The Importance of Data Interoperability Before Implementing AI for Building Permit Review

• Data interoperability is crucial to implementing AI in government. One tech executive at an industry summit said a state’s five-year AI roadmap should spend the first phase on data interoperability.

• County IT leaders described how data fragmentation can make it harder to finalize an AI policy and how automated processes running with “dirty data” just generate more dirty data.

• Permit review involves parcel records, zoning, land use history, prior permits, and code, and AI tools for permitting deliver most when all that data is connected, current, and searchable as a single entity.

At the recent 2026 Los Angeles Digital Government Summit (LADGS), it seemed that all roads lead to AI. Many of the day’s sessions focused on what AI has done, can do, or will do for governments. However, another common theme arose: The journey toward successful AI use begins with data interoperability. If a municipality’s various sources and systems of data don’t play nice with a particular AI technology, that tool can’t perform to its full potential.

Whether it was the futurist public intellectual delivering the summit’s keynote address, a county CIO, or a vendor executive, three unrelated presentations supported themes that backed the importance of yes, adopting AI, but getting your data in order first.

During the LADGS 2026 “Rethinking Enterprise Architecture for AI” panel, the moderator, LA County Assistant Auditor-Controller Majida Adnan posed a hypothetical scenario to Joshua Northcott, Co-founder and CTO, Hounder. She asked him if he were made California’s chief enterprise architect, how would he lay out state and local government’s AI roadmap for the next five years?

Northcott answered with a multi-stage process beginning with the most important part. “The first year wouldn’t even be about AI,” he says. “It would really be all about data interoperability, because the true magic behind AI applications is the data. You can’t AI-enable a system that can’t talk to the system right next to it.”

After establishing data interoperability among systems, Northcott prescribes a statewide unified identity layer shared across jurisdictions and departments. “Imagine having a state of California login that gives access to your MyDMV, your business registration—a unified security layer across all logins,” he says. “All networked together.”

Northcott says his third step is “absolutely crucial:” AI governance and procurement rules that apply very specific security standards and evaluations across the board. “Because if not,” he says, “who knows, departments will just start buying your own licenses, right?”

Not only is data interoperability required for good implementation of AI, but its placement as the first step is also ideal to avoid unpleasant consequences.

Two members of the LADGS 2026 panel “AI One Year Later: Lessons from Implementation” discussed their data governance challenges for implementing AI.

Lynn Fyhrlund, Chief Innovation Officer, San Bernardino County, described governing data and AI across a large county with 24,000 employees, 15 separate IT departments, and wildly divergent data-sensitivity circumstances. “We have a hospital, a health division—many different requirements,” he says, “so we’re struggling with what we can use AI with. We try to have an overarching governance, but yet have enough autonomy within the department to be able to do what we need to do.”

Departmental fragmentation contributed to the long timeline of putting a San Bernardino County-wide AI policy in place. “It’s taken us a year and a half just because of all the departments,” Fyhrlund says. “It’s been like a soccer ball going back and forth across the field. We’re about to shoot the goal… but we understand that we’re not going to have it perfect.”

An IT architect at the Metropolitan Water District of Southern California, Jon Houck, traced the problems with data interoperability back to business processes and staff mindset. Automating a process that runs on bad data doesn’t fix the data, but rather it intensifies the problem. “If you don’t look at the system’s processes that made that dirty data, you’re going to just end up with more dirty data,” he says.

Houck recommends pushing data governance as far upstream in an organization as possible. “Change the culture, change people’s mindset, increase AI and data literacy — kind of all at once, ideally,” he says. “And the floodgates are just going to open.”

While the LADGS 2026 keynote speaker, bestselling futurist author and AI inventor Jonathan Brill, did not speak directly about AI for permitting or building departments, part of his message indicated that the time is growing short for straightening out your data governance for AI. Brill spoke to the collapsing cost of building and experimenting with AI, to the point that individual employees can launch tools that used to require a specialist team and a budget line. “The cost of doing that innovation is going through the floor,” he says.

Brill intended that as good news, for the collective potential that it will grant to every work group. “Every single person in every single organization is about to get dramatically smarter as a result of AI,” he says.

However, if an organization is to follow the steps for data interoperability and implementing AI that Northcott described above, they ought to do it soon. When cheaper, more accessible AI is in everyone’s hands, ungoverned data can be a serious risk when a staffer using rogue AI uses whatever data they can reach, clean or not.

The multi-faceted types of data involved in permitting workflows make data interoperability paramount for building departments looking into AI for plan review. A single building permit plan review may incorporate parcel records, zoning designations, land use history, prior permits, inspection records, and of course all the applicable building codes and zoning bylaws. All this data may run through systems that were built decades apart by different vendors and for different departments. They likely were not built in order to work together.

Like other forms of AI, automated building permit review depends on the quality data for the best outcomes and needs to be able to read across that entire spread of data types. An AI model that can flag a code issue on page 3,000 of a complex plan set also needs to be able to determine reliably which version of the code applies to that parcel.

That’s why the data interoperability that Northcott laid out as a common data model deserves to be scoped out ahead of an AI permitting pilot. When parcel, zoning, and code data is actually connected, current, and able to be queried as one thing, building departments tend to get useful results from AI review. And it’s a better use of time than trying to sort out the data after the pilot begins. If an AI tool needs data to flow freely through systems it currently can’t, that’s not a reason to abandon the AI project. It’s a sign to start the project with data interoperability.