From Hype to How: A Practical
Playbook for AI in Government

Hello and thank you for joining us for today’s webinar, From Hype to How: A Practical Playbook for AI in Government. I’m Caroline Boyd. I’m principal at MeriTalk and I’m going to be moderating this discussion. So before we get started, I’d like to thank PayIt for partnering with us on today’s discussion and supporting this panel and important conversation. So as digital services become the norm, state and local leaders are looking to improve AI efficiency and deliver better results and experiences, but there are a lot of possibilities, from service portals to back office operations, and the big question is where do you start, and importantly, how do you bring people along? So in this webinar we’re going to hear from PayIt and public sector peers who are going to share practical frameworks for choosing the right AI pilots, building trust, transparency, and scaling that works, most importantly scaling that works securely and responsibly. So hopefully you’ll leave our discussion with a checklist of real-world lessons that you can put into action right away. All right, so with that I want to get started. We’re excited to have two great panelists with us today that bring a wide range of experience from both the public and private sectors. First panelist is Adam Christensen. He’s the senior vice president for product at PayIt. Adam has more than 20 years of leadership experience across technology, payments, and fintech. At PayIt, he leads the product and design teams. He creates consumer-grade experiences for clients and their customers. So, welcome Adam. We’re glad to have you with us.
Thank you. Great to be here.

And then I’m also pleased to be joined by Christian Napier. He is the director of artificial intelligence for the Utah Division of Technology Services. Christian oversees AI implementation across the state’s executive branch and was recently recognized as one of the inaugural AI 50 award recipients by the Center for Public Sector AI. So, welcome and congratulations, Christian.
Well, thank you very much. Very happy to be here.

Great. So, why don’t I ask each of you to just briefly say a little bit about your role in relationship to our discussion today. So Adam, why don’t you go first?

Yeah. So, there’s an exciting, transformational time. I’ve seen a lot of trends come and go. I haven’t seen anything hit this hard and fast and transformative as AI is hitting. So, I know there’s been a lot of buzz, a lot of buzzwords out there, but there is a lot of reality to the gains here that we’re seeing both internally as well as with the agencies that we’re working with. So, very exciting time. So as head of product, I am not only in charge of really helping transform our own development processes, you know, how we envision and really end to end build and launch products, but also how do we inject AI into the products that our agencies can get the benefit of. And so, very, very exciting time.

It is exciting, the reality of the promise, right? So that’s great. Christian?

Yeah, I’ll just draft off of what Adam just said there. It really is. This generative AI and AI in general is once-in-a-generation technology, and we are at an inflection point where we can leverage this kind of technology to the benefit of our state agencies and our residents and citizens like never before. So super excited to join this conversation.

Well, we can start right there, Christian, with you, and maybe you can kick us off by talking a little bit about what problems you are solving first, right, with AI, or looking to solve first with AI, and why did you choose that path?

That’s a great question. You know, ChatGPT launched in November of 2022 and took the world by storm, and everybody was scrambling, like, how can we use this? In the state, we initially came up with two fundamental questions, really foundational questions. Number one, given the challenge of hallucination, can AI be deployed in a responsible way that it can be trusted? So that was number one. Number two, we were looking at really internal facing use cases and what could be impactful for our workforce. We have 22,000 employees. So we embarked on two different pilots in mid-2024, let’s see, mid-2024 I would say. One is we looked at something like tax, where accuracy is paramount, and we conducted a unique pilot with the Utah State Tax Commission. What we did there, again with a core focus and goal of determining whether we could engineer it in a way that the responses could be accurate on a consistent basis, so we conducted a bake-off between four vendors and their solutions to see if we could get a chatbot that would be used by internal tax call center employees to help them answer citizens’ questions related to tax. And we did that, and the first responses were promising but showed us that we had a long way to go. And so we did a second phase of that project, and by the conclusion of that second phase we were able to get the bots to respond accurately, as judged by the tax commission experts, equivalent to or better than a knowledgeable human 97% of the time. So we felt like, okay, we can do this, and we have now rolled that solution out into production in tax, and we are starting to deploy that in other agencies as well.

Then on the employee productivity side of the fence, we are a Google shop when it comes to office productivity. And so we conducted a pilot of Gemini for Google Workspace starting in June of 2024. And the results of that pilot were promising, with 257 folks participating across all state agencies. What they showed was that on average employees were stating, the people that used it in a regular, consistent way, were self-reporting, you know, self-attestation, that they were saving on average three and a half hours a work week. And so we decided we were going to roll this out to everybody. So currently in the state we have not rolled it out to everybody, because agencies opt in, and that’s still a work in progress, but the majority of agencies have opted in. In fact, all agencies at least have some segment of the workforce that are using AI. Currently, as of today, out of the 22,000 employees there are about 9,500 of those employees who use it on an active basis, meaning they’ve used it at least once in the last four weeks. We conducted a survey of those employees last month, and they reflected the same indications that we got in the pilot. So for the high-volume users, they are saving a significant amount of time and they feel better. Those were kind of the two things, and that was probably a long way to answer that question, so I apologize, but really foundationally we wanted to see: can AI be tuned to be trusted, and can it have some type of impact on employee productivity? And we think that the answers there are yes, but there’s still a long ways to go.

Well, we’ll come back and dig into a few of those things that you’ve highlighted as we go through the conversation, because I think there’s some important lessons for our listeners to learn there. But Adam, I know that you’re working with state and local governments across the country. Just piggybacking on to that, why don’t you talk a little bit about, sort of, is this typical of the types of problems that you’re seeing state and local governments address first with AI?

Yeah, I think Christian is right on. I think there are a lot of individual experiences out there where the benefits are clear of, first and foremost, removing some grunt work out of life. And I think what we’re seeing is that that pulls those type of use cases forward into agencies. You know, the Gemini use case is a very salient one that we see. In other words, hey, can you just record our meeting, take notes, can you give me a draft of a paragraph that I want to send in an email? I mean, these are the sorts of things that we see increased comfort with across agencies. Not all, but definitely kind of as a starting point for people to say, “Wow, I don’t have to do that. I can add higher order value.” And so I think those are definitely exciting to see, and we certainly do. We put a lot of AI to use internally as well in our operations, but then even in our user research, you know, when we’re testing and interviewing users and being able to synthesize and bring together, I mean, the hours and hours of savings from not having to go back and type in their answers and things like that, it’s just really extraordinary, because we can actually do more research and better research and apply those insights to our products a lot more quickly. Yeah, we see really, even though residents are excited to get that benefit, I think there’s a little higher risk to some of those resident facing types of products that we’re working on, and others definitely want to launch. I think agencies though are really starting first with really how do we increase our own both operational efficiency internally, but as well as our confidence in AI, before we then expose that to our residents that we serve. So, I think what Christian’s saying is really footing with a lot of the agencies that we’re working with in terms of where the starting points are with AI.

It’s great to hear that there’s like a commonality here in the types of projects that state and local governments are diving into first. Christian, let’s go back and dig in a little bit and maybe talk about kind of your rubric for picking the pilots that you did. You know, looking at maybe impact versus feasibility. Adam mentioned risk of course being one of them. Is there anything that you would change about the way that you went about choosing these pilots in this first cycle?
Yeah, that’s a great question, Caroline. I think back to it. You know, we’re pretty new. I got into this role about a year ago.

We hired our first AI analyst almost exactly a year ago today and the second one shortly after that. In March, we stood up what we call internally the AI factory, which was a streamlined, standardized way for agencies to take pilots from ideation, or take these ideas, right, from the initial phase, this ideation phase, to pilot to production. And I’ll be honest, after a few months we were looking around and we’re thinking, you know, we’ve done a lot of these kind of chatbot pilots, one, because we figured out how to do them, you know, by going through that tax experience, and two, because agencies were like, hey, you know, we heard about this cool thing called AI, and it typically manifests itself in the form of a chatbot, right? So we had a pretty fair number of these chatbots. But as we were going from pilot and looking to go to production, we started running into challenges. And I’m not going to go too deep into all of those, but I think there are two fundamental things that really stood out. Number one was it was very inexpensive to pilot, because vendors were very generous with providing free proofs of concepts and pilots and things like this, but it can be expensive to go to production, and they realized, oh well, maybe I don’t have the budget right now to cover this. We may need to put that off. I need to go to the legislature and get an ask for some funding or something like this. That was one. The other is that by and large these chatbot style use cases are very much sidecar processes. Like, I’ve got my existing business process that I’ve been running for 20 years and I can still run that process, but oh, by the way, I have this other chatbot that can be over here to the side that can help me if I need it.

But there are a lot of people who feel like, well, I don’t really need the help, like, I know my job, I know what I do. And so, you know, we were looking at it and thinking, oh well, the uptake, the adoption, isn’t as high as we would like or we would anticipate for the amount of investment that we’re putting in, just primarily in man-hours, you know, into these pilots. And so we have tweaked our factories so that we focus much more at the beginning on understanding what the business value is of deploying AI, rather than, oh, we can do it, it’s a chatbot, we know how to do it, let’s give it a shot and try to roll it out. So I think we’ve been trying to change our focus to more higher impact, higher business value use cases than perhaps we did initially. I’m glad that we went through the experience. It’s helped us to tune our AI factory process and make it more effective. And I’m sure we’ve got a long ways to go still on that. It is still early days. But those were some of the important lessons that we’ve learned over the last few months.

That’s the beauty of something like an AI factory or a pilot, right? That you learn from it, you make the tweaks, so that you can roll out higher impact opportunities for other agencies across the state as you go. So it’s a good approach. Adam, as you listen to this and knowing your design background, what do you think is like the smallest, most useful pilot for state and local governments to be looking at? Maybe like what would that scope look like? What does that timeline look like? What does success look like?

Yeah, I love these questions. So, I mean, one that we’re working on right now and in the process of launching is really, if you think about one of the core problems that a lot of agencies have, is how do I validate that this is actually a resident that really is meriting the services that we provide? And so right now there are several agencies that, you know, part of that process, if I want to unlock certain services, I have to bring in one or two documents and prove that I’m a valid resident of your city, county, state. So this is something that we’re in the process of launching right now, and this does have a resident side of it. And so it sounds big, but really AI has really come so far that this is a very feasible and straightforward use case. So the resident, instead of bringing in documents and then sometimes being turned away: oh, it’s not recent enough, or it’s the wrong document, or oh, go home and go get another document. And so that’s one frustration, is just, is this the right doc? Second of all, the problem of just the time it takes for an agent to look at that doc. Many agencies are still needing to walk over, scan the document, and upload it into their own system as part of the workflow. All of this takes time and just effort and human toil that’s really unnecessary. And so instead, the pilot we have is both the resident has a way to be able to upload documents and do a document verification beforehand, and allow the AI models to just say, hey, is this within our county or within our state? Is this a valid doc with a valid timestamp? And can we basically let this resident know that, yep, you’re good to go on this residency check? Or it looks like the right docs, the agent will need to take another look in office, but go ahead and go into office. So that saves potentially multiple trips into office. And then on the agency side, they can simply look, and if there’s a high confidence, that step is done. If it’s, say, on the border or on the bubble, the agent can quickly take a look online in the workflow and say, oh yeah, that looks good, that’s good enough.

We can verify, but avoid any of the physical document problems or scanning or any of those steps that are unnecessary, and then be able to just go on with actually the items of business. I mean, this is just a residency check before the actual substance of the visit. So that’s a pilot. I would say, you know, a 60-day check is: is the agency able to upload, you know, a variety of test documents? I mean, we’re doing a bunch of testing on our side, of course, but are they internally able to upload and see for themselves? Yeah, we have confidence in those documents that they are rejecting, those that they’re accepting, and those that might look good but might require that human review in office. And then at 90 days, be able to start doing a soft launch or even a hard launch of a certain type of agency service that says, yeah, here you can upload this. And then even offer, I mean, what we’re offering is, hey, you can go to this, you know, FastPass type of line in office, cuz we’re going to reward you for doing that up front, and here you can get streamlined service over here in this line, where those agents are trained to just speed ahead and look at a doc as needed, but more quickly because it’s already digitized. So, that’s a use case we’re very excited about, and also very confident in, in terms of the ability to deliver this and really help, really allow residents to spend less time in those lines and agents to also spend less time per resident to get the job done.
As a resident of a state, I know that I will appreciate as those pilots, you know, come to fruition, for sure. There’s high value in that type of pilot turning into real-time action for citizens, for sure. One of the things that we haven’t yet talked about, which actually is unusual I think in some of the AI conversations, is data and the data governance piece. So Christian, you know, before you’ve moved into these builds and these pilots, what data is in bounds and out of bounds, so to speak? What are the checks that you’re doing on data? Like what’s mandatory for that before you move to sort of that go/no-go decision on the pilots that you’re putting in place?

That’s another great question, Caroline. When I reflect on this, really the guiding principle for us, it’s more than a principle, it’s actually embedded in the legislation and statute here in the state: the privacy of the individual is paramount. That’s the most important thing. And so we have an Office of Data Privacy. Each agency has its own privacy officer. So they own the privacy elements, but we make sure that we are adjacent to them and aligned with them in our AI factory. In addition to us conducting an initial AI risk assessment when we look at any potential pilots, we also ensure that the Office of Data Privacy is informed and they have a say. So they conduct their own initial privacy impact assessment for any pilots. And then in addition to them, our security team will do an initial security assessment as well. If we’re doing something in the cloud or something like that, our hosting team may have some involvement as well. We all provide our recommendations to leadership and then they make the decision on whether to move forward or not on the pilot. So as an AI team, we don’t act as gatekeepers. We’re not saying yes, no. What we are doing is providing a recommendation, along with the other areas, to ensure that if we’re going to go down this path, that we’re going to do it responsibly.

Once a pilot is done, if there’s a decision to move to production, then a full privacy impact assessment is required. So the data privacy officer for the agency works with the teams to conduct that assessment. A full security assessment is also undertaken, and then all of that information is brought to the Application Review Board, the ARB, to make their ultimate decision as to whether this can be launched into production or not. So it’s not necessarily cut and dried, this data is in bounds, this data is out of bounds. It really is on a case-by-case basis. Our principle is that citizens provide us data to perform the transactions that they have requested and nothing more. So we take that very seriously. We don’t say, okay, well, they’ve given us data for this, maybe we could use it for this other thing. Absolutely not. If they have not given us consent to use the data outside of the way that we are using it just to process their transaction, we are not going to use it. The other thing I’ll say is the agencies have more expertise in the regulations than we do in our Division of Technology Services. And so we really defer to the agencies when they tell us, even though we say, okay, Gemini can be used for this, that, and the other, and it’s FedRAMP High and it’s HIPAA compliant or whatever, they may come back and say, well, there are other regulations that really prohibit us from using it in this particular way, and so we want to put additional restrictions on it. And we say, okay, that’s totally fine, you know, we defer to them. So, just to kind of sum it up: privacy is paramount. Agencies are the experts in their regulatory environment. We don’t do anything on our own. We partner with our counterparts in privacy, security, etc., to make sure that we’re rolling it out responsibly.

Great to hear that the collaboration with the agency is sort of number two in your list there, because I think that helps keep that responsibility clear and prevent you from going too far in a pilot, right? For sure. And Adam, I guess on the design and the product side, who’s accountable during the pilot and sort of the post-launch? What’s the kind of operational governance that’s directing you and your team’s day-to-day as you work with government? Are there sort of some checks like human in the loop or approval gates, things like that, audit logs?

Yeah, you know, absolutely. I mean, on the product side, I mean, I’m ultimately accountable, but each of my product managers are very hands-on through any initial launch. We take that very seriously, and ultimately it’s on the product manager who’s launching in what we call an early adoption phase. As we get to more general adoption, or GA, that general availability, that gets to more of our operational side, but that’s really after passing all of the criteria that we have established. But that’s really on the product side. You know, I think it’s really important on the agency side that we’re working with the right leadership, that, you know, it does take an innovative mindset. We try to find those that are really ultimately accountable for the positive result of the launch and, you know, have all the incentives aligned to make it successful, say versus sitting in a seat of, you know, disproof. So I think that’s been really important, both at the overall leadership level, the day-to-day operational leadership level in the agencies, and then, you know, in the case that I mentioned, selecting the right offices that would give the right, say, tailwind of excitement, that they’re really looking to remove grunt work out of their day-to-day life. And you know, let’s face it, not everybody feels the same way or sees this the same way. For some, it’s threatening. So I think that’s where really that right level of leadership is critical to how do we serve a growing population with the same staff, right? Or how do we unlock our staff to be able to do more useful and helpful things for our residents, for example? So that’s where we really are, you know, we really make sure that we’re working with the right folks. I think that, you know, all of those things you mentioned in terms of audit logging, you know, being able to give the agencies the ability to see everything, we certainly look at that from a product perspective, but to be able to give the agencies confidence on the back end to say, “Hey, look, we’re going to look at, say, all of the documents that have been approved, all of those that have been rejected, you know, is this the right level of accuracy and validation that we would expect?” and to always be able to go back and validate and look at anything we’ve done beforehand, to make sure that that confidence continues to build and goes really in one direction as you improve and refine the product and the models to meet the needs.

Well, all that kind of takes us to training, and as you talk about leadership, too, change management, making sure everyone is on board. So, Christian, I’ll just turn it back to you. What’s the training path for the frontline staff actually involved, the supervisors, leadership, and how do you communicate those changes, you know, to the staff? Like, how are you rolling that out?

Yeah, that’s a great question, and I think it ties in nicely to what Adam was just talking about. So I’ll kind of break it into some different components. The first component is policy. So, you know, after ChatGPT was released in November 2022, everybody was scrambling trying to figure out what does this mean for us? Certainly that happened in the state, and in the fall of ‘23 the state issued its first enterprise generative AI policy. One of the first requirements for all employees was to read and acknowledge to abide by the policy. So that was kind of the first step. But included in that policy was the responsibility for the Division of Technology Services to provide some responsible use AI training. For us to be compliant with our own policy, we had to make sure that we made some training available to all state employees. And so we decided to link that to our security awareness training, which was already going on, an annual mandatory training for all statewide employees. And so we basically sidecarred the AI responsible use training to the security awareness training, making it an annual mandatory training for all state employees. That launched in 2024. So that’s still going on today. And then aside from that, on the office productivity side of things, we’ve done a lot of work to help employees get up to speed. We’ve had a number of statewide lunch and learn sessions. We have cross-agency AI user groups. We have customized, bespoke AI training on request for agencies who say, hey, my employees have specific needs, can you develop training that is tailored especially for our agency or our division? So we’ve done a lot, but there’s still a lot more that we can do. Communication in the state is a bit challenging because people don’t want to be inundated with things, and agencies don’t want our division to be inundating their staff with a lot of stuff. And so, we’ve got communication channels that we need to go through to get things out to agencies. And so, we’ve tried to balance, you know, overwhelming people, but at the same time making sure that they’re informed. We’ve also had open office hours and some other things like that. But one thing that I think we undersold is the power of organic growth, of employees discovering for themselves how to use it rather than top down, kind of, here are all the cool ways you can use this. They’re figuring it out themselves, and their natural inclination when they figure out some cool use case is to share that. You know, for example, last week I got a note from our networking team, and they shared an experience where they used generative AI to create a bash script to go through and evaluate 2,000 different web links and websites in the state, and how it saved them literally thousands of hours. So like, if it, you know, we were thinking we would have to do this manually and it was going to take a huge amount of time and resource, and what do they do? As soon as they do it, they’re so excited about it, they start sharing it: look what we were able to do with this tool. And so I think one of the things that we can continue to do better is to promote this kind of organic, bottoms-up learning about AI, as opposed to trying to, you know, centrally plan the entire thing.

I love the organic approach, too. I know in our own organization that has provided some of the best ideas and enthusiasm as well for things that can really improve efficiency. Do you have any, like, adoption metrics or gates that kind of determine when you roll things out, you know, more broadly?

In terms of adoption metrics and gates, as I mentioned earlier, we took an agency opt-in approach. Even though we’re a centralized IT shop, we said, okay, when your agency feels like they’re ready, then we will enable certain features. You tell us. What we try to do is educate the agencies on the responsible use of this. You know, some had some misconceptions. For example, oh, our state data is going to be used to train models, and we’re like, no, we have this ring fence. Oh, but it will be used to train just for our state so that it knows how to answer our own employees’ questions better. No, that’s not actually how it works, you know. So we’ve had to do some agency education, you know, on leadership just to kind of level set expectations for what the technology does and does not do. But really it’s been kind of on an agency-by-agency basis in terms of opting in. I will say, and Adam, you may have seen this as well, you know, this is different because generally speaking agencies, with a few exceptions, have been knocking on our door saying, hey, we want this, how do we get this in? Now, this is not like it was when you were migrating from on-prem to cloud and you’re like, you’re going to pry my server from my cold dead fingers. Like, this is not that, where you have to go through and do a lot of persuasion. I mean, it’s kind of the inverse. It’s like, hey, people are super excited. How can I implement it? How can I get my people to work more effectively so that we can serve our citizens better, etc.? So, probably a long-winded answer again to your question, but for us, it’s really kind of driven by the agencies, and the agencies have a very keen interest in it.

Well, actually, you ended on a good note, because what I wanted to do is turn back to Adam for a minute and talk about, we’ve talked about getting everybody on the inside on board, but let’s look a little bit about what happens when there is output, right, to the residents. What steps are you seeing with the customers that you’re working with that help ensure things like explainability, accessibility, language, the bias issue, you know, equity in general? What are the mitigations for citizens and residents on the other end?

Yeah. No, I completely agree with everything that Christian just said, and certainly exciting to see all of that and the commonalities probably across all of our organizations with that internal groundswell. Yeah, a really important question, a set of questions here on the accessibility and bias and these sorts of things. I mean, first of all, we’re very selective on multiple approaches. I mean, the starting point is really critical for us. I mean, first of all, we are starting with the most transactional type use cases that we can, just to unlock the value. We’re not providing therapy or counseling or anything like that. It’s like, hey, get in, get out, and that helps really narrow the field in terms of a lot of the verbiage and types of potential pitfalls there when residents look, I just want to get this done. So that’s where we’re really trying to start with AI and allow the models to continue to evolve that we’re using underneath, but really first and foremost start with very transactional jobs to be done for the resident, and agencies for that matter. Second of all, I mean, the models themselves, the companies we work with, we’re very selective there in terms of really understanding how these models came to be. And then I think third of all, you know, we definitely have experts, the humans in the loop, right? Accessibility experts, compliance experts, these sorts of humans that are continually looking, and you know, there’s still a level of human rationality and read-through that we need to apply, both within our own product and development but also in partnership with our agencies, to make sure that, you know, that they’re very confident on those dimensions, and that from an accessibility, hey, we’re not just meeting the base compliance accessibility, for example, but we are making this also very useful and usable and, you know, meeting the higher order goals here over time. I mean, there will be, you know, models that keep other models in check, right? And that’s, you know, we’re certainly looking and experimenting with those sorts of things as well, to use AI as a counterbalance, you know, against some of these areas where other AI models can overshoot, and that’s very exciting as well. I think there’s a whole frontier where multiple models will be used to really provide that pristine end result.

So on the inside, Christian, I know you talked, actually in one of our earlier questions, you mentioned how important trust is, right? And you’ve talked about a couple of the communication tools and paths that you’re using, but how do you keep that ongoing trust building with these engagements, and the confidence not just for citizens but also for your leadership or policymakers, you know, who you may need to go back to for additional dollars?

Oh, that is a really important question, and I’ll kind of preface my response there to say that it’s in part aspirational, because we have not launched any citizen-facing AI, but we are now developing proofs of concept for that. So I’ll take the policymakers and the citizens kind of as two separate things. So on the policymaker side, our leadership meets frequently with the governor’s office, with the legislature, to build relationships of trust. The channels there are open, the communication is pretty good. On the citizen side, our vision is intent to leverage existing systems or capabilities or initiatives that are already in place to serve citizens without reinventing the wheel. So the first one is what’s called the Utah Design System. So the Utah Design System was established a few years ago to create a standard look and feel for all of the state’s websites, because in the past all the agencies had their own websites and they all look completely different and they behave very differently. And so the Utah Design System was established to create a more consistent experience for citizens across various agency websites. So they all share the same kind of header and footer and the way buttons look and feel and so on and so forth.

That’s still in the process. Not all of the agencies are on board with Utah Design System. But our view is that the citizen-facing solution, the UI, should be constructed using elements of the standardized Utah Design System, and they should be the people who should actually design that front end rather than us. We are not front-end people. We are AI people, and they have all of the accessibility elements and multilingual elements already kind of baked in to that platform, and that’s really important. We don’t need to be reinventing that. Element number two is, again, several years ago an initiative was launched by the legislature called the citizen portal. It’s now been renamed MyUtah. The idea behind that is that as a citizen, there’s one place I can go to do any transaction with the state, right? So rather than go to all the different agency websites, I can go to MyUtah, and the transactions that are relevant to me are there. I can find them. I can order a birth certificate. I can renew my vehicle registration. I can get immunization records from the same place. I don’t need to go to different agencies.

So, what we’re looking at there from a citizen-facing AI perspective, as Adam was talking about, if you’re looking at processing transactions, rather than have agencies build their own interfaces to some chatbot front end that allows people to do transactions conversationally, let’s have a unified experience where the MyUtah platform is the back office for the conversational interface. Any agency wants to have citizen-facing transactional capabilities, then it goes through the MyUtah platform, and they don’t have to worry about doing multiple integrations. They do one integration to MyUtah and that’s that, and then they’ll get the conversational integration added on to it, no big deal. The third thing is something that you mentioned earlier, which is about the citizens being engaged and having some kind of a say, like, there’s already a process in place called the, there’s a citizen experience initiative that’s been going on now several years in the state. Again, this is kind of a standardized statewide framework for getting citizen feedback on things. Rather than us standing up our own kind of feedback loop and trying to figure out what do citizens like and don’t like, let’s leverage what the state is already doing in this area to get that kind of feedback when it comes to AI. So for us, I think building on these three existing elements has really been important for us. And I would say the fourth component and final component is we are looking, you know, we are preparing an ask for funding from the legislature to have this be centrally funded, because we want the citizens to have a consistent experience across the entire state, a no-wrong-door approach, and also provide a financial incentive for agencies to buy into this standardized framework, because if we have a centrally funded mechanism, then agencies don’t have to carve funding out of their own budgets or go to the legislature for an additional ask to get AI for their citizens, their constituents. We will have it all in kind of a unified approach. So, that’s our aspiration, and we’re going through the process now, and hopefully things will come out favorably for us.

Well, I hope so as well, and I think it’s great to have an aspiration like that, that is, you know, baking it in, having a unified platform. I mean, Adam, I just want to pull on that a little bit. If you could talk about what you’ve learned about combining AI into, you know, digital wallets or reminders or that proactive side on the payment side, since that’s your specialty area, that processing of transactions. What are you seeing with other customers and what’s changing as they approach that? I mean, what are the synergies you hear with how Christian and Utah are approaching that integration?

Yeah. No, he’s speaking my language. From a product perspective, reuse as much as possible, and patterns and systems and underlying platform capabilities that can unlock multiple use cases. So, love that all. Yeah, so on the payment side, you know, from a core payments perspective, you know, AI hasn’t been as critical, because, you know, really there are clear patterns out there, and you have those that pay immediately, and then there’s those that pay at the last minute. I mean, there’s kind of an inverse bell curve of payments when it comes to meeting obligations such as property tax or deadlines with government agencies, right? So, we haven’t found as much criticality. I mean, our core platform, you know, make sure that residents are aware of, hey, look, you’re 48 hours from a penalty, you know, make sure that you make this payment. I think, though, where we’re seeing some exciting frontiers are really just unlocking the, okay, what else, while you’re in flight of making this particular payment, what else do we know about you that we can surface up right now and have you just get it done? And that’s really, you know, part of our core platform is really that multi-agency, you know, really the single view of the resident, but really the single set of consistent experiences with the resident, where they can come in and get multiple things done. And that’s where, you know, in the process of a payment, we can then say, hey, knowing what we know about you, well, if you’re registering this car, why not also be able to register this other vehicle as well while you’re here, or pay other obligations, and show you, like, hey, here’s everything that you’ve got with us, and also celebrate completion: you know, hey, you are good, you’re good for like 90 days, so go enjoy your life, you know, and then come back. So those are the types of things that we’re seeing AI just be able to connect dots across multiple data sets, but really being able to focus on, hey, how does the resident see the world across agencies, to make those experiences better.

I mean, connecting the dots is, I mean, as citizens, all three of us, right, like, that’s a big one, right, for any of the transactions we do with our state government. So glad to hear that that’s a powerful use for AI. All right. As we wrap, I want to ask each of you, we’ve covered a lot of things and probably didn’t dive deep enough into any one of them and could talk a lot more about a couple of these, but I’d like to ask each of you sort of what is one AI capability that you are expecting to really change things in the next, say, 12 months for agencies, and maybe why. So, let’s see if you both have the same thing, but Christian, you can go first, and then Adam, you’ll have to think about yours if he takes yours.

Automation. You know, so on the internal space, instead of having these sidecar AI capabilities, it’s more rip and replace, like, here’s the old way that we did things and we’re not doing it this way anymore. We’re creating a much more automated approach to doing this. So, through, you know, agentic AI, or, you know, whatever the buzzword is, as Adam was talking about, there’s a lot of buzzwords flying around, but I think automation for us is what we’re really looking at as being significant. It’s going to be significant in the state of Utah. We have our own version of DOGE now called GRIT. Thankfully it was not really driven from the top down necessarily, where cuts were prescribed, but rather agencies were asked for their input on what can we do, agency by agency, to operate more efficiently, and AI is going to be a key driver for a large percentage of the initiatives that are coming in from the various state agencies, because they are all looking at ways that they can automate their business process. So I think for us automation is going to be really, really big, in addition to the citizen-facing stuff.
All right, Adam.

Automation, love it, love it all. I think I see a real bending of the curve of how we use data. So from a back office perspective, I really see a powerful change of behavior, which is tough to pull off, but I think consumer experiences that we’re having with AI are going to change expectations also, even in day-to-day back office. I can’t tell you how many products I’ve built that deliver reports to people, and reports, you know, the usage curve is, hey, I thought I needed this report every week, and they look at it for a month and then the trend just goes down and down, and we generate the myriad of reports. I think that behavior is going to change, and I think the behavior is going to be, look, tell me the answer I’m looking for. I’m not going to call for a report that I’m going to scan and have to go find the answer on a weekly basis or on some one-off basis. Just going to ask a question of the data and let it come back to me in more of that agentic type of way. And I think that’s going to become the norm and the expectation, and we’re very excited to be on the forefront of that. And I think for residents, I think on the resident side, I don’t think there’s going to be as much patience for this wall of links and every agency has their front door and, hey, come get to know us. I think there’s just going to be a lot more experience that says, “How do I get this thing done? I’m looking for this type of permit,” and be able to surface back up: here’s a warm handoff to how you go and make that happen. It may involve one agency, it might involve multiple agencies, but I think the expectations and the behaviors are going to change pretty quickly. I mean, COVID digitized everybody, and so it kind of laid the foundation for, and the expectation for, really what I envision as really full digitization of almost everything that we do, even those things we go into office for. So, starting that with that question of how do I, but then the expectation is going to be, and certainly I can do that right here, right now, with this same interface. And so I think those are two major arcs of transformation that I’m seeing hit pretty quickly here.

Oh, that’s great. I love that you both were able to share different arcs, but obviously they complement each other. So I’d like to thank both of you for your time on this webinar, sharing your expertise and your insight. We’re at time. Like I said, I know we could dig into a lot more, but I think we touched on a lot of great topics and some good guidance and advice as other agencies and states are looking at their pilots and how they move forward. So on behalf of MeriTalk, I’d like to thank both of you for your participation. I’d like to thank PayIt for supporting these dialogues and discussions, and I’d like to thank the audience who’s listening for their time, and hope everyone has a great day.