do something we couldn't do before, and here's what it's worth." That's directional ROI. It tells you whether you're transforming or just optimizing your way deeper into the old model.

When you measure cost savings, you're confirming the status quo. When you measure new capability, you're confirming transformation. Most organizations never ask the second question.

The Change Management Trap

Here's the distinction that matters most. The one that organizations get wrong consistently.

A tech project with a change management plan is still a tech project.

I know that sounds like semantics. It isn't.

Change management addresses layer three. It helps people adapt to a change that's already been decided. Communication, training, resistance management. That's valuable work.

But transformation requires a complete top-down review starting at Leadership. Can the leadership team envision a future that's genuinely different? Can they define it with clarity? Can they identify the gaps between where you are and where you need to be? Can they implement across all six layers, including changes to their own behavior?

That last part is the one that kills most initiatives. And I say this with some self-awareness, because I've been the person in the room watching it happen and not having the authority to stop it.

I've watched executives greenlight AI transformations and then continue making decisions exactly the way they did before. They approved the project. They funded the technology. They hired the change management team. And they never changed a single thing about how they lead.

The organization noticed. Of course it noticed. People watch what leaders do, not what they fund. The message was clear: this transformation is for everyone else.

Across 80+ locations, the transformations that worked required leaders to change first. Not just endorse the change. Demonstrate it. Their own decision-making. Their own relationship with data and accountability. When leaders went first, the transformation had credibility. When they didn't, it had a budget and a timeline, which is a polite way of saying it had a failure date.

Where Your AI Initiative Actually Is

Most AI initiatives I see follow the same pattern:

Start at Technology. Pick a platform. Reach backwards to Data when you realize yours isn't ready. Launch a change management program that covers Workforce. Never question whether Governance exists for the conflicts ahead. Never ask whether Leadership has the capability or the will to drive actual transformation. And measure ROI as cost savings, which confirms you're making the old model faster.

That's a tech project. It might deliver value. It might save time and money. But it won't transform anything. And when it "fails," everyone will blame the technology, the vendor, or the change management execution. Nobody will point at the three layers that were never addressed. (This is also what you're really paying for when Big Four shows up: someone to run the 5-How test on the layers you skipped.)

Run your current initiative through the six layers. See where you actually stand.

The Question Behind the Assessment

The assessment above shows you where your initiative has gaps. But there's a deeper question it doesn't answer.

Transformation doesn't fail because of frameworks. It fails because the people leading it don't have the execution skills to drive change at this scale. Vision without execution capability is just aspiration.

The hardest question isn't "does our initiative cover all six layers?" It's "do I, as a leader, have the skills to operate across all six?"

I built the Execution Index to answer that question. It's a 3-minute assessment that measures the leadership capabilities behind transformation: strategic clarity, execution discipline, adaptability under pressure. The things that determine whether you can actually drive what you just assessed above.

It's free for the first 25 people. After that, it goes behind a paywall.

If you scored well on the assessment above, the Execution Index tells you whether your leadership team can sustain it. If you didn't score well, it tells you where to start building the capabilities that make transformation possible.

The Uncomfortable Pattern

Here's what I keep coming back to after 20 years of watching this.

The failure rate has hovered around 70% for decades. Better technology, better methodologies, better consultants. The number doesn't move.

That should tell us something. The problem isn't the tools. The problem isn't the frameworks. The problem is that organizations keep starting at the wrong layer and expecting a different result.

AI doesn't change this equation. It amplifies it. A successful AI transformation creates capabilities that didn't exist before. A failed one burns millions and confirms what the skeptics always suspected: that the initiative was never about transformation. It was about appearing current.

Don't run a tech project and call it a transformation. Start at Leadership. Work through all six layers. Measure whether you're heading somewhere new.

That's transformation. Everything else is a tech project with a bigger budget.

FAQ

Why do AI initiatives fail at a higher rate than other technology projects?

AI initiatives require organizational readiness across more dimensions than traditional software. Implementing a new ERP system primarily needs data migration and user training. AI transforms how decisions get made, which means leadership clarity, governance for new types of conflicts, and workforce skill development all become critical. When organizations skip those layers, AI projects fail because they're trying to change decision-making patterns without changing the structures around them.

What's the difference between change management and transformation?

Change management helps people adapt to a specific, decided change. Transformation redefines what the organization is and how it operates across six layers: Leadership, Governance, Workforce, Data, Technology, and ROI. Change management lives in layer three. You can have excellent change management and still fail at transformation if leadership hasn't defined a genuinely different future and governance can't absorb the conflicts that change creates.

How do you know if you're measuring transformation ROI or just optimization ROI?

Ask this question: are we measuring whether we do existing things faster, or whether we can do new things? Cost savings and time savings are optimization metrics. They confirm you're heading deeper into the current model. Transformation metrics measure new capabilities: decisions you couldn't make before, markets you couldn't serve, speed of response that wasn't possible. If your ROI dashboard only shows efficiency gains, you're measuring optimization.