How Healthcare Enterprise Architects Are Carrying AI Governance Council Decisions All The Way To Production

AI governance councils decide which AI projects an organization funds, reviewing each proposal for security, compliance, and business value. Once a use case is approved, it often stalls because no one owns the work of building it across infrastructure, security, data, and business systems. Enterprise architects already work across all of those systems, which makes them the team best placed to carry approved use cases into production.

Owning that work is a core part of Chetna Mehrotra’s role as Senior Enterprise Architect at Ascension, one of the largest nonprofit health systems in the U.S. She has spent more than 20 years in healthcare IT, including time as a solutions architect at Ascension and senior consulting roles with CommonSpirit Health and British Columbia’s Provincial Health Services Authority, where she worked closely with clinical teams and EHR systems. She sees the CIO as the right leader for an AI governance council and enterprise architecture as the team that should drive execution.

"Somebody has to lead from the front and make sure all the stakeholders are coming together. EA can be a great team to lead these efforts because it hits every tower," said Mehrotra. In enterprise IT, a tower is a major domain like infrastructure or security, and in a health system the clinical side counts as one too. That same view shapes who belongs on the council.

• Seats at the table: The CIO should lead an AI governance council, with the CISO and chief data officer alongside. Two other seats matter just as much. One goes to the business sponsor who brings the use case, and the other goes to the people who will use what the AI produces, since they’re the ones who will judge whether it helps. Security needs a strong voice, because it touches every step of how data moves through an organization. "Security is the foundation. Every aspect of the data movement will touch security," she said. With users and security both in the room, problems come up during review, before anyone starts building.

• Start to finish: In this model, enterprise architecture runs intake for AI use cases and stays with each one through delivery. "The enterprise architecture team becomes the gatekeeper of the use case. They take it to the governance council, they do the architecture and the design, and then they provide leadership and technical oversight to the implementation team," she added. That gives each use case a single point of accountability from first request to launch. Developers, testers, and project managers still handle day-to-day execution, with the architect keeping their work tied to what the council approved.

Some of an architect’s most important work happens early, while a request is still a business problem. The team asking for help usually knows what it wants to fix but not which systems or data the fix would touch. Turning that request into technical requirements is a translation job, and architects are trained for it. A use case that arrives in front of the council already written in technical terms is easier to approve and easier to staff.

• Move before it’s perfect: Once a use case has a technical spec, the team can begin building, even if parts of the design are still open. "I’m a big believer in starting something before figuring out the final perfect design. Once you start, you’ll see a path forward on its own, especially with a new technology like AI," said Mehrotra. Every organization is still learning what AI can do in its own environment, and many of those lessons only appear once work is underway. A running project surfaces them faster than a design review. The architect keeps the full picture in view during that phase and steers teams back when they drift off course.

• Aim before volume: Architects also decide how much data a use case needs, and the answer is usually less than teams expect. A healthcare use case might run on six to 10 targeted APIs covering patient, encounter, visit, and diagnosis data, with no need to pull every table in the organization. That data then moves through stages, starting raw, getting cleaned and harmonized, and ending in a version shaped for its end user. "I want to make sure my nurse sees what she wants to see, not my raw data or a bunch of tables and columns," she noted. Scoping data this way keeps projects smaller and cheaper to run, and the architect who designed the use case is the right person to make that call.

AI governance councils will field more requests as AI spreads through clinical and business operations, and each request will touch more systems before it goes live. Councils that keep pace will need someone who can explain what a use case solves and what it takes to build, in terms both executives and engineers understand. That skill sits at the core of enterprise architecture. "A CIO will always ask one question. Where do I put my money? What are you trying to solve?" said Mehrotra.

The views and opinions expressed are those of Chetna Mehrotra and do not represent the official policy or position of any organization.