We sat down with Pooneh to talk about why so many AI initiatives struggle to deliver ROI, why change management needs to begin on day one, and how leaders can bring people on the journey as AI transforms the way work is done. 

Why do you think so many AI initiatives are failing to deliver ROI? 

There are several reasons, but the most obvious one to me is that organizations rarely think through the people side of change in enough depth. What will the roles be? Where are the handoffs? How do humans and agents actually interact day to day? Those questions tend to get answered after deployment, not before. They roll out tools and technologies before thinking through what the organization will look like on the other side — for example, what agents will handle versus what remains with humans.

The mindset I see in many organizations is: We need to compete. We’re behind. So, they keep pushing tools into the organization, saying, “Go play and experiment with these tools. Tell us where you’re going to gain productivity,” without first having drawn a clear picture of the desired future state. 

Also, transparency matters just as much. When people understand what’s coming and why, apprehension about change can turn into curiosity about the possibilities — and, eventually, a desire to help build the future state rather than simply be pushed into it. 

Another challenge is that organizations lose sight of the business outcomes they set out to achieve, and the anticipated ROI never materializes. The lens has to be transformation, not a lift-and-shift of existing processes into a new technology. That means optimizing where agents are deployed — and transforming the rest of the organization around them. 

Historically, we approached this as process optimization and end-to-end re-engineering, viewed through a business lens. In the age of agentic AI, that framing isn’t enough. The work is no longer about making an existing process more efficient — it’s about reimagining what the process becomes when agents and humans work together.

When should change management enter an AI initiative? 

Change management belongs alongside the program leads from the very beginning — from the initial framing of the business case, not after the solution has been designed. 

Start with people, processes, and technology. Look at the functions where you plan to deploy AI and work through the people angle deliberately. You won’t have a perfect picture of where everything lands — no one does at that stage. But you can still bring people along and say, “Here’s where we’re trying to get to.” That’s the work: drawing the picture, however provisional, and communicating it widely and consistently. Building awareness and desire across an organization requires the change lead, the program leads, and the change owners sitting at the same table from day one — not the change lead being handed a communications plan at the end. 

How do you build trust and excitement around AI for employees? 

Transparency and honesty are the way to go. No one has a crystal ball and so no one can say, “This is exactly what the future is going to look like.” But the communication has to be something along the lines of: “Look, we know that some of the tasks within your responsibilities are going to be replaced by agentic tools, by agents. But that is going to help make you more productive so that you can do more interesting and higher value work.” 

That message only lands if the second half is true, which is why the human side has to be designed as deliberately as the technology. We need humans to do more complex work and to have domain knowledge, because AI can never replace human experience and judgement – it augments it. Agents can produce a confident answer that is simply wrong, and only someone with real experience can pressure-test whether the output actually holds up in context. Judgment isn’t replaced by AI; it’s what makes AI usable. 

Human skills like emotional intelligence, critical thinking and creativity remain essential. AI cannot replace the importance of empathy and context in communicating and training – knowing what a particular team needs to hear, and how. No agent does that. When employees can see where their own value sits in the future state, trust follows, and excitement usually follows trust. 

One of the promises of AI is higher-value work. What does that actually look like? 

There’s no single answer — it depends heavily on the function. But there’s a common thread: domain knowledge, end-to-end fluency in how the work actually gets done. Whether you sit in HR, operations, or technology, understanding your process front to back is what lets you see where an agent belongs, where it doesn’t, and what breaks if you get that wrong. 

I’d focus on skills rather than where a role sits on an org chart, because the organizational model itself is going to change. This isn’t a clean split where humans do X and agents do Y. It’s a question of how you use these tools to do your own job better and faster — and then what you do with the capacity that frees up. 

That last part is where the real value is, and it’s the part organizations most often leave on the table. If you use AI to do the same work with fewer people and stop there, you’ve captured a cost saving. The higher-value work is the set of problems, customer needs, and products you never had the bandwidth to get to before. 

Are organizations measuring AI adoption the right way? 

In a recent agentic AI course at Harvard, one point stayed with me: adoption is a dangerous measure. Consider what it usually means in practice — did people log in, did they click the app, did usage go up quarter over quarter? Every organization is eager to report that its people are using the tools. 

But usage and adoption are not the same thing. A high login count tells you people opened something. It tells you nothing about whether the work got better, or whether anything changed in the outcome the business actually cares about. Those are two very different questions, and most dashboards only answer the first one. 

The adoption worth measuring is when AI becomes part of how someone works. Not “I tried the tool this month,” but, “I use it to scan my inbox each morning and it tells me what needs my attention, and I’d notice immediately if it was gone.” That’s the threshold — when the tool is embedded in the daily routine rather than visited. It’s harder to measure than logins, which is precisely why so few organizations do it. 

AI is moving very quickly. Does that require a new change management playbook? 

The core elements hold. What breaks is the shape of the cycle. The traditional sequence assumes an endpoint — you build awareness, roll out, go live, sustain, and monitor. That works when the thing you rolled out is still the same thing a year later. With agentic tools, it isn’t. What one of these tools could do three months ago and what it can do today are meaningfully different capabilities. 

So, the model becomes a rolling change management framework. You define a wave: here’s what we’re trying to achieve, here’s what the tool can do today, the business outcomes it can deliver and here’s what changes for the people doing the work. You deliver it. Then you come back and ask a second set of questions — what can the tool do now that it couldn’t when we scoped wave one, and what does that make possible in terms of additional business value? And what that does look like for the people of side of change. 

The hard part is that sustainment, in the traditional sense, never quite arrives. You’re stabilizing one wave while scoping the next, and the people in the middle are absorbing both at once. That’s the real discipline: sequencing the waves so the organization can keep pace, rather than treating each new capability as a reason to restart. The tools will keep getting better on a very short cycle. The change function has to be built to move at that cadence without exhausting the people it’s meant to support. 

If an AI initiative is struggling, how should leaders reset it? 

What I am seeing is that some organizations are favoring the risk of getting it wrong against the risk of holding off and waiting for the organization to be ready. So, they deploy with little change planning, because waiting until the organization is ready feels like falling behind. 

The pattern that follows is fairly consistent. Experimentation and rollout proceed, adoption comes in low, and the returns stay stubbornly distant. Meanwhile consumption costs climb. At some point someone asks the uncomfortable question — we’ve spent this, what did we actually get? — and budgets get cut before the initiative ever had a fair test. 

So, the reset means going back to the beginning and rebuilding the strategy and operating model with people treated as an integral part of the future-state business case, not an afterthought to it. Done well, a reset isn’t an admission of failure. It’s the wave-one retrospective that should have been on the calendar from the start. 

Why does change management still come in so late, and where should it sit in the organization? 

Because it’s widely misunderstood at the leadership level. Change management is still treated as the communications workstream — “the change is coming, here’s the announcement, here’s the training deck.” That framing puts it at the end of the process by definition. If your role is to tell people about a decision, you can’t be in the room where the decision gets made. 

The irony is that the evidence has been consistent for years. Prosci’s benchmarking research finds that initiatives with excellent change management are seven times more likely to meet their objectives than those with poor change management — and simply moving from poor to fair roughly triples the odds. Few disciplines can point to that kind of correlation, and yet it remains among the first line items cut when budgets tighten. 

On structure: a central function alone isn’t enough, and neither is a federated model. Purely central teams end up too far from the work to know what’s actually changing on the ground. Purely embedded ones drift — different methodologies, different standards, inconsistent results, and no way to compare one initiative to another. The model that works is a hub and spoke: a central team that owns the principles, methodology, tooling, and sponsorship model, with experienced change practitioners embedded in each line of business, close enough to the work to see what’s really happening. 

Could I see a change leader in the C-suite? Yes — and it’s already happening. Chief Transformation Officers and Chief Change Officers have moved from novelty to a recognized appointment; driven by the same forces we’re discussing here. The point isn’t the title. It’s that change management should be a genuine partner in defining the future-state operating model, with full standing on the talent dimensions of that design, rather than a service function brought in to explain a decision after it’s been made. 

What is one issue around AI that leaders are not focusing on enough right now? 

Governance & regulatory oversight— how we deploy Agentic AI and how we control its use.  

These tools can hallucinate. They can reach deep into your systems. And they can be very effective at working around rules and ungoverned processes, which is dangerous in any regulated environment.   

Regulators in Europe are already probing this closely. The EU AI Act, which entered into force on August 1, 2024, imposes a level of control and governance we don’t yet see in the US. European Works Councils also play a prominent role in overseeing the use of AI, given the impact on people and the potential for job losses — and there is no direct equivalent in the US. Organizations need to focus far more on proper governance and controls in the near term.  

Over the next couple of years, governance and the ethical use of AI will matter a great deal more than they do today. 

What is a piece of advice you would give leaders today? 

Through all of it, people still have to be brought along. The pace is fast; the targets are ambitious, and the impact on organizations — and on the people inside them — will be enormous. So, the question every leader should be asking isn’t only what the technology can do. It’s whether we are giving our people what they need to meet it: clarity about where we’re going, the skills to get there, and a genuine place in what comes next. That doesn’t happen on its own. It has to be built deliberately. Technology will keep advancing on its own timeline. People need us to walk the journey with them.