You won’t find a lottery with a 5% chance of winning. The odds of winning a lottery are typically in the one-in-a-million range, but the prize can be millions of dollars. Implementing AI seems to be a similar game. Millions in winnings are beckoning, but the odds are significantly better: 5%.
However, in a business context, a 5% chance of success is a death sentence for any initiative. Unless, of course, an organisation is forced into it from the outside—but no one is forcing AI projects, at least not for now.
A 2025 MIT study conducted a thorough empirical analysis of real-world projects and initiatives, which included a review of over 300 initiatives, structured interviews with 52 organisations, and surveys with 153 senior leaders (MIT, p. 2). The results are devastating: only 5% were successful, while 95% were not (MIT, p. 3), and billions were sunk without any measurable economic benefit (MIT, p. 3).
However, the study also reveals the differences between successful and unsuccessful projects. The winners ultimately implemented adaptive solutions that were customised to the needs of users in the business context (MIT, p. 3).
Understanding the following five myths is the first step to significantly increasing your own odds of winning.
5 Myths Sabotaging AI Integration Success
Myth 1: “The current AI just isn’t reliable enough for our teams’ practical needs.”
The Reality: This thought is tempting because it shifts responsibility onto the technology. But the facts tell a different story. If the technology itself were the problem, there would be no winners. Yet, the MIT study clearly shows that 5% of companies are exceedingly successful (MIT, p. 3). It’s like in school: even if the exam was extremely difficult, one or two students still got a perfect score.
The problem isn’t the maturity of the technology, but a fundamental misunderstanding of what AI actually is and where its strengths lie (MIT, p. 3). At its core, every AI, no matter how complex, is purely an information processing instrument. Its only job is to take in data and output it as processed information. It cannot control a machine, make a booking, or decide on its own—it can only prepare for these actions by supplying information to an actuator that performs the action. This leads to a compelling logical conclusion: the only sensible place to apply AI in a functioning business workflow is at the very beginning—at the information supply stage. The success of the 5% is not due to having a better AI, but because they use AI correctly: not as a universal task-doer, but as a highly specialised information supplier that strengthens their workflows at the most critical point.
Myth 2: “We need a flashy ‘lighthouse’ use case to win over our employees.”
The Reality: We definitely need successes to showcase. The error, however, lies in the assumption that we need to convince employees of AI’s power in the first place. That’s no longer necessary. According to the MIT study, workers from over 90% of surveyed companies report regular use of personal AI tools for work tasks (MIT, p. 8). What they need to be convinced of is something much more important: that AI can also work reliably in their specific business context to make their professional lives easier. In most companies, this proof has yet to materialise. A staged lighthouse project in some distant department cannot build trust, but a small, perfectly integrated solution that solves a genuine, widely-felt pain point in the daily workflow can.
Myth 3: “We must hire top-tier AI experts to succeed.”
The Reality: Of course, AI expertise is necessary. But experts in the company’s own needs and workflows are even more important. The motto is: company first, AI second. An AI model that doesn’t understand the subtleties of your internal workflows is worthless. What companies really need are partners who understand both: the business mechanics and the AI technology. According to the MIT study, the most successful projects often arise from collaborations with external partners who, by necessity, start with no knowledge of the company’s specific context and are therefore forced to listen and immerse themselves deeply in the client’s workflows (MIT, p. 19). This also opens up enormous opportunities for the internal IT department: it can transform from a general support unit into a strategic partner for business units, building the bridge to the external AI world.
Myth 4: “We have to ‘bring our employees along’ for the AI journey.”
The Reality: This phrase sounds like good management, but it implies an outdated, top-down approach from the change management textbooks. It suggests that management knows the way and just needs to get the workforce on board. With AI, this leads straight to failure. Anyone who tries to force a generic AI solution into a finely tuned clockwork of workflows like a new part risks damaging the entire process. The reality is: your employees are already on their way. In the “shadow AI economy” uncovered by the MIT study, they are already using privately funded tools that help them at work (MIT, p. 8). Management’s role isn’t to drag employees along on a top-down AI journey. It’s to champion the bottom-up initiatives that are already happening in the shadows. This means discovering which tools teams are actually using, understanding the mechanics of their workflows and their own learnings, and channelling that grassroots energy into a cohesive, company-wide strategy that’s built to win—not to gamble.
Myth 5: “We must build an in-house AI solution to succeed.”
The Reality: Behind this myth lies the correct idea that a solution must precisely match internal conditions. However, the conclusion that this is only possible through in-house development is wrong. The findings of the MIT study show the exact opposite: internally built solutions fail twice as often as those from external partners (MIT, p. 19).
The key is not who builds the solution, but how it is built. The successful 5% have focused on making the AI solution adaptive—shaping it into a learning companion that learns from feedback and continuously adapts to workflows, instead of forcing employees to adapt to the AI system’s needs (MIT, p. 14).
Conclusion
AI looks like software, feels like software, is implemented like software, and appears to work like software. But it is far more than mere software—it is a game-changer for information management at the start of any workflow. The companies that are successful with AI integration have understood and proven this.
Those who manage AI like software will inherit its worst flaws: scrambled processes, angry workers, and limited returns—or even downturns.
Those who integrate AI into the information core of their workflows as a learning partner will secure the future—and reverse the odds of the AI integration game.
About the author: Silvio Gerlach (https://silvio-gerlach.com/) is an AI integration explorer, trainer, and coach, supporting companies in embedding AI into their workflows. His upcoming book, “Leverage, Not Lottery! – How to Make AI Work for Business”, will be published soon.