Drawing on more than 50 combined years of experience in the public and private sectors, Millan and Zaborda recommended:
- Don’t start with the finish line in mind: “You know when they say begin with the end in mind? I think with this, that’s not the advice. I think it’s see what’s achievable and build on your success because if you begin with the end in mind, I think you might be asking too much of these systems that are still in their infancy,” Millan said. “If you can achieve some success and then push a bit further, I think you could build on that success and prepare the people around you on why that’s a good thing.”
- Have good quality data and start with use cases: “You need to have good quality data for AI. If your data is not good, the answers or responses you receive will be as good as your data,” Laborda said.
Another concept he discussed is how to define AI’s business value.
“How can we get value out of this? For us, the most important thing is, where is the business value, right? I love technology, so I will implement anything you put in front of me,“ he said. “But which problem am I really solving? I mean, figuring out the ROI of what you are implementing is really hard. We are taking an approach where we define use cases and real-life problems that we can solve and start building towards that.” - Keep humans in the loop: “I’m a big advocate for human-centered design,” Millan said. “Remember who your audience is. Remember what it is you’re trying to achieve. If you’re making something in consideration of someone else, right? Don’t forget what that objective is, and use these tools to enhance things so you can be a better human, right, and not replace the human element.”
"We started very simply and said, 'OK, we have documents with policies, and God knows what else. There are thousands of documents. Well, how can you be smarter in trying to get information out of them?'" he said. "So, we put an AI in front of it and said, 'OK, I’m just gonna ask you questions. You go find it and come back.' So, if you want a smarter search, that was a good pilot we ran for a while. We got some good results.”
The department found that databases are beginning to incorporate AI tools directly, raising questions about what role third-party AI should play when working in those environments. Laborda and his team started comparing results and data from third-party products versus integrated AI models by testing non-confidential or non-privileged customer data.
In Orange County, Millan said she and her team use AI through robotic process automation and language models.
“So, how many of you had to bring forward a budget with a 0 percent increase this year? Right? We did. So, we have to find ways to use our resources more efficiently. I was not able to add additional staff in pretty much the last two years, and yet, we have new regulatory requirements coming down from the state, new federal requirements rolling down. We have an influx of federal funding, right? So, I have to find ways to do more with less,” Millan said. “For us, we’ve leveraged a lot of robotic process automation and the use of language models to help ourselves.”
In procurement, for example, the county handles thousands and thousands of transactions.
“If I have a step that takes five minutes, right? Twenty-five thousand transactions, 125,000 minutes, 2,000 hours, that’s a full-time employee, right?” she said. “You don’t get the return on investment. But in fact, it is 2,000 hours. So, we look at even small improvements we could make. We look at those five-minute improvements, 10-minute improvements, the things that we do over and over, and that’s how we’ve been quantifying the ROI. So, to date, we’ve done small RPA projects that have saved 1,000 hours.”
Another use of the technology by the county includes using language models to catch specs that shouldn’t be included in construction plans.
“In my role, I oversee construction and professional services. I oversee transactional procurement, and I oversee formal procurement, right? So, even for things like construction plans, I need to check those construction plans against requirements, right? We have a prototype that’ll do that,” Millan said.
Lastly, other uses of the technology in the county include using prototypes to load agendas, requests for proposals and proposals to better access and share information during commissioner briefings and to conduct pre-legal reviews.
*The Central Florida Digital Government Summit is presented by Government Technology, a publication of e.Republic, Industry Insider — Florida’s parent company.
