June 1, 2026
We’re talking about agentic AI a lot more than we’re doing it. In regulated industries like healthcare and finance, risk aversion is logical, but staying still is also a risk decision. True execution requires building measurement and attribution infrastructure before building agents.
Gregory Ng | EVP, Decision Science
There is a particular kind of organizational paralysis that looks like strategy.
Meetings get scheduled. Frameworks get built. Pilots get announced in all-hands presentations with slide decks that use words like "phased" and "thoughtful" and "responsible."
And then not much happens.
That's where regulated industries are right now with agentic AI. That’s the honest version of what seems to be happening these days.
Bottom line: we’re talking about agentic AI a lot more than we’re doing it.
In healthcare and financial services, the hesitation around agentic AI is not irrational. It is, in fact, logical given the environment these organizations operate in.
Agentic AI systems don't just generate content or surface recommendations. They take action.
They make sequential decisions with downstream consequences. They interact with customers and patients in ways that affect financial outcomes, health decisions, and institutional trust.
In industries where a single compliance failure triggers regulatory scrutiny and a single bad customer experience ends a relationship, the instinct to move carefully is not timidity. It is institutional self-preservation.
No one wants an AI agent sending the wrong communication to a patient navigating a serious diagnosis. No one wants an agent making an incorrect determination in a loan workflow and leaving a paper trail that resembles discrimination.
The stakes are asymmetric. The upside of moving fast is competitive positioning. The downside of moving without discipline is existential.
So organizations do what organizations do when the risk calculus feels uneven.
They wait for more certainty. They benchmark against peers who are also waiting. They commission research, attend conferences, and build internal committees.
Meanwhile, the gap between where they are and where the technology is going widens in the background.
What doesn't get said in those same conversations: inaction is a risk decision, too. It feels safer because the consequences are slower and harder to attribute. But, customers and patients are not waiting.
They are using AI agents in other parts of their lives right now. They use them to book travel, manage personal finances, navigate insurance questions, research medical symptoms, and coordinate complex logistics. Their expectations about what a responsive, intelligent, personalized experience looks like are being shaped by those interactions every day.
When they encounter your organization and find something slower, more generic, and less intuitive by comparison, the gap registers. They may not articulate it that way.
Over time that gap becomes a retention problem, an acquisition problem, and a loyalty problem that shows up in the numbers long after the window to address it has closed.
The CMO or CDO who waits for perfect conditions is not avoiding risk. They are trading one kind of risk for another.
The question is not whether to move. The question is how to move in a way that is defensible, measurable, and designed to build confidence rather than erode it.
The agentic AI conversation in regulated industries gets framed as a technology problem or a risk problem. Those things are real. Integration complexity is real. Governance frameworks are still developing. Vendor capabilities vary and the definition of what "agentic" means is still being standardized across the industry.
Underneath all of that is a more fundamental problem that doesn't get enough attention: Most organizations don't have the attribution infrastructure to know whether their AI investments are working.
Without clear attribution, every agentic deployment is a leap of faith.
You launch something, watch some metrics move, and try to connect the dots back to business outcomes in ways that are more narrative than rigorous. When outcomes are ambiguous, the internal case for the next investment is weak. When the internal case is weak, momentum stalls.
In a regulated industry, stalled momentum tends to stay stalled.
Attribution infrastructure is not a glamorous investment. It doesn't generate keynotes or press releases.
The crawl-walk-run framework gets used so often in AI conversations that it has lost meaning. It has become shorthand for "we are being careful" without specifying what careful looks like or how you know when you are ready to move to the next stage.
The crawl stage is about building measurement before building capability.
The walk stage is where you deploy in constrained, visible contexts with human review built into the workflow.
The run stage is not a destination. It is a posture.
The framework only works when each stage is tied to real business outcomes with real attribution calculations. Without that, crawl-walk-run is just a metaphor for going slowly.
Slow for its own sake doesn't build anything.
The organizations that lead on agentic AI in regulated industries will not necessarily be the ones moving fastest right now. They will be the ones building the infrastructure that lets them move with confidence later.
That starts with data foundations. Agentic AI is only as good as the data it acts on.
Organizations with clean, connected, governed data architectures will have a structural advantage. Those without that foundation will discover the problem at the worst possible time.
The conversation around agentic AI in regulated industries is not slowing down. The question is whether your organization is building toward something or participating in the conversation.
Those are two different things. And the distance between them is growing.
