MIT Says 95% of GenAI Pilots Are Failing at Companies - Here’s the Reason Why

Insight & Guidance For CEOs & ELTs

A recent report came out from MIT called The GenAI Divide: State of AI in Business 2025 which reveals the startling statistic that 95% of companies are unable to drive revenue growth through their AI pilot projects.

Can the 95% Failure Rate Really Be True?

Yes, the results of the MIT study are as certain as a CFO asking about ROI.

Talk to CEOs directly and you’ll find the same story: very few are seeing meaningful business impact from their AI initiatives today.

Every now and then you’ll hear a success story, but when you look closer it usually falls into one of two categories:

• A very narrow, niche use case where AI is helpful in one small corner of the business.

• A situation where the real results came from something other than AI, and AI just happened to be in the mix.

What you don’t hear is CEOs saying, “We plugged AI into Sales and our numbers took off.”

That kind of large-scale, transformational impact isn’t happening yet.

Why are companies failing with AI?

Let’s go over the 10 key possibilities of why companies are failing with AI pilots:

10 Reasons

1. Choosing the wrong use cases

2. AI technology isn’t capable enough

3. Not putting enough money into the AI pilots

4. Don’t have the right talent/skills/knowledge

5. Data problems (without the right data AI usually fails)

6. Choosing the wrong AI vendor

7. Internal team isn’t ready for the AI (for example, too much else going on)

8. Cultural dislike of AI

9. Change management not being done right

10. Company dysfunction / lack of ownership

What does MIT indicate the reason for the failure is?

The MIT study says that a key driver for failure is #4: companies are not understanding / learning the AI tools.

For example, companies don’t realize what ChatGPT should be used for vs. a 3rd party specialized AI tool.

MIT also suggests another driver is #1: that companies are using AI for the wrong use-cases.

For example, organizations are trying to increase Sales instead of using AI for back-office efficiency projects.

Is MIT correct about the reasons?

The short answer is: no.

MIT is right about the failure rate, but they’re off on why it’s happening.

Most of the companies responding to the survey probably didn’t give the full story.

Companies rarely admit when their AI initiatives aren’t working — there’s too much pride, pressure, and sometimes embarrassment.

On top of that, the researchers at MIT may have read the survey data in a way that doesn’t reflect the real issues with AI.

What’s the REAL reason behind the 95% failure rate?

It’s a simple answer and everyone (especially CEOs) wish it wasn’t true.

The real reason company AI pilots are failing at an alarming rate is that AI as a technology is just not very good (yet).

This is #2 from my list above: AI technology isn’t capable enough. It’s not mature enough for the use-cases most companies care about.

But there’s a big, big caveat.

If you spend lots of time and money you can MAKE it good enough.

In other words, “out of the box” AI is definitely not going to make results happen for your pilot. But if you spend enough you can make it work.

And most companies just aren’t spending enough on it. They are buying into the hype, tossing it into the mix and hoping for the best.

The harsh reality about AI adoption in 2025.

The truth is simple: GenAI is overhyped relative to its current maturity.

If you’re a CEO, you’ve likely been pitched AI as a miracle drug that will cure sales, marketing, operations, engineering, and customer service all at once.

The narrative goes: “Plug it into workflows, watch the results flow.”

But AI isn’t that easy. Not today.

Instead, what we really have is:

• A set of powerful but immature tools (ChatGPT, Claude, Gemini, etc.) that work well as general question & answer machines.

• Some impressive AI vendors in industries like healthcare, sales, or legal — but they require heavy investment & customization.

So when MIT reports that 95% of companies are failing, it’s not because business leaders or companies don’t know what they’re doing.

It’s because the AI tools don’t yet deliver at enterprise scale — unless you’re willing to spend like Uber, McKenzie, or JPMorgan.

What do the 5% do that makes them successful?

The 5% that are winning at AI pilot projects are spending more money on more of the right AI capabilities vs. the 95% that are failing.

Let’s go through what the typical AI spend areas are:

• A lot of product management time on discovery

• Raw AI engineering talent, straight up; or at least good training

• Access to multiple LLMs / models (not just 1) from OpenAI and other providers

• Access to AI platforms outside of just the models like Amazon Bedrock

• Access to plenty of AI developer tools like LangFlow/Chain

• Development of RAG

• Development of MCP

• Development of custom workflows & glue code

• High-end 3rd party tools, for example for voice AI

• Data clean up and set up

You have to be willing to spend on these.

That’s why large companies like Uber, McKenzie, and JPMorgan are seeing wins. They have the budget, the talent, and the patience to grind through the process.

But if you’re a typical mid-sized or even medium-large enterprise trying to implement AI with just a couple $100K and some semi-trained engineers you won’t see great results.

You might get a bit of what you want, but you’re almost certainly not going to see the big impact you’re looking for.

What should CEOs takeaway from all this?

All hope is not lost — AI will still eventually live up to much of the hype.

And even a 95% failure rate doesn’t mean AI is useless.

It’s the early days and you simply have to reset expectations.

Here are 10 things CEOs should do to succeed with AI:

1. Keep developing & hiring AI talent in your organization — nothing is more important.

2. Reset your timelines. Give your projects plenty of runway to succeed. Think multi-year.

3. Carefully increase your spend in the AI areas described above.

4. Do smaller scoped projects and get smaller wins. But keep aggregating those wins & building momentum.

5. Focus more on efficiency than revenue at first. AI is better at the former.

6. Get your data house in order. This is a must as AI thrives on good data.

7. Tie AI adoption to KPIs. Incentivize the organization.

8. Study other companies failures carefully & avoid repeating their AI implementation mistakes.

9. Run lots of small experiments…it’s the only way to know what’s going to work for your business.

10. Stay patient. New technology adoption is always challenging. Playing the long-game will serve you better. Unless of course, you need to show a big shiny new AI object to an investor. That’s different and you should totally go for that for the short-term win.

Closing Thoughts

MIT’s study wasn’t wrong about the failure rate — 95% of GenAI pilots are failing. That’s the reality of where we are in 2025.

But here’s the part the headlines don’t capture: failure at this stage doesn’t mean AI is a dead end. It means we’re still in the messy adolescence of the technology.

The companies that recognize this, that reset expectations, invest in talent, get their data ready, and play the long game, will be the ones that break free from the 95% and join the 5% who turn AI into real business value.

So don’t dismiss the MIT numbers. Use them as your wake-up call.