Federal agencies looking to scale artificial intelligence need to focus on more than technology. Mission alignment, governance, organizational structure, and workflow design will determine whether AI investments translate into lasting results, according to data leaders from the State Department and ServiceNow.
In a recent MeriTalk webinar, Paula Osborn, deputy chief data and AI officer at the State Department, and Patrick McGarry, federal chief data officer at ServiceNow, outlined how agencies can connect data investments to mission priorities, build governance into daily work, and scale AI without multiplying security and compliance risk.
At the State Department, that approach is already helping reduce repetitive work. The department’s Northstar tool, which uses AI to produce media summaries and translations, has cut a process that could take embassy employees four to six hours a day to less than 30 minutes, Osborn said.
“Most agencies treat data as a technical modernization problem when it’s really actually an organizational alignment problem,” McGarry said.
Mission needs should drive data strategy
Agencies that want to treat data as a strategic asset first need to connect their data strategies directly to mission outcomes.
A data strategy should clearly explain how it will advance the agency’s core mission. Otherwise, “it’s a technical roadmap and it’s not a strategy,” McGarry said.
Clear ownership and sustained investment are also critical, Osborn noted.
“We need actual … data owners. We need … leadership to put money towards data management and governance and data catalogs and all of that,” she said.
At State, bureau-level chief data officer positions were deliberately established as senior-level roles so those officials could participate in executive decision-making.
That mission focus also shapes how the department looks for AI opportunities.
While serving as a bureau-level chief data officer, Osborn initially asked employees what data they used and what additional data they wanted. Those questions produced limited answers. Asking employees about their pain points proved more useful.
That approach helped identify repetitive workflows such as media monitoring at U.S. embassies.
Northstar supports that work by searching for relevant media coverage, producing summaries, and translating content into English. The process is performed across more than 270 State Department posts worldwide, Osborn said.
“Before you build anything … go watch how decisions get made,” McGarry said.
The most useful intervention points are often where employees need more information, wait for data, or make decisions using incomplete or outdated information, he noted.
Governance has to move into the workflow
As agencies move data and AI deeper into mission operations, governance cannot remain a separate compliance exercise.
McGarry said agencies should build controls such as role-based access, automated data tagging, audit trails, and clearly identified data owners directly into workflows.
“Real governance is sort of embedded architecture,” he said. “It’s designed into how data flows, who can access what, how decisions are logged.”
At State, that governance work is being paired with a more federated operating model that keeps policy and oversight centralized while allowing posts to develop tools locally.
The department’s Funhouse data science sandbox has more than 700 users who can build tools for their posts. State has also launched a global scaling program to identify locally developed tools that could be expanded for broader use.
Another effort, the AI Champions Program, trains employees at individual posts who can then share AI knowledge and practices within their embassies.
Osborn said she sometimes refers to data governance as “AI-ready data” because that framing can help leadership connect foundational investments with the AI capabilities they want to deploy.
The broader challenge is turning pilots into sustained operating capabilities.
Unclear ownership, weak incentives, data literacy gaps, governance disconnected from daily work, and pilots that never scale can all signal deeper organizational problems, McGarry said.
“You don’t want to be AI ready, you want to be ready to be accountable for AI,” he said.
Agencies can move fast without cutting security
Speed and security do not have to be competing priorities.
At State, employees focused on innovation, responsible AI, technology, and cybersecurity are working together as the department develops AI governance.
Some of the biggest delays come not from cybersecurity controls but from data-sharing agreements that require legal and organizational coordination, Osborn said. State is working to automate portions of those processes to reduce delays while maintaining necessary safeguards.
Organizations often assume “you’re either, moving really fast unsafely, or you’re doing nothing,” she said. “These are not dichotomous. You can … you can do both.”
Architecture can provide another lever for speed.
“You have to build composable modular architecture from the start,” McGarry said. “Not for its own sake, but because modularity gives you the speed and control.”
A modular approach can allow agencies to replace individual components or adopt new AI capabilities without rebuilding an entire architecture or governance framework.
“The speed comes from that modularity, not from cutting corners on security,” McGarry said.
For higher-impact AI uses, accountability also has to remain visible. Decision-makers should be able to understand how an AI system reached a recommendation and retain the ability to override it when necessary.
“It shouldn’t be a black box,” McGarry said. “It has to, by design, be a glass box.”
View the full discussion: https://www.meritalk.com/event/turning-data-insights-into-mission-outcomes/