Government AI adoption: Best practices for public service transformation

As artificial intelligence (AI) begins to impact every aspect of life in the U.S., government agencies across the country are increasingly exploring how AI can help them deliver public services in quicker, innovative and more efficient ways. Unfortunately with legacy technology and practices still at play, incorporating AI into operations is a complicated endeavor that many are realizing requires serious overhaul of data practices and IT environments.

So, where can government organizations start? In a recent roundtable discussion hosted by The Advanced Technology Academic Research Center (ATARC) and sponsored by Iron Mountain, experts across government and industry gathered to discuss how they found their footing when it came to government applications of AI, and offer advice on how other government organizations can do the same.

Start small

It can be tempting to begin with a large project, but the best way for government organizations to begin their AI journeys is to start with small , manageable projects that can help them build the knowledge base and data foundations necessary to grow their use of the technology. In fact, it’s key that agencies not try to take on too much at once or create solutions that aim to be wide-ranging. Instead, organizations should focus on very specific, manageable use cases that can grow their understanding, confidence, trusted data repositories and AI infrastructure.

Establishing a strong data foundation

Data fuels AI innovation, which makes a strong data foundation a cornerstone for developing powerful AI solutions. Here are four key steps to establishing a strong foundation:

1. Digitize paper records and standardize digital data and digitized records in a centralized repository.
2. Implement data quality control through cleansing and validating data and enforcing data entry standards.
3. Generate metadata for each dataset and create structured taxonomies to define relationships and categories within your data.
4. Implement data governance by defining ownership and responsibility and establishing and enforcing policies related to access, security, and privacy.

“The key to successfully integrating AI is focusing on targeted, specific use cases — one at a time,” said an IT executive at the Pennsylvania Treasury. He noted that rather than trying to apply AI broadly, organizations should identify practical, day-to-day processes where AI can improve efficiency and build from those successes with clear frameworks and guidelines.

“As you build one degree of efficiency at a time, eventually we’ll get to a much more efficient overall process, but it has to be one use case at a time,” he said. These types of use cases, while small at first, can ultimately grow the underpinnings necessary to build more efficient enterprise solutions.

A prime example of this is Miami Dade County, which launched its use of AI with a tool for the county’s pet shelters that matched potential pet owners with pets that might be a good fit for them based on their lifestyle and preferences. For example, a person could ask the AI tool for a small, low-energy pet that’s good with kids and the tool would search the database to provide them with potential matches.

“That use case, as innocent and as fun as it sounds, helped us mature our understanding of the data needs for AI tools. We saw what was both needed and what we wanted to have in a data set for an AI tool,” said an IT leader for Miami Dade County, noting that it was the first use case that depended on a live view of a database which was, in this case, the notes within the Animal Services database. “It gave us an idea of what we needed if we’re going to … start building our governance around AI, one use case at a time. It starts with that data governance foundation.”

Invest in R&D

It’s important to note, however, that organizations don’t always have to learn by launching public-facing projects. Within the U.S. Treasury Department, the Office of the Comptroller of the Currency has seen the benefits of investing time and budget toward experimenting with AI.

In particular, the office wanted to build out its already strong data governance and security programs by tapping into AI to ultimately drive the office’s data accessibility.

To test the best way to do this, the office set up a sandbox where IT teams could play around with AI and figure out not only what the technology could do, but also how the office needed to structure and regulate its data — and the data from institutions like banks that fed into the Treasury Department’s database — to support that use.

“We realized, as we mature, we need to go towards an agentic AI where you have a data orchestrator, you have little [AI agents] that have a specific function,” said the executive. To support those functions, the team used the AI sandbox to interrogate both the internal and public-facing policies that informed its current data governance and provide insight into the areas that needed to be modified to support future AI use in the ways that would be most beneficial to the office.

The office has found that by investing in R&D for AI, they could quickly and safely experiment in ways that help set them on the right path for AI innovation. The investment, while new, has certainly been worthwhile and one the executive encourages other agencies to make space for.

“We were missing that real, true research and development piece,” said the executive. “So, we set up a little bit of money [for an R&D function].”

Robust data is key — but tricky

An AI solution is only as valuable as the data that backs it.

The challenge is that not all data is clean, organized, of high quality, or even in a structured format that can be easily incorporated into AI platforms.

For starters, many institutions that gather the data that will ultimately inform AI solutions simply don’t have the understanding or ability to collect strong, quality data.

In Pennsylvania, for instance, the IT leader for the state’s treasury has found that as the organization works to create and introduce AI-backed solutions that aim to serve the business community, the businesses that would benefit most struggle to create the data foundation necessary to support these solutions.

“It’s a challenge for the business community to get up-to-date and accurate data so that it can become the most valuable tool possible,” he said. “As we’re introducing these tools to the business community, they’re saying: ‘Hey, wait a minute. We’ve got to get our house in order for us to make this a valuable tool.’”

And it’s not just user data that’s a challenge. State agencies looking to synthesize all relevant data sources to inform AI solutions must also often pull information from federal sources, which present similar challenges — data required to reach a decision or complete a process can be dispersed among different agencies and departments. While some of that data might be of higher quality and better organized, it’s also a challenge for many organizations to develop the foundational data-sharing and security protocols necessary to ensure the data is being used responsibly, not to mention the immense work it takes to analyze that information and incorporate it in useful ways.

Finally, not all data sources are even in a format that many organizations can access, but unlocking that data is critical to creating a full, well-rounded picture for AI solutions to pull from.

“As you can imagine, a lot of the historical data is on analog records, so it’s locked away on a piece of paper or fiche or film. And therefore, one of the exercises [government organizations] need to go through is to digitalize and capture some of the metadata off of those analog records to have a complete picture of a citizen’s record across agencies,” said an executive for Iron Mountain Government Solutions, a leader in innovative storage and information management services.

Chart a path to strong data governance

With so many data challenges at their feet, it can be difficult for a government IT team to know where to begin.

In Miami Dade County, the IT team found it has been best to not focus too early on the specific in-the-weeds data challenges at hand — like breaking down data silos or uniting data — but to zoom out and focus on the user as a way to draw a roadmap to the types of data governance necessary to inform these AI tools. Along the way, it’s also crucial to ensure the government staff understand their role in collecting and categorizing the data that informs these tools.

“Largely, that approach helped our teams realize: We need to start with our customer. We need to start with the person that this tool or the solution is for, and then as we map that journey, we can identify the data points along the way that will inform that use case, or inform that that practice area of AI, but also inform our understanding of the customer’s business, so that we can build upon that foundation to do great things for them,” said an executive for the county.

Once an organization has a concept of what solutions and data they may need to deliver, it’s important to establish strong data governance that can support those solutions.

“You need to establish that unified data governance framework, especially if you’re going to try to improve public services and work across organizations, so we don’t have separate silos,” said the executive for the U.S. Treasury.

To create those data standards, key leadership and stakeholders need to be on board with data policies early and often.

“It all starts with getting the decision-makers and the executive leadership behind it,” the U.S. Treasury executive noted, stating that this is especially crucial when multiple agencies with the same mission are involved. “If you really want to make an impact, the point is to improve decision-making and make high-impact, quality decisions. So, the best way to do that is to have your leadership behind it.”

Finally, while getting the stakeholders and data structure necessary to create the foundation to enable AI solutions may seem like a tall order, it’s also an opportunity for agencies to revolutionize the way they deliver public services.

“AI is only as good as the data that’s going to be behind it. Let’s not take that as a fear factor, or a reason not to do it, but actually use it as we learn, going forward, to build that robust data foundation and information governance,” said the Iron Mountain Government Solutions executive, “so that we can service all these citizens and even our workforces to have that digital app kind of experience not too far in the near future.”

Iron Mountain Government Solutions can help your government organization build the data foundation necessary to make the most of AI.

Learn more