Why 95 Percent of AI Pilots Fail—And How to Avoid It Happening to You

The corporate world's artificial intelligence revolution has a secret: It's not going well. Executives are flocking to implement AI, but the vast majority of these initiatives collapse under their own weight. Marc Malott witnessed this firsthand, running into an invisible wall as he tried to lead an AI transformation at a consultancy. But there’s hope—studying the rare organizations that have successfully rewired their operating systems for the AI age, Marc uncovers the blueprint for sustainable transformation that doesn't just deliver ROI, but reimagines what's possible.—Kate Lee

Was this newsletter forwarded to you? Sign up to get it in your inbox.

Companies everywhere are racing to integrate the world-changing power of AI into their businesses—and 95 percent of them are failing.

Such were the headlines as a study out of MIT went viral late last month that seemed to puncture the exuberance around corporate AI adoption.

I spent the last two years spearheading an ambitious AI transformation program at a mid-sized consulting firm. I don’t claim to know better than the study’s authors. But from what I’ve seen, a 95 percent failure rate is believable. A McKinsey report from earlier this year is similarly bleak: More than 80 percent of executives surveyed said generative AI has not yet moved enterprise-level earnings in a tangible way. Within companies, the employees leading the AI charge often hate their jobs the most: A study by Upwork published last month found that the most productive AI users are twice as likely to quit, and most don't understand their own firm's AI strategy. Eighty-eight percent reported feeling burned out.

All of these datapoints are due to the same root cause. Companies that are investing in AI are so eager to make their money back that they are sacrificing long-term payoff.

I learned this the hard way. My team started tailoring our organization’s AI adoption plan in 2023—we launched targeted pilots, overhauled workflows, and thought deeply about how to manage the change. We unlocked 40,000 hours of human capacity. Clients were thrilled.

But when we started chasing a hard return on investment, progress stalled.

The good news is that when AI transformation stalls, it often follows a pattern. It can be hard to spot, but I’ve managed to tease out some common traits, and found a recipe for how anyone can identify the warning signs and set a course toward lasting change.

Bland Is the Most Powerful Voice AI in the World

In fact, chances are, you’ve already spoken to it without even noticing. Companies from Fortune 500’s to startups have been using Bland to build AI voice agents for sales, support, and much more. It can:

1. Answer phone, SMS, and web‑chat requests 24/7—no hold music, no queues
2. Scale to 1 million-plus simultaneous conversations while slashing support costs
3. Speak any language and connect to any backend CRM, ticketing, or data source
4. Stay on‑brand with guard‑railed reasoning that prevents hallucinations and off‑script replies

Hundreds of enterprises and high‑growth startups already trust Bland to boost CSAT, cut resolution times, and free up agents for higher‑value work.

Curious? Call the number in the image, or Every readers can get started here for free.

Want to sponsor Every? Click here.

Why success triggers failure

Last year, the 300-person consulting firm I was working at had strong momentum following our aggressive, early AI implementation. We implemented what has become a best-in-class tool for surfacing insights from research calls. It was like having a junior employee—it automated the preparation of transcripts and made client-ready summaries of our research. Teams were able to focus on the work that mattered—like overhauling critical project workflows and distilling actionable insights faster. We were delivering higher quality work and wowing clients.

We believed our success would speak for itself, validate AI’s potential, and accelerate progress across the company. Instead, progress stalled.

This is not a critique of my former company. It’s an illustration of the powerful systemic forces affecting nearly every legacy business trying to adapt to AI.

To reflect our shop’s newfound AI efficiency, we were told to charge clients a higher hourly rate. This would naturally entail fewer hours spent per project. But that made sense—we’d get the job done more quickly and better, and move on to the next project faster. It was an easy way to capture ROI from the hard implementation work we had done.

But the decision to begin harvesting ROI resulted in an unforeseen shift. After we raised performance targets and hourly prices, progress became exponentially more difficult. Slack for innovation vanished as everyone focused on hitting their new numbers.

The pressure to hit near-term targets created immense friction. The bar for approving any new expenditure was raised, slowing decisions and timelines to a crawl. The rollout of a key AI-enabled product expansion stalled for nearly a year because teams felt too overwhelmed to commit to more work.

These symptoms describe the classic problem in game theory known as the stag hunt. Applied to the age of AI, the "stag" is the massive, shared prize of forward momentum and compounding capabilities. On the other side is the "hare"—an immediate but temporary gain companies win by harvesting a productivity bump, or employees gain by hoarding small wins.

Anyone pursuing the stag without cooperation catches nothing. As a result, belief in continued cooperation is the deciding factor between outcomes. When companies harvest gains, they signal they are chasing the hare—shattering the trust required to ever hunt the stag collaboratively again.

This pattern is everywhere. A pioneer invests discretionary effort to learn, create, and share something that helps everyone (e.g., a new workflow, an automated process, a faster tool). In a truly forward-thinking company, this would become the starting point for rebuilding how work gets done. But in the vast majority of companies, the innovation gets absorbed without changing the underlying structure. They pocket the efficiency gain. Targets ratchet up. Slack vanishes. The pioneer gets boxed into the old system, judged by the old rules. Less time for innovation. No reward, no promotion. These pioneers are left with two rational choices: give up on the stag or leave.

Everyone else learns the lesson that leaning in is career-limiting. So they make their own rational choice. Harvard Business Review research shows that employees protect themselves by hiding AI use and thereby hoarding productivity gains—they hunt for hares.

A theory of leverage

After my experience, I started trying to figure out how to prevent this kind of breakdown from happening again.

That led me to Donella Meadows. Meadows was a formative figure in systems thinking, a MacArthur “genius” grant recipient, longtime Dartmouth professor, and lead author on the influential book The Limits to Growth.

In her seminal essay "Leverage Points," published in 1997, Meadows explained that creating lasting change requires finding the right places to intervene. The most obvious interventions are like trying to move a boulder by hand: They require immense effort for relatively little impact. Less obvious interventions are powerful levers, where a strategic push can have massive, lasting change.

Meadows warned that in novel systems, our intuitions systematically mislead us. We not only focus on weak leverage points, but when we do identify powerful ones, we reflexively push them in the wrong direction.

This describes the current failure pattern with AI. Overwhelming effort is spent on surface-level solutions and implementations. And when companies go deeper, they invariably push powerful levers backward: tightening control when they need to release it, extracting more when they need to reinvest, accelerating work when they need to create space for adaptation.

Leading companies do the opposite. They build healthy and adaptable long-term value creation systems, where ROI emerges as the natural byproduct. Here are three examples that show how companies have addressed high-leverage parts of their organization in ways that have led to lasting impact and sustainable ROI in the age of AI.

SharkNinja: Trusting in decentralized decision-making

In the old world of predictable business, critiquing a flawed plan was risky. Pointing it out often got you labeled as "not a team player," so the rational move was to keep quiet and not rock the boat, leading to collective dysfunction.

Fast-growing household products maker SharkNinja is a perfect example of a company building an agile, high-trust operating system. Its CEO, Mark Barrocas, attacked the rigid old paradigm head-on. After realizing there was no playbook for the kind of innovation SharkNinja needed to achieve, he put radical trust in his employees. At a company town hall, he said, "I made a change because previously I was being stupid and I’ve decided now to be un-stupid."

"Being stupid," he signaled, was stubbornly sticking to a plan after reality proved it wrong. Instead, he made it clear that every employee has the agency and the responsibility to help correct bad decisions. This reinforced decentralized decision-making and redefined failure as a data point for course-correction, thereby building an agile, high-trust operating system that invites cooperation. SharkNinja more than doubled year-over-year net income (105 percent) in its latest quarterly earnings, which drove its stock to an all-time high last month.

Johnson Hana: Moving from billable hours to client value

Johnson Hana is a non-traditional, Dublin-based law firm. Instead of billing by the hour, it provides corporate clients with on-demand legal professionals and managed services for a flat fee or project-based rate. Its mission is not to bill more hours, but to make the lives of legal professionals "better and happier."

By changing the goal from "maximizing hours" to "maximizing talent and client value," the company inverts its relationship with technology. AI is embedded into every aspect of delivery to enable work that is better and faster than humans or technology alone. The lawyer becomes “happier” by automating the soul-crushing drudgery of routine document review. They become "better" by focusing on what clients actually value: complex negotiation, creative problem solving, and sophisticated judgement.

In July, the AI-powered legal platform Eudia acquired Johnson Hana for $50 million, pairing its advanced AI with a firm that has already proven it can embrace and metabolize the technology. Eudia bought a high-trust, stag-hunting operating system with a strong client list, rather than build one from scratch.

Shopify: Making AI use a performance metric

In a low-trust system, employees rationally hide AI use, hoard productivity gains, or refuse to learn AI as a means of (paradoxically) asserting their own indispensability.

Shopify's leadership, driven by a conviction to build a truly "AI-native" company, chose to rewire this. In a now-famous memo, CEO Tobi Lütke decreed that hiring managers must prove AI cannot do a task before they get approval to hire a human. And reflexive, effective AI use is now a part of performance evaluations. The incentive is no longer to hide AI to protect your job, but to master it to create a new, more valuable one. It forces the organization to hire for complementary human skills that AI cannot easily replicate: creativity, strategic thinking, and innovation.

Shopify has engineered a system where the rational choice is to collaborate with technology to scale capability. They are making the stag hunt mandatory. As Lütke has said, the goal is to use AI to "accomplish 100 times more work," which frees up human talent to tackle previously impossible challenges.

What does transformation really look like?

Meadows’s diagnosis rings true because what is required for sustainable AI ROI is nothing short of an operating system change.

For the last century, most companies have operated like a centralized power grid: built for top-down control and linear predictability. The goal was often replicability to scale and achieve maximum efficiency.

AI, however, is not a more efficient fuel for this old grid. It is a new form of energy entirely. Much of its power is generated at the edges, by individual employees and teams experimenting and innovating. This is why leaders like SharkNinja are effectively calling out the old grid as obsolete, like Johnson Hana are replacing old meters of value consumption with modern ones that also measure creation, and like Shopify are incentivizing everyone to be a net-positive energy producer. These companies are building a new grid, one that rewrites the social contract of work to match where and how value is now created. They have recognized that trust between management and employees is existential.

While 95 percent of employees see AI’s potential, their biggest concern is they don’t believe their organizations will share the benefits. The result is a shadow workforce disconnected from the grid. Everyone is hunting the hare, often hiding the fact that they’re using AI at work. Without trust, no one risks cooperation.

The new grid requires significantly deeper trust that says "we will recognize and reward the value you create, even when we cannot easily predict or control it.” The firms that build this trust by changing their operating model will own the future—not through temporary efficiency gains, but by enabling and harnessing compounding innovation that competitors cannot replicate. Because sustainable ROI does not come from what AI can do. It comes from what people can do when they trust that everyone is on the same hunt, and they will share in the prize.

Marc Malott advises leaders transforming operating models to thrive in the AI era. Follow him on LinkedIn.

To read more essays like this, subscribe to Every, and follow us on X at @every and on LinkedIn.

We build AI tools for readers like you. Write brilliantly with Spiral. Organize files automatically with Sparkle. Deliver yourself from email with Cora.

We also do AI training, adoption, and innovation for companies. Work with us to bring AI into your organization.

Get paid for sharing Every with your friends. Join our referral program.