Amadally’s view is that many companies are still early in the AI maturity lifecycle. They know they have to do something, but they are sorting through hundreds or even thousands of possible use cases, trying to determine which ones will actually matter to the business. At the same time, the conversation around AI is getting more demanding: CFOs want to know when the investment will pay off, leaders are confronting questions about operating models and governance, and organizations are beginning to see that simply bolting AI onto existing workflows will not deliver the full value.

What emerges from the conversation is a picture of the next phase of AI adoption: more disciplined about ROI, more ambitious about redesigning work, more robust about governance, and increasingly focused on what it means to become an AI-first organization.

The Pilot Era Is Giving Way to ROI Discipline

Amadally says a good number of companies are still at the very early stages of AI, evaluating which use cases are really going to impact the business. The pressure is shifting from experimentation to return on investment.

“Initially it was, oh, AI is like this new cool thing that we’re going to play around with. I think that the CFO office is coming down quite hard; it’s costing a bunch of money, both in terms of infrastructure, people and tokens. When is this going to pay off?”

That changes how companies choose where to begin. In Amadally’s words, “you have to pick your spots.” Not all use cases will deliver the same ROI or deliver it in the same time frame. A project that takes two years can lose support long before it produces value.

“Time frame is a very critical thing here, because if something’s going to take two years to do, people lose interest along the way. I think it’s key to get a couple of quick wins out the door, lay the foundation for how things can transform, and then kind of build on the more complex use cases.”

He also cautions companies not to turn data readiness into a reason to wait indefinitely. Data quality remains an issue, but that problem predates AI – and AI can sometimes help improve the data itself.

AI Should Change the Workflow

For Amadally, one of the biggest mistakes is treating AI as a tool that can simply be attached to an existing process. Incremental automation has value, but the more consequential opportunity comes from stepping back and rethinking how the work should happen in the first place.

“You can’t just put band-aids on your workflow. I think for AI to be truly effective, you have to try and take a step back, look at how AI can transform it. You have to re-think the way you’re solving that problem and the workflow you have in place.”

He points to document intelligence as one practical area where the technology is already changing what is possible. GenAI can move beyond extracting information from documents to helping create credit memos, appraisals and risk evaluations, with a human still in the loop where judgment is required.

“I think the workflow of the future is going to be a combination of human and digital labor and figuring out what that mix is. It is finding the optimal mix to be able to scale effectively and bring the right efficiencies to reduce costs.”

The human role does not disappear. Amadally says the right level of human involvement depends partly on regulation and partly on the gravity of the decision. If the outcome has to be right before it is executed, a human still needs to be involved. For less mission-critical work, he expects organizations to push further toward autonomy – potentially even using AI as a second checkpoint on other AI systems. It is something he already sees happening in the market right now.

The Right Enterprise Model Sits Somewhere in the Middle

AI also reopens an old organizational debate: how much should be centralized and how much should sit with the business? Amadally has seen organizations swing in both directions. When AI and technology sit close to individual products, teams can move quickly – but they can also end up solving the same problem repeatedly in different parts of the company.

He is specific about the things that should remain connected at the enterprise level. Culture and risk need a common view, and he sees a strong argument for centralizing tooling so companies do not create an integration problem by proliferating too many platforms. Product development and innovation, however, need room to live closer to the business.

The ideal, in his view, is a feedback loop: individual businesses have enough autonomy to act, but what they learn comes back to the center and helps to continually refine the enterprise’s business model.

“That’s the ideal. You give your individual business lines enough autonomy to go and kind of do their business. And then as they learn, they feedback and then you refine your business model by the findings on what you’re seeing on the edges.”

Becoming AI-First Means Putting AI at the Front, Not the Back

The distinction Amadally draws between using AI and becoming AI-first is important. AI-first companies do not begin with the existing operating model and ask where AI might fit. They start with the assumption that AI will change how the organization operates.

That is easier for startups than mature companies with decades of infrastructure, processes and systems already in place. Amadally compares it to the difference between being cloud-first and becoming a late cloud adopter: starting without legacy infrastructure creates an inherent advantage.

“It’s much easier when you don’t have an infrastructure to go on the cloud, right? It’s the same dynamic, but even more so with AI.”

For established organizations, he sees a more practical path. A company may start a new function, enter a new geography or pursue a new opportunity as AI-first from the beginning, and then allow what it learns to spill back into existing operations.

Amadally says leaders are often a step or two behind where the technology is moving, which is understandable: they are still running a company or a function and dealing with today’s problems while being asked to rethink the business for tomorrow.

“I think leadership needs to move beyond the buzzword of AI and think, really think about what is going on around them? What is going on in the market?”

That same expectation extends to the workforce. Amadally does not believe everyone has to become an AI expert. He does believe the whole organization needs enough fluency to understand what AI can do and how it can change the work.

“Not everybody needs to be an expert on it. Everyone needs to know that it’s there, what it’s capable of, how it can impact the way they do their job.I don’t think it’s a one-off task either. This is like a real ongoing educational journey.”

Looking Ahead: Three Priorities for Leaders Make the ROI harder.

Make the ROI harder. 

Amadally says AI ROI has often been described as “soft ROI”, such as freeing up a percentage of someone’s time so it can be spent on something more productive. He is seeing CFOs push for a more concrete definition of value.

“There’s now a demand, more coming out of the CFO. The CFO is thinking “If my business is going to grow 50%, I don’t need 50% more people to do it, I need to be able to do it with the existing number of people, and utilize AI to be able to scale more effectively.”

Treat governance as an ongoing operating capability.

AI cannot simply be implemented and left alone. Models change, risk evolves and more autonomous systems can find ways around constraints that organizations did not anticipate.

” You have to make sure that you’ve got the right governance around it, you’ve got the right rail guards. You need to know you’re addressing model drift and bias as the model evolves over time.”

Amadally expects demand for AI governance and risk capabilities to grow significantly as organizations build, adopt and use the technology.

Build AI fluency across the business.

Specialized AI talent will remain important, but Amadally’s point is broader: becoming AI-first cannot be the responsibility of the data and AI team alone.

“It’s not just the data and AI team that needs to be fluent in this. I think it’s this whole organization.”

The Bottom Line

The next phase of AI is about choosing the right use cases, proving the economics, redesigning workflows, deciding what belongs at the center of the enterprise and what belongs at the edges, and building the leadership and workforce fluency to keep adapting as the technology changes.

Amadally’s central point is that the organizations that get the most from AI will be the ones willing to rethink how they operate, not just automate how they operate today.