Why 90% of AI Pilots Fail and What the 5% Do Differently
By Bill Goodwin, MBA, Certified Scaling Up Coach · Updated November 2026
Most AI pilots fail because companies apply AI in a lot of different places, like individual productivity or problems that maybe aren’t their top problems. The 5% apply AI to their top growth constraint first and design it very well.
The 95% figure comes from MIT Project NANDA’s 2025 report, The GenAI Divide. Critics have pushed back on it for good reason: it drew on 52 interviews and 153 survey responses, and it counted a pilot as a success only if it showed measurable P&L impact within about six months. The stronger evidence is McKinsey’s 2026 State of AI survey of 1,719 leaders, which found that only 6% of organizations attribute at least 5% of their EBIT to AI. Call it 94% or 95% then that didn’t see a strong EBITDA improvement from AI initiatives.
I hear it from CEOs all the time. One told me, “I look at the bill, I see what we’re spending, and I’m wondering if it’s worth the cost.” Another put it this way: “We’re deploying AI in a lot of different places. Our spend is fairly high and everyone’s using it, but I don’t see it returning economic value to the business.”
Why do most AI pilots fail?
When I’ve worked with client companies, I’ve seen them trying to apply AI in a lot of different places, and it’s a little bit diluted: individual productivity, or problems that maybe aren’t their top problems. In McKinsey’s 2026 survey, 80% of respondents said AI improved their own productivity, but only 37% said it had a meaningful impact on EBIT, unchanged from a year earlier.
PwC’s 29th Global CEO Survey of 4,454 chief executives found that 56% had seen neither higher revenue nor lower costs from AI in the previous 12 months. Only 12% saw both.
One CEO summed it up for me: “We’re deploying AI at an individual level to improve productivity and looking at a couple of problems, but it doesn’t seem to be generating any results.” That is the 95% in one sentence.
What do the 5% of successful AI pilots do differently?
What I’ve started coaching clients on is this: you need to apply AI to your top growth constraint first, and design it very well. I’ve run some of my own AI projects, and when I push them after the biggest growth constraint, that’s where I get the most value, including economic value.
| The 95% | The 5% | |
|---|---|---|
| Where AI goes | Wherever someone has an idea | The company’s biggest growth constraint |
| Who owns it | IT, an enthusiastic employee, or nobody | The CEO, with one accountable leader |
| The process | AI added on top of the current workflow | Workflow stripped down and redesigned first |
| How success is measured | Hours saved, logins, usage | Revenue, margin or cash |
| Timeline | An open-ended pilot | A target number and a date |
Nearly three-quarters of McKinsey’s AI high performers have fundamentally redesigned their workflows, up from 55% a year earlier, compared with about a quarter of other companies.
Why does the CEO have to personally own AI?
The CEO has to personally own AI because you can’t lead what you don’t understand. Many CEOs, including leaders I work with (and me at first), were slow to adopt AI personally. That makes it extremely hard to implement something you don’t have a basic comfort level using. It’s either IT owns it or somebody else owns it.
That is starting to get better, but it isn’t where it needs to be. I’m a good example of what changes when a CEO commits. I made it an intention to learn AI and use it in my business every day or every other day, as a thought partner to work through problems and optimize how the business runs. That comfort level helped me understand it and apply it in my own business. It also lets me work with clients on where they can get the best return leveraging AI.
One thing that helped me early on is the CRIT framework from Geoff Woods’ book The AI-Driven Leader, which, coincidentally, has a partnership with Scaling Up. You give AI the context, you tell it the role you want it to play, you have it interview you with clarifying questions, and then you assign the task on what you’re looking for. I started doing that early on, 10 to 15 minutes a day, and I always use CRIT whenever I use AI for anything. Leveraging a framework like that improves the results. I wrote more about this in the Phoenix Business Journal: Mid-market CEOs who don’t use AI personally set their companies up to fail.
How should a CEO choose where to use AI?
Revenue is vanity, profit is sanity, and cash is king. So if a CEO asked me where to start, I’d pick in this order: cash first, then profitability, then revenue. There’s probably an opportunity in billing or collections to almost immediately improve cash on hand by making some changes in that part of the cash conversion cycle. Second, I’d look at profitability, whether that’s improving gross margin or lowering general and administrative expense. The last one is revenue.
That’s not always the order. With the client story below, we did revenue first, because we knew that was the biggest growth constraint for a fast-scaling company to keep growing at 50% or more a year. But cash is often a really good place to go. It’s the thing CEOs worry about running out of, the thing they have to manage most tightly, and growing your business costs money.
Step 1: Find your biggest growth constraint
The real constraint is not something you typically uncover in 5 minutes. Through pretty deep questions and analysis, we can figure out the biggest 1, 2 or 3 growth constraints, prioritize them, and then start on the top one.
I start with a very simple question: “Is there one area of the business, one process you could improve, one change on the front end, that would drive more revenue, higher profitability or more cash?”
Then I separate it into three levers:
- Revenue: What is the top constraint to driving more revenue?
- Margin: What is the top constraint to increasing your margin?
- Cash: What is the top constraint to generating more cash, or having more cash on hand to fuel your growth?
Looking at it from a P&L and balance sheet standpoint makes the constraint easier to see. Ultimately, anything we do, we want to see the results show up on the financials.
Step 2: Fix the process before you add AI
Once we identify the problem or process that AI needs to go after, we strip out the steps that don’t belong and revise the process. Then we point AI at it.
Step 3: Put a number, an owner and a date on it
When a CEO personally sponsors an AI project, here’s what it means to me:
- Stay accountable. Someone else is responsible for doing the majority of the work, but the CEO is accountable for the outcome. That mindset change is big: it’s not somebody else, it’s me.
- Stay involved. That doesn’t mean being in every meeting. It means getting sufficient updates and being able to interject with questions, almost as a working team member, but with a greater degree of accountability and authority.
- Keep using AI. Otherwise, you can’t add as much value to the process.
- Clear blockers. Blockers come up, and the CEO has the ability to clear them for the team.
That’s where a CEO becomes a strong sponsor, not just a name on the project.
Define it up front in a business case for attacking that growth constraint. When do we expect it to generate positive contribution margin? When do we expect it to break even?
Add a standing AI review to your monthly executive meeting, and make a point of asking, “Which constraint are we attacking next, and what will AI change in how we work?”
What does a successful AI project look like in a mid-sized company?
A successful AI project in a mid-sized company changes one constraint and shows up in revenue. One client of mine, an $8 million company growing 50% a year, had its growth constraint at the top of the sales funnel: new customers and revenue coming in.
They intentionally built a lead aggregation system via AI that pulls in leads across multiple states. AI then profiles each lead for buying signals across different platforms and ranks them by how likely they are to buy. It also serves those leads up to the reps much faster.
So the team now has more leads, calls them faster, and works the highest-probability buyers first. The sales team’s throughput is up 2.5 times.
It took a couple of months to really dial in the top of the funnel. Once they did, they started seeing results almost immediately.
The CEO made the difference. This CEO had a very good understanding of AI from using it firsthand, and challenged the organization to point AI at the top growth constraint. Their approach succeeded for three reasons. The CEO bought in. The CEO gave a mandate: we’re going to figure out a way to do this. And they identified the top growth constraint before they built anything.
Frequently asked questions about AI pilots
Is the “95% of AI pilots fail” statistic accurate?
I think it’s overstated, since MIT’s 2025 GenAI Divide report used a small sample and a strict six-month test. But from my experience talking to CEOs, one on one and in groups, 75 to 80% would tell me their teams are using AI to be more productive, but they don’t see it in the financials. Probably only 15 to 20% would say it’s really improving their financials. McKinsey’s and PwC’s 2026 surveys point the same direction.
I’ll say it over and over again: if you do anything, whether you hire people, build a new process, launch a new product or leverage AI, it has to show up on your P&L or your balance sheet in a positive way. Otherwise, why do it? The CEO’s job is to drive valuation, and to drive valuation, you have to make your financials better.
Is it too late to get AI right if we’ve already started?
No. Every CEO, I think, has started AI. It’s very rare that they haven’t done that. They may have started and are frustrated. They may not be wasting real money yet, but many are. And the ones who are just starting don’t want to waste a lot of money, time or resources either.
How long should it take to see a return on AI?
It’s hard to answer, because quite frankly, it depends on the growth constraint and the business model. It could take months to see results. It could take weeks. You need to know when you expect to see results, have those be realistic, understand the cost and really calculate the ROI. What’s the business case for investing in this AI initiative? When do we expect it to generate positive contribution margin? When do we expect it to break even?
What should you do next?
If you showed me your current AI project, I’d ask you a couple of things:
- What value do you see created from this, and by when? For example, “I expect to see $500,000 more in cash on hand,” or “I expect to see 20% growth in revenue starting in the second quarter.” If you can’t articulate the economic value and by when, I’d question whether it’s defined well enough.
- Is this your number one growth constraint? That means revenue growth, improvement in margin or profitability, or improvement in cash.
If it doesn’t check one of those boxes, then we have a real problem.
There’s a lot of AI noise in the marketplace. Everyone’s trying to do it and trying to leverage it, with different levels of success. If you want more comfort on how to get a hold of AI and point it at your top growth constraint, book a 30-minute call with me.
About the author
Bill Goodwin is a Certified Scaling Up Coach and founder and CEO of Growth Optimized, where he works with CEOs and founders of $5M to $100M companies across the country on strategy, execution and growth. He also leads Complete CEO, an in-person peer group for CEOs in Phoenix and Scottsdale. Before coaching, he was CEO of MeMD, which he scaled and sold to Walmart at 9X revenue.
Sources
- McKinsey, The state of AI in 2026: On the road to ROI, August 2026, as reported by Fortune, September 2, 2026
- PwC, 29th Global CEO Survey, January 2026
- MIT Project NANDA, The GenAI Divide: State of AI in Business 2025, methodology as summarized by eCommerce Fastlane and Marketing AI Institute
