Forgetting change management undermines AI ROI

One statistic jumps out in the MIT Media Labs report on the state of generative AI: 95% of enterprise AI pilots never reach full production. However, that headline obscures a more revealing fact: 'unwillingness to adopt new tools' ranks highest at nine out of 10 when respondents rate barriers to AI success, and 'challenging change management' follows at roughly six out of 10. Years ago I crafted the equation: change management = expectations + accountabilities. The current imbalance in that math brought on by the hype around AI has introduced a familiar dynamic for anyone with a background in change management: employees, afraid for their jobs, are reluctant to engage in new processes and new technology as a means of job protection, companies unclear on expectations are challenged to communicate clearly, and uncertainty in the job market creates a perfect storm. With AI, executives are caught in a bind. They can't answer questions about how AI will impact jobs because they don't know what the future holds. Historically in major transformations, executives are able to provide some early guidance. They can tell employees what skills the company will need and retrain employees to adapt to new work. AI is evolving so rapidly that it is very hard to guess just where it will be within months, let alone years. As a result, mapping a clear AI relevant change and talent strategy is difficult. Still, paralysis isn't a foregone conclusion in the face of this uncertainty. Quite the opposite. Acknowledgement of uncertainty may form the seeds of a new approach. There's no crystal ball If the current state of AI gives executives little ability to speak clearly about what the future holds, what is actually happening in companies? Deloitte's 2026 'State of AI in the Enterprise' report provides some answers. While executives see insufficient skill as the biggest barrier to integrating AI into the business, less than half are making significant changes to their talent management strategies. On that front, most commonly, about half are educating their workforce on AI fluency. At this stage, AI is largely a complement to existing worker skills, as 84% of companies have not (yet) redesigned jobs around AI capabilities. Executives don't know what can be automated, and therefore, they don't know what skills will be needed (or not) in the future. That situation provides little comfort for employees, and it might be a reason why a survey from the Pew Research Center last year found a huge disparity in enthusiasm for AI. Among experts, 56% believed AI will have a positive impact over the next 20 years. Among the general public, that share fell to just 17%. For starters, automate the boring stuff For most enterprises, the aim is not merely to bolt AI onto existing roles, but to change how work is done. Consider procurement, where managers once compared vendors and negotiated contracts based on experience. Given the current state of AI adoption at many companies, that manager is probably using an AI tool to draft emails more quickly. Executives want something more consequential: an automated system that weighs cost, risk, and reliability – with better outcomes. That does not mean the procurement role disappears. It means the role changes. Someone must still evaluate whether the model is wrong, if, for example, a geopolitical shock alters the risk calculus faster than the system can grasp. So, how can executives get there? Automation doesn't replicate human judgment. It targets the boring stuff: organizational basics like paying invoices, order-to-cash, and procurement. These processes are repeatable sequences of steps, decisions, and handoffs that keep an enterprise running (emphasis on repeatable). And therein lies the challenge. Reality is a messy accumulation of exceptions, workarounds, and hidden steps. Official process documentation rarely reflects how work actually happens. Automation strategies often stall because of this reality gap. A digital twin of your organization, built with process intelligence, can provide the roadmap organizations need to build AI and clearly manage expectations and accountabilities. Without a map, you're just wandering Process intelligence uses system data to provide a real-time, end-to-end view of how work actually happens, and has become a key tool for companies to understand how work flows across their organization. When used to build a digital twin, it identifies bottlenecks and inefficiencies in how work is truly done, not as it is hoped in outdated organizational flowcharts. In recent years, it has evolved. Now showing the best opportunities for automation with context of what requires a human touch – the boring stuff, and the exceptions. This visibility also impacts talent strategies. When organizations have an idea of what can be automated, they have a better idea about how jobs will change. That gives a window into what talent is needed in the future. That information can be used to plan and intentionally signal to employees the skills that will be in demand. Workers are desperate for this guidance. The World Economic Forum found in 2025 that 50% of workers have completed some degree of reskilling despite the widespread perception of skill instability. Perpetual transformation AI will get more capable – but that doesn't mean humans step out of the equation. In mission-critical processes, the objective isn't to remove human involvement, it's to elevate it. A human in the loop ensures accountability, context, and judgment where it matters most. Preparing for that shift means embedding change into day-to-day operations. AI adoption isn't a linear project with a clear start and end; it is reshaping organizations from all directions. AI introduces constant disruption – processes, roles, and decisions are continually being redefined. Likewise, it highlights a demand for a perpetual approach to change management, one in which transformation is part of the culture. A strong foundation, grounded in practical 'boring' AI, will give organizations the stability to adapt, without losing control. We are asking our employees to maintain a high tolerance for ambiguity, it is our responsibility to work smartly to provide context and guidance for the future ahead where we can.