AI’s Hidden Change Management Debt
Adapted from The Digital Project Manager podcast, hosted by Galen Low. Listen here.
Large enterprises rolling out AI often treat it as a technology problem first and an organizational one second. That order of operations is a mistake, according to Deborah Ketai, organizational change manager and board member at the Project Management Institute (PMI).
Ketai spent years leading AI-related change initiatives inside United Health Group, a Fortune 5 healthcare company, and she said the companies that skip the work of breaking down silos between teams pay for it later. “If they skip the step of breaking down silos, they’re also going to incur a lot of what we call debt, technical debt, knowledge debt, and so on,” Ketai said. “That is gonna make future projects more expensive and more difficult.”
Continue Reading for Free
Create a free account to finish this article, plus get ongoing access to timely insights and practical resources.
Have an account? Log In
If they skip the step of breaking down silos, they’re also going to incur a lot of what we call debt – technical debt, knowledge debt, and so on. That is gonna make future projects more expensive and more difficult.
That debt shows up in duplicated effort, competing tools and teams that don’t know what the group down the hall is already building.
Three Goals for AI Alignment at Scale
To address that, Ketai built her program around three goals: talent mobility and retention across the AI workforce, cross-team collaboration, and a broader cultural shift toward considering AI as a legitimate business solution. “We [at United Health Group] really had three goals — talent, mobility and retention across the AI space,” Ketai said. The second goal addressed a common failure mode in large organizations, where pockets of AI work spring up independently and duplicate each other’s efforts. “My second goal for the program was to encourage collaboration between these pockets of AI, so that they’re not reinventing the wheel,” she said. The third goal was cultural: “The third is a cultural shift, encouraging business champions to consider AI solutions to their business challenges.”
Achieving that meant building programming — conferences, hackathons, internal content and community channels — aimed at different stakeholder segments across the company, rather than a single top-down mandate.
Measuring Change When the Data Is a Mess
Even with clear goals, proving the program’s impact turned out to be one of the hardest parts of the job. Ketai’s company had grown largely through mergers and acquisitions, which meant the underlying data she needed to track career mobility and engagement was inconsistent at best. “I will acknowledge that the metrics were the most difficult part of the program for me,” Ketai said, “because this was a Fortune 5 company that had grown largely through mergers and acquisitions, the relics of old HR systems still existed in the fact that the same type of role often had 15 different titles, job titles or job codes across the organization, and that made life very difficult.”
Without clean systems to draw on, Ketai’s team leaned on proxy metrics such as content engagement and event attendance rather than the deeper workforce data they would have preferred.
Should Project Managers Also Be Change Managers?
As AI initiatives multiply, more project managers are being asked to take on change management responsibilities as well. Ketai said the two skill sets overlap but aren’t identical, and the difference matters most in the timeline. “Not all projects require change management,” she said. “The skillsets are different but overlapping.” The real distinction, according to Ketai, comes down to who sticks around to see a change through. “The reason why you should have separate roles is the timelines, because traditionally, project managers are not given the opportunity to follow through and really make sure that a change gets sustained.”
Not all projects require change management. The skillsets are different, but overlapping.
That gap — between finishing a project and sustaining the change it was meant to produce — is, in her view, the strongest argument for keeping the two roles distinct, even as the boundary between them continues to blur in practice.
AI Isn’t Just a Tool — It’s a Stakeholder
Ketai also pointed to a shift in how project and change managers need to think about AI itself. Rather than treating it purely as infrastructure, she argued it increasingly behaves like a stakeholder in its own right. “Increasingly with generative AI, AI is itself a stakeholder,” Ketai said. “It has its own goals, its own attitudes, and it will act differently and sometimes contrary to the way you want it to based on those internal realities.”
Increasingly with generative AI, AI is a stakeholder. It has it’s own goals, its own attitudes, and it will act differently and sometimes contrary to the way you want it to.
That framing has practical implications for governance: If AI systems can behave unpredictably or diverge from expectations, they need to be accounted for in planning and risk conversations the same way a human stakeholder would be.
Where the Program Manager Role Is Headed
Looking ahead, Ketai said program managers will need to move closer to where business strategy is actually formed, not just execute against it. “Program managers need to get closer to where strategy is formed and be able to influence strategy,” she said, “and to be transparent in reflecting back to leadership where there are likely to be misalignments.”
Program managers need to get closer to where strategy is formed and be able to influence strategy.
That shift also means expanding how program managers think about risk, to include upside opportunity alongside the traditional focus on threats, and investing more deliberately in relationship-building across the organization over time.
The Trust Problem Nobody’s Solved Yet
Despite the momentum behind AI adoption, Ketai said the biggest remaining barrier isn’t the technology itself. It’s trust, particularly around data. She recounted a recent PMI Regional Leadership Conference session that debated whether it was appropriate for a PMI chapter to use AI to draft member email blasts. “It was not ethical if it meant releasing your members’ data,” Ketai said, “unless you were using some kind of completely internal tool.”
That distinction — between AI tools that keep data contained and those that don’t — is likely to shape how organizations approach AI governance long after the current wave of experimentation settles down.
Want more insights like these? Sign up for a free DPM account to hear from more experts like these.
