SAN FRANCISCO, CALIFORNIA – SEPTEMBER 04: Anthropic Co-founder and CEO Dario Amodei speaks at the “How AI Will Transform Business in the Next 18 Months” panel during INBOUND 2025 Powered by HubSpot at Moscone Center on September 04, 2025 in San Francisco, California. (Photo by Chance Yeh/Getty Images for HubSpot)
Getty Images for HubSpot
While AI leaders like Dario Amodei are giving interviews that predict that AI will upend employment as we know it, the actual changes on the corporate level are happening at a much slower pace. Over the summer, a widely covered MIT report revealed that 95% of generative AI projects were failing at large organizations; more recently, an EY report found that while 88% of employees are using AI at work, the tasks are primarily limited to using it for basic things like search and summarization. In fact, the report found that companies are missing out on up to 40% of AI productivity gains and only 5% of employees are maximizing AI to transform their work.
Some of this can be attributed to gaps in the actual capabilities of the technology; as much as some companies would like to replace customer service teams with bots, those that have tried this have reversed course quickly. And lots of employees could benefit from training in how to use AI for deep research and automation. But beyond that, most companies are flailing for one simple reason – they don’t actually know what problems they are trying to solve. And until they are clear on that, as well as the reasons the problems exist and how to scale solutions quickly, they’ll continue to basically “perform” AI adoption without actually making any changes.
Let’s take a best case scenario for how doing something meaningful with AI could work. Company X starts digging into its biggest problems and determines that employees are spending hours doing administrative work that is outside their actual job description and zone of genius. Filing expenses, updating decks, tracking tickets…we’ve all had to do it, and it sucks the life and joy out of everyone. And it’s a problem because it takes away from the time an employee could be productive and contributing to the bottom line.
But just understanding that a problem exists isn’t enough; a company needs to understand why the problem exists. In a best case scenario like this one, it’s because, until recently, the tools to successfully automate most of this work haven’t been developed or widely available. In many other situations, the problem is structural, and no amount of AI can fix it. A company with a high turnover rate and employee morale issues can implement all the whizzy tech solutions it wants, but if the corporate culture is toxic and the pay is inferior, no amount of AI can undo that.
If the reason for the problem is that the solution was heretofore unavailable, then there is a real opportunity for change. The next step is to find vendors and put together a pilot class, with clear outcomes that must be met. For instance, a certain amount of hours must be redirected from admin work to productive work, and the tools must be easy enough to learn and use that people want to continue with them, and the cost must come in at a certain number. If all those conditions are met, then the firm needs to do the hardest part – scale rapidly.
“Scaling based on a pilot’s initial ROI is the classic ‘Surface Wave‘ trap,” says Dr. Markus Bernhardt, Principal at Endeavor Intelligence. “That early productivity surge never translates to business outcomes unless leaders recognize that organizational inertia is the real obstacle. Real transformation requires shifting the focus from tools to the ‘undercurrent,’ which means redesigning the system around the tool.”
“Across organizations, executives still feel pressure to prove that ‘AI-driven performance improvement’ delivers tangible results,” Bernhardt continues. “Multiple 2025 market reviews point to the same issue: the tools are strong, but outcomes stall. A new ‘State of AI’ report from McKinsey confirms this: while 88% of organizations report using AI, the majority are ‘still in the experimenting or piloting stages’ and have not yet reached the ‘scaling phase.”
The failure is not technical; it is operational. A 2025 Deloitte report on AI ROI explains that adoption depends on people: how cultural resistance is managed, how effectively employees adopt new tools and how workflows adapt. This requires more than just better prompts. An IBM report on 2025 adoption challenges further proves this, listing concerns about data accuracy or bias (45%) and iInsufficient proprietary data (42%) as the top two barriers, meaning the human system is failing to provide the context and credible data the AI needs to function.
This confirms the core finding: The performance plateau is not a technology failure, but an operational one. “The tools are excellent at building knowledge, but they are insufficient for building judgment, “ says Bernhardt. “Until leaders address this structural human and data challenge, the plateau will persist.”
The essential takeaway is that the executive mandate has shifted. Leaders must stop acting as technology approvers and start becoming system designers. This means viewing the successful pilot as a blueprint for operational change, not just a win for the technology team. Until that foundational shift happens, the performance plateau will persist.