When
But that leaves an important question: What about the 5% that succeed? Their lessons offer a roadmap for everyone else.
MIT's research highlights why startups often succeed where enterprises fail: They pick a single, tangible pain point and apply AI relentlessly until it creates value. In contrast, many enterprises try to "do AI everywhere" without solving a single problem well.
In our own pilots, a breakthrough came from focusing on something narrow, but painful: annual engagement letter renewals. This repetitive process, involving tens of thousands of documents, was reduced by a factor of 20 when AI was embedded throughout the workflow from the bottom up. That single success became a catalyst for wider adoption because it solved a real business problem employees cared about.
This mirrors what McKinsey tracked across industries in a report from last year. In its
Another challenge MIT flagged is the "learning gap": the mismatch between expectations of AI and the reality of how it delivers value. Organizations often overcomplicate adoption with heavy training frameworks that become outdated almost immediately.
We found progress by simplifying. Instead of teaching people prompt engineering like a software manual, we focused on helping professionals understand how AI "thinks," what it does well, what it doesn't do well and how to test it in context. Micro-learning, onboarding bots and feedback loops helped turn AI from an abstract idea into a daily tool.
The cultural effect of this shift is as important as the efficiency gains. Employees who once approached AI with skepticism began seeking it voluntarily when they saw what it could do for them in their workday. Trust was built through corporate mandates and protocols, yes but they quickly translated into visible, practical benefits.
Invest in what's next
MIT's study also underscored that purchased AI solutions succeed twice as often as homegrown ones (67% vs. 33%). Many early internal experiments became "science projects" that couldn't scale. The key is not reinventing the wheel but choosing tools that fit business needs.
In our case, we wanted consistency across tax, audit and consulting teams, so we prioritized platforms that reduced friction and created a common language. Some of the most transformative gains come from less glamorous back-office processes, where automating administrative and compliance-heavy tasks quickly increases efficiency. If done right, it frees people up to do higher-value work, the work they'd rather be doing.
The other unavoidable reality is disruption. AI will not fully replace professionals, but it will reshape tasks, roles and in some cases careers. Pretending otherwise undermines trust.
Be candid: Acknowledge what AI can automate, identify where human expertise remains indispensable, and prepare teams to adapt. That preparation requires moving beyond tool-specific training toward durable skills. We emphasize building "T-shaped consultants," professionals with deep subject-matter expertise and broad AI fluency who can collaborate across disciplines.
Netscape Navigator was once a leading browser in the early days of the internet; today it's a footnote. The same cycle will play out with today's AI platforms. The only safeguard is developing adaptable thinkers who can evolve with the tools.
Most importantly, it is still the early days and a 95% failure rate shouldn't discourage leaders. It should motivate them to rethink how they pilot, how they train, and how they invest. AI success doesn't come from chasing hype or spreading resources thin. It comes from choosing a clear problem, embedding AI into real workflows, and preparing people to grow with the technology.
