Why Most Company AI Projects Fail, Despite the Hype

In Part One of this two-part series, AI expert Robert van der Zwart (EO Netherlands) explains why following AI influencers and chasing new tools is a losing strategy. Research from MIT, McKinsey, RAND, and BCG shows most AI initiatives fail because companies underestimate the organizational, cultural, and operational changes required to make AI work.

A man types on a laptop computer screen using ChatGPT. — Photo courtesy Canva

I need to get something off my chest.

My Instagram feed has turned into a non-stop parade of AI gurus. Every morning, somewhere between the coffee reels and the sunrise shots, there’s another influencer looking straight into the camera, promising me that their framework, their prompt template, their “one tool that changes everything” will transform my business overnight.

And I will admit, for a while, I fell for it. I tried the tools. I watched the tutorials. I copied the prompts. Some of it was genuinely useful for small tasks. But when I looked at the bigger picture, at what it actually takes to make AI work inside a real company, with real teams, real processes, and real stakes, I realized something uncomfortable.

These influencers are not wrong that AI is powerful. They’re wrong about what it takes to make it work. And the gap between their promises and reality isn’t small. It’s a chasm.

Let’s Talk About What is Actually Happening

I started digging into the research, not the blog posts, not the Twitter threads, but the actual studies from institutions whose reputation depends on getting this right. What I found surprised me:

95% of AI pilot programs fail to achieve rapid revenue acceleration. — MIT, 2025

MIT’s Sloan School of Management looked at over 300 real AI deployments, talked to 350 employees, and interviewed 150 company leaders. Their verdict: for all the billions being poured into enterprise AI, only about 5% of companies are seeing the kind of results the influencers promise.

Now, that number deserves a caveat. MIT set a high bar: They measured “rapid revenue acceleration,” a tough standard. Some critics argue this undercounts projects that deliver real but incremental value. Fair point. But here’s the thing: Even if you soften the definition, every other major institution lands in the same neighborhood.

80% of AI projects never reach production. — RAND Corporation, 2024

The RAND Corporation, the people who advise the U.S. military on some of the most complex systems in the world, found that AI projects fail at twice the rate of regular IT projects. And the root causes? Overwhelmingly organizational, not technical.

74% of companies can’t achieve meaningful scale with AI. — BCG, 2024

A BCG survey of 1,000 C-suite executives found that only about 1 in 4 is gaining real value from AI. McKinsey’s 2025 global survey highlighted this even more: While 88% of organizations have adopted AI in some form, just 6% are actually realizing enterprise-wide value.

And the trend is not improving. S&P Global reported that 42% of companies reduced most of their AI projects in 2025, more than double the rate from the previous year. Companies aren’t just quietly failing; they’re actively pulling back.

Whether the true failure rate is 80% or 95%, the message from every credible research institution is the same: The vast majority of AI projects are not delivering on their promises.

We Have Seen This Movie Before

Here is what clicked for me: This is not a new story. We have been here before.

Remember the early 2000s? Companies “went digital” by scanning paper forms into PDFs. They took a broken paper-based process, added a computer screen, and called it a transformation. Nothing truly changed about how the work was done.

We’re doing the exact same thing with AI. We take a chatbot and bolt it onto a workflow designed 20 years ago, then wonder why the results are underwhelming.

The World Economic Forum confirms that, “Layering AI on top of broken workflows and bad data is a recipe for failure. Value shows up when you redesign workflows, roles, and handoffs — not when you add a chatbot to a dysfunctional process.”

McKinsey discovered something that makes this even clearer: Companies allocate only 10% of their transformation budgets to change management, even though cultural resistance remains the top barrier to success. Think about that. They’re spending 90% of the money on technology and just 10% on the people who actually need to use it.

I see this pattern everywhere. AI speeds up one isolated task — maybe it drafts an email faster, or summarizes a report in seconds. But then you spend 20 minutes prompting it, reviewing the output, correcting the tone, reformatting the result, and running it past someone who does not trust it. The tool got faster. The work did not.

The Three Things That Actually Matter

So what separates that small minority of companies that make AI work from the rest of us? I dove deep on this, and the answer from McKinsey, BCG, MIT, Harvard, and RAND is remarkably consistent.

It comes down to three things. None of them is sexy. None of them will get you likes on Instagram. All of them are essential.

1. Redesign How You Actually Work

This one feels most significant, because it’s the one most companies skip.

McKinsey’s 2025 survey covered nearly 2,000 people across 105 countries. The companies that reported real financial returns from AI were 3.6 times more likely to have pursued fundamental change in how work gets done. More than half of the high performers had completely reworked their workflows before plugging in AI.

Meanwhile, across all companies using generative AI, only 21% had redesigned even some of their workflows. The other 80%? They were just layering AI on top of whatever they were already doing.

McKinsey was clear about this: Workflow redesign contributed more to meaningful business impact than many other factors they tested. It was more influential than the choice of model or the size of the investment.

Harvard Business Review picked up the same thread in its January 2025 cover story, calling it “Kaizen 2.0.” The big insight: AI’s natural-language interface means frontline employees can now drive process improvements directly; they don’t need to be engineers. But this only works when you actually redesign the process around what AI can do. You have to put people at the center of new workflows, not ask them to use a chatbot in a process built for fax machines and filing cabinets.

2. Train Your People Like You Mean It

This is where the gap between what leaders say and what they do becomes almost comical.

SHRM’s 2025 survey found that 75% of American workers expect AI to change their jobs in the next five years. But only 45% have received any recent upskilling. Nearly half the workforce is already using AI on the job, and just 31% say their employer provides any training for it.

It gets worse — 89% of business leaders say their people need better AI skills. How many have actually started meaningful training programs? Six percent.

Let that land for a moment. Almost nine in ten leaders know there’s a skills gap. Fewer than one in ten are doing anything real about it.

And even when companies offer training, they usually get it wrong. A study on Microsoft 365 Copilot adoption found that 90% of people agreed that formal training would help, yet 70% ignored the onboarding videos entirely. People don’t learn by watching tutorials. They learn by doing.

McKinsey nailed the distinction: Effective upskilling isn’t a learning initiative. It is a change initiative. People adopt AI when they know what to do differently, believe in why it matters, feel supported by their managers, and see the change reflected in the systems and incentives around them. A two-hour webinar on “How to Use ChatGPT” doesn’t check any of those boxes.

3. Fix Your Data Before You Do Anything Else

I almost didn’t include this one because it sounds so obvious. But Gartner attributes 85% of AI project failures to poor data quality, so apparently it needs saying.

The pattern is always the same. A pilot project works beautifully on a clean, curated dataset. Everyone gets excited. Then the system meets real-world data — messy, inconsistent, duplicated, poorly labeled, constantly changing — and it falls apart.

Data readiness is not a nice-to-have. It is infrastructure. It is plumbing. And like actual plumbing, nobody wants to pay for it until something breaks. But the companies that get AI right build this foundation first, not after.

But there is another problem quietly making all of this worse: Entrepreneurs and leaders are being conditioned to chase tools instead of building capability.

Every week brings another “must-have” AI platform, prompt library, automation stack, or viral workflow. The result is that many companies are trapped in a cycle of experimentation without integration, constantly switching tools without ever fundamentally improving how work gets done.

In Part Two, I will break down why tool-chasing has become one of the biggest hidden traps in AI adoption, what the research says about sustainable AI implementation, and why the companies seeing real gains are focusing less on tools and more on operational discipline, human behavior, and organizational design.

Contributed by Robert van der Zwart, an EO Netherlands member, who is a coach, keynote speaker, founder of AIPO Network, co-founder of The Clever Innovation Box, which assists companies in successfully transitioning to an AI-enabled organization. Robert recently spent five days as the member host of EO’s inaugural executive education program, the EO Stanford Graduate School of Business: AI Integration Lab and wrote about it in 3 Reasons Most Companies Are Getting AI Wrong (And What To Do About It).

The AI Illusion: Why Most Companies Fail at AI Transformation (Part 1)