---
id: "workshop:scaling-ai-adoption-in-the-public-sector-lessons-from-utah-s-journey-from-pilot-to-implementation"
type: "InnovateUS workshop"
title: "Scaling AI Adoption in the Public Sector: Lessons from Utah’s Journey from Pilot to Implementation"
url: "https://innovate-us.org/scaling-ai-adoption-in-the-public-sector-lessons-from-utah-s-journey-from-pilot-to-implementation"
date: "2025-01-14"
duration: "60 minutes"
language: "English"
instructors: ["Christian Napier","Jake Windley","Carter Saar"]
recording: "https://www.youtube.com/watch?v=WLlYO13upOY"
transcript: "automatic, from the recording"
source: "InnovateUS (innovate-us.org)"
licence: "CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org"
---

# Scaling AI Adoption in the Public Sector: Lessons from Utah’s Journey from Pilot to Implementation

InnovateUS workshop · 14 January 2025 · 60 minutes

How can states move from piloting AI tools to successfully scaling their implementation across departments? In this workshop, Christian Napier shares his experience transitioning from an AI pilot program to full-scale adoption, offering insights into the strategies and tools that made the process successful. With a focus on practical solutions, this session will highlight Utah’s approach to general workforce training, customized support for power users, and the development of tailored workshops for key stakeholders such as adjudicators and procurement professionals.

Utah’s experience underscores the importance of partnerships, such as its collaboration with Google, to support its rollout of AI tools like Gemini. Participants will learn how structured environments, targeted training programs, and stakeholder engagement have driven innovation and ensured alignment with public sector priorities.

Whether you’re in IT, public policy, or workforce development, this workshop will provide a roadmap for states and organizations looking to scale AI responsibly, leveraging training, collaboration, and feedback to maximize impact.

## Learning goals

- Explore Utah’s strategies for transitioning from AI pilot programs to full implementation

- Learn about the role of general workforce training and tailored workshops in preparing teams for AI adoption

- Understand how Utah utilized partnerships to enhance its AI implementation efforts

- Gain insights into stakeholder engagement and effective communication strategies for scaling AI within the public sector

## Speakers

- **Christian Napier**, Director of AI, Division of Technology Services, State of Utah
- **Jake Windley**, Artificial Intelligence Analyst, State of Utah
- **Carter Saar**, AI Engineer, State of Utah

Recording: https://www.youtube.com/watch?v=WLlYO13upOY

## Transcript

Automatic transcript of the recording: names and terms can be misheard, and the times can be a few seconds off. Each time links to that moment of the video. The transcript alone: https://innovate-us.org/scaling-ai-adoption-in-the-public-sector-lessons-from-utah-s-journey-from-pilot-to-implementation.txt (plain text) and https://innovate-us.org/scaling-ai-adoption-in-the-public-sector-lessons-from-utah-s-journey-from-pilot-to-implementation.json (every segment with its start and end second).

**Christian Napier**

[0:00](https://youtu.be/WLlYO13upOY?t=0) It's a pleasure to be here. We're honored to participate in this workshop and thankful to InnovateUS for allowing us to share our story. Before we get into that story, I'll say that we're in the process of incorporating the InnovateUS curriculum into our LMS at the state. So we've completed our testing in our sandbox environment and looking to roll out production over the next few weeks. So I'm super excited about that. Basically, what we're going to do in our presentation today is we're going to talk about some of the foundational aspects to this Gemini for Google Workspace pilot that we've been conducting since June of 2024. And then we'll share some results that we've had to date. And then we'll conclude the presentation by discussing our production rollout.

[1:02](https://youtu.be/WLlYO13upOY?t=62) And I will have our analyst tackle the first two items. But before I hand it over to Carter Saar, one of our analysts, to talk about foundations, let me just give a little bit of background on the state of Utah. Within the state of Utah, AI, particularly generative AI, is bifurcated amongst two agencies. One is the Department of Commerce, who oversees the Office of AI Policy. This is an office that is dedicated to generating AI policy for the state, looking to balance innovation and regulation in the state. So they're very heavily involved in legislative initiatives and the regulatory framework for AI in the state. Then you have the Division of Technology Services and AI within the Division of Technology Services. DTS, or the Division of Technology Services, is a centralized IT shop. We service IT. We provide IT services for all state executive branch agencies. And our mandate in our office is to oversee the rollout of AI solutions within these state agencies. And that's the area where we are. This project, the Gemini for Google Workspace project, is the office productivity element of the rollout of AI to state agencies. And so we're very happy to speak to that. We are a Google shop in Utah, so we use Gemini for Google Workspace.

[2:57](https://youtu.be/WLlYO13upOY?t=177) We're not a Microsoft 365 shop. That being said, I think the lessons that we've learned in our rollout apply irrespective of the office productivity solution that you implement. I think that whether you're using Google Gemini or Microsoft 365, I think the lessons are relevant. And with that being said, I'm going to now hand the time over to Carter Saar, one of our AI analysts, to give an overview of this Gemini for Google Workspace project, what it is and what we are

**Carter Saar**

[3:34](https://youtu.be/WLlYO13upOY?t=214) looking to accomplish. So Carter, over to you. Thanks, Christian. So hello, everyone. My name is Carter Saar. I'm an AI analyst for the state of Utah. Happy to be here today to discuss our Gemini for Google Workspace pilot. So starting off with our pilot objective. So the main goal for our pilot is to determine how a Gen AI tool like Gemini for Google Workspace can enhance employee productivity within the state of Utah. But throughout this pilot, we want to explore the potential benefits of Gemini, see how the tool can fit in with the work that we do, and then identify a path forward for overall statewide adoption. So going into our pilot approach, the first thing that we did is DTS, which is our division at the state, procured 257 licenses for a year. Then we divide up these licenses among the various Utah state agencies, and we do this based off of the size of the agency, as well as overall interest from those agencies. So throughout the pilot, we made sure to provide continuous support. So similar to the type of support that we would do in a full production rollout, this was a way for us to build trust with these agencies throughout the pilot and the exploration of new AI tools, potentially for the first time with some of these folks. Since we didn't want to just throw them off the deep end, our team provided support through a variety of methods, some of those being, you know, trainings, lunch and learn sessions, and then user groups. In addition to providing this overall support, our AI team managed license allocation, we did user surveys and interviews, and then we briefed our DTS leadership team on the progress throughout the entire pilot.

[5:16](https://youtu.be/WLlYO13upOY?t=316) So here's just an overall visual of our pilot timeline. You can see this started back in May of 2024. This is where we were focused on launching the pilot. So we're trying to get everything prepared for a successful launch. We defined the goals of our pilot, procured and allocated licenses, defined our communications plan, and basically did everything to support a successful launch. Then we moved on to the familiarization phase, which started in July of 2024. That was focused on getting our users familiar with these tools. We're trying to deliver on that support plan we talked about in this phase as well, and getting the users to really test out the technology for their particular use cases. From there, in the fall, we moved on to the assessment phase. And this is where our goal was to find ways to quantify the usage and identify the use cases of the tool, while also polishing up some of our internal processes, like while also polishing up some of our internal processes, like updating our trainings, refining documentation, doing a security review, and also suggesting some revisions to our state AI policy. And then finally, that puts us where we are today in our deployment and retrospective phase, which began in January of this year. Our team is focused on effective deployment of this tool. So this is where we're engaging with the state agency leaders, following on from the assessment phase, and basically are offering them an opt-in option to deploy Gemini across their teams.

[6:43](https://youtu.be/WLlYO13upOY?t=403) With this phase currently ongoing, we're focused on continuing our support and ensuring a successful launch of the Gemini technology to the broader, you know, government market within our state. And then this is a chart just showing a visual of the teams that were involved with our pilot. So in the center in the green are all the teams within our division. So you can see we have, you know, our leadership team providing vision and strategy. We have our IT directors as agency liaisons within the different government agencies. We have our PIO team kind of leading our communications. We have a Google workspace team that's helping, you know, be an admin for all the Google suite across the state. We have our finance and admin team helping with the procurement and license management. And then finally in the middle, you can see we have our own AI team, which is kind of serving as, you know, project management of a lot of these different teams, making sure that we're doing things effectively and efficiently within the DTS, like, broader organization. And then we're also, you know, doing a lot of communication directly with Google and our folks over there and conveying the important bits back to the agencies and, you know, vice versa throughout the pilot. So with that, I'll hand it over to my colleague, Jacob Winley, to

**Jake Windley**

[8:00](https://youtu.be/WLlYO13upOY?t=480) discuss some of the results. Thanks, Carter. I'm Jake Winley, also an AI analyst here at the state of Utah. And we're just going to go through some of the results based off of numbers we were able to pull from our Google workspace team and then also from some surveys. So just basic usage level here as of December 11th. So there were four breakdowns of different users, high, medium, low, and none. And the way those categories were determined was just by the number of prompts in a four-week period. So for high users, it's about 10 or more prompts a week, and then it kind of tapers off from there. It is worthy to note that, like it says on the slide, the usage of Notebook LLM and Google Chat was not included in these statistics, just based off of what we were able to pull. But as you can see, you know, over a quarter of users that were part of this pilot were in the high category, and a relatively low percentage of a little over 13 percent were not using it at all over this four-week period.

[9:22](https://youtu.be/WLlYO13upOY?t=562) So just breaking it down by volume, just to get a better grasp of how many prompts were actually being input by the users. For a four-week breakdown, we had over 8,000 total prompts, so a little over 34 per user. And the high users were categorized as 40-plus prompts over the four-week period, but the average was over 100 for those high-level users. So the high-level users were putting in over five prompts a day into one of the various Google Workspace tools, and we're able to later, we'll see, we were able to take a survey of those high-level users to see kind of what they were using it for and how effective they felt it was. But overall, you know, 34 and a half prompts per user. The tools are being used pretty regularly on a weekly basis and even daily basis, which allows us to get a good picture of how effective or how impactful these tools have been on our Workspace users. So the majority of users who interacted with Gemini did it via just the Gemini chat, which is just gemini.google.com, which is your typical chat GPT-type interface where you can just ask questions, generate images, input documents or files, and get responses from Gemini. And then the next highest was Docs, where in Docs, there's a lot of interesting features to the Gemini. You can use it to summarize text, you can use it to generate text, you can use it to generate outlines, generate tables, a lot of different use cases for Docs. And then Gmail being the third largest

[11:26](https://youtu.be/WLlYO13upOY?t=686) subset here, where Gmail, you can use Gemini to summarize emails, to prioritize emails, but there's still limited functionality. And for the usages here, we think that the functionality of Gemini in each of these Workspace tools is what really determines the usage. For Docs, the functionality of Gemini is definitely the highest in terms of integration into the Workspace tool. For Docs, in terms of what you can do and what you can modify, whereas something like Sheets, you can generate formulas, it can help you generate tables, but it's pretty limited in what it can do for you as of now. There's not a lot of functionality for Gemini to alter your spreadsheet or create a complex spreadsheet. But those are features that we're looking to see this coming year and would potentially boost interaction in those other Workspace tools.

[12:28](https://youtu.be/WLlYO13upOY?t=748) Chat with Gemini, that one is just really kind of the sandbox where you can go in and ask whatever questions you have, whether it's coding or brainstorming or other use cases that you have. We'll kind of see the results of how the chat was utilized, specifically by the high level users. So, as seen previously, Docs and Chat with Gemini kind of dominated the app usage. And again, this boils down to just how useful Gemini is within these different apps. And I'd imagine as Gemini is improved and there's more integrations to specific parts of the different Google Workspace tools like Sheets and Slides, there'll be a more broad number of users using those. However, it is notable there are some employees or users who maybe they don't use Sheets at all or maybe they don't use Slides at all. And so, Docs being one of the most dominant apps in the Google Workspace suite, it makes sense that it would be up there. And then Chat is just something people are familiar with in terms of interacting with the large language model.

[13:54](https://youtu.be/WLlYO13upOY?t=834) And so, I think that's another natural way for people to utilize Gemini. So, this data here now is taken from a survey of the high level users asking what the main usages they had of the Gemini for Google Workspace, whether that's with Docs or with the Chat with Gemini. And there's a pretty good distribution of responses. They could just choose any that applied out of these categories. And the average user chose over five responses. So, that shows that users are really getting a broad use out of the Chat with Gemini or the Gemini for Google Workspace integration into Docs or Slides or Sheets to accomplish a lot of different tasks. And I think this really speaks to the versatility of these generative AI tools and how how they can impact and influence the efficiency and efficacy of an employee in a lot of different areas. It's not just for programmers or, you know, just for someone who heavily uses docs.

[15:16](https://youtu.be/WLlYO13upOY?t=916) These applications, you know, pretty much anyone is going to need to do brainstorming at some point or note-taking or learning. And GemIIni can help users with all of these, and the high-level users who are inputting five plus prompts a day, we're using it for many of these different categories. For me, I really find it helpful for brainstorming. I use it for coding a lot and then also just kind of generating content or outlines to help me, you know, tying into brainstorming to help me think through problems or think through solutions to questions I have or tasks I'm trying to accomplish. It's just kind of a campaign you can bounce ideas off of, you know, you can edit documents, you can have it help you with specific language. So it's just really a lot of use cases and a lot of really helpful ways that even just the gemini.google.com can be utilized in state agencies and other scenarios.

[16:30](https://youtu.be/WLlYO13upOY?t=990) So this was another question on the survey, another couple questions where we asked how much time was saved per week and then what the quality of improvement was to the work where zero would be the no improvement and five would be an immense amount of improvement. So for the average time saved per week, the majority of respondents were in the one to two hour range or one to five hour range really, which is still significant in terms of time savings. Personally, I was part of the six to eight hours and that's mainly because, especially for something like code generation, I can come up with an idea or framework of what I want some code to do. I can tell Gemini how I want it to work and it will output some code for me that I can then go and test and revise and polish so that I can focus on the actual functionality of the code and not the intricacies of how it's being written or how I don't have to jump on Google and figure out some syntax for a specific Python function I'm trying to call or library I'm trying to use.

[17:51](https://youtu.be/WLlYO13upOY?t=1071) So Gemini is really helpful there and I think it's helpful even if you have a tool like GitHub Copilot. I've found that using Gemini for code generation for larger code generation tasks is often more accurate and more helpful than GitHub Copilot, while I still think it's valuable to have both tools at hand. It's also really easy if you have a document or a big long report that you're trying to ingest and maybe you don't need to know all the details but you just kind of want to get the gist of what the report is saying, you can throw it into Gemini, have it summarize it. If you use NotebookLM, you can even have it create things like a table of contents or a study sheet or just summarize it in a page. And so it really makes large documents and large amounts of data digestible in a relatively short amount of time. Now it's important to note that with these tools you should always verify information that they're giving you, but with something like a direct summarization or categorization of a document, you can get fairly accurate results and very useful results in my experience. And then lastly, just touching on brainstorming again, helping with brainstorming, just thinking through problems, coming up with frameworks for solutions, coming up with a plan, even something as simple as laying out a daily schedule to be able to accomplish a certain goal, Gemini can be used for all these different tasks.

[19:40](https://youtu.be/WLlYO13upOY?t=1180) And I'd say for me, brainstorming is probably my primary use of Gemini.google.com. And then in terms of quality of improvement, I think with this tool, it just allows a higher level of efficiency and a focus on the quality of work that you're doing because it can remove some of the monotony of certain tasks. Going back to the coding example, it can remove a lot of kind of the syntax-y and cumbersome parts of coding and help you really focus on the architecture of the product you're building or the development stack that you're working on to help you focus on the product and what the code is accomplishing rather than how to write specifics of it. And as you can see, the majority of people felt that the quality improvement was on the high range. So with that, we'll turn over to Christian for talking about how we rolled it out and how we were able to communicate that to state agencies.

**Christian Napier**

[20:55](https://youtu.be/WLlYO13upOY?t=1255) All right. Well, thank you, Jake and Carter, for giving us those overviews. So we've been in that pilot phase a long time, and then something happened in I think it was late September that caused us a lot of excitement. Jake talked about the standalone chat with Gemini app. It's also called the Gemini chat bot or the Gemini app. As he mentioned, it's like chat GPT. You just go to gemini.google.com and you ask questions and it gives you answers, right? The difference being that unlike just using regular off-the-shelf or publicly available Gemini or chat GPT or Claude or these other solutions, the Gemini application is ring-fenced in our environment. So that means that none of the prompts or the data sources that we use or the responses that are generated by Gemini are used to train Google's models. They're not accessible by Google engineers. So it's in a very controlled space. The Gemini app is available in browser on the desktop and also there's a mobile app for Gemini. The Gemini app does require a Google login. So unlike, for example, chat GPT, you can go on there and actually put in a few prompts without even having to create an OpenAI account. But for using Gemini, you have to have a login.

[22:46](https://youtu.be/WLlYO13upOY?t=1366) Now, of course, you could log in with your own personal account and there would be different terms and conditions. But if you log in with a utah.gov account, then you're part of our environment and so you can use it in a secure way. The game changer for Gemini happened in September when Google announced that this standalone Gemini chatbot would be made available throughout the state for any state employee that has a Google Workspace license at no additional cost. We wouldn't have to pay any more money. We wouldn't have to pay any license cost for our 23,000 employees to be able to have access to generative AI capabilities. And as a result of that announcement, we thought, well, do we want to wait until we conclude this Gemini for Google Workspace pilot to roll it out or do we want to go ahead and make this capability available to our state employees? Because as the surveys have shown, more than half of state government workers are using generative AI as part of their work, many of them just using the freely available tools. So could we bring them into a more secure environment and give them those benefits? So we met with our leadership and we presented this opportunity and they enthusiastically agreed, let's roll this out. So we started coordinating that effort, working closely with Google, working with our own IT Google Workspace team and other parties, our security team, for example.

[24:43](https://youtu.be/WLlYO13upOY?t=1483) We laid out a plan, basically a three-month plan, a very aggressive rollout plan. Starting in October, we started preparing the communications plan. How are we going to train people on using the tools? And then we had our security team under the direction of our chief information security officer do an in-depth assessment of the Gemini standalone chatbot. Could they do things that would break it? Could they do prompt injection? What are the different things that they could do? Certainly there were things they could do to get inaccurate responses, but were there some security loopholes or threats? They conducted this assessment, concluded it in November, determined that the risk of rolling this tool out was minimal. So we got to go ahead. And we also, in preparation for the launches, we decided to update our responsible AI use training in the state of Utah. In July of 2024, we launched a small responsible AI use training course that all state employees have to take.

[26:01](https://youtu.be/WLlYO13upOY?t=1561) And we updated that course and we actually rolled it into our overall security awareness training, which meant it was a lot easier for all state employees to take the training because instead of having to take two trainings, one security awareness and one responsible AI, it's rolled into one comprehensive course starting this year. We launched that on Monday of this year. Then in December, we worked with all of the IT directors. Keep in mind we're a centralized IT shop, but every agency has an IT director. They are employees of the division of technology services, but they oversee the teams, the IT teams that service those agencies. So we worked through those IT directors to liaise and communicate with their agencies, let them know this is coming down the road. They'd actually had those communications back in October and November, letting them know that this was coming. But we created formalized communications for the IT directors to then liaise with their counterparts in the agencies at the beginning of December. We decided to launch, do a soft launch of the tool on the 16th of December. That was a Monday. Agencies were given opportunities to opt in.

[27:32](https://youtu.be/WLlYO13upOY?t=1652) We thought if an agency, for whatever reason, maybe they're highly regulated, they've got some security concerns, every agency has its own security officer, you know, if there were some concerns that they wanted to have a little bit more time to manage communications with employees in their agency, by all means, they could do that. So agencies were asked, do you want to opt in to this December date? And out of the 29 agencies, about 22 agencies opted in. And the remaining agencies said, well, we want to have a little bit more time to manage our communications. So the remaining agencies are going to be opting in. Some have already opted in. The remaining will be opting in either later this month or in February as they get that around. And there were some who said, well, I've got a targeted group of employees that I don't want to have access because they have access to highly sensitive information or something like this. So we sent out welcome emails to the employees.

[28:35](https://youtu.be/WLlYO13upOY?t=1715) Actually, the agency sent those, we crafted the language of those emails and then sent those to the directors, who then sent those to the executive directors of the agencies. And in a parallel effort, starting from September, October on, we started to do a transition of the management of this from ourselves, the DTS project managers to the Google workspace team. And this was in parallel with the Google workspace team also restructuring itself. The way the Google workspace team operated in Utah, historically, it was part of the desktop support team. And so it was very much a break fix or a fix break, break fix kind of solution. And DTS thought, you know, we need to have more of a product management, product ownership type of approach to managing our office productivity suite instead of a technical desktop support approach. So in November of 2024, they hired a product manager to oversee the Google workspace team. And we've been working very closely with her and the Google workspace team to roll this out together and eventually transition the ownership of this project from the AI team to the product management team who are going to be operating this thing for the foreseeable future.

[30:08](https://youtu.be/WLlYO13upOY?t=1808) I think that's an important lesson to learn. So that happened in December. We did the launch. This month, we are holding a launch event for all of the agencies who have opted in and a series of lunch and learn training sessions. So we're going to be providing ongoing support, doing a lot of knowledge sharing. We've created user groups and ways for people to submit through forms like, hey, here's a great use case. Here's a great use for Google Workspace, and we'll continue to monitor and assess. I will say, Jake presented some statistics on usage. The reporting capabilities that Google has made available for Gemini are quite limited. We can only report on activity over the past 28 days. That activity doesn't include all of the applications.

[31:08](https://youtu.be/WLlYO13upOY?t=1868) So for example, Notebook LM or integration with chat is not available in the reporting. We currently don't have any way of reporting usage statewide, only for the pilot group. And so there's some things that we're working with our friends at Google. We need some additional capability to see what's going on there. But the one thing I didn't put here on this slide, this was the rollout for the Gemini production app. The Gemini for Google Workspace, which includes the integration with Docs and Gmail and Sheets and Slides, very similar to Copilot for Microsoft 365, that's a paid service, and we're working on the transition from a pilot to production. The number one thing we're working on right now is negotiating pricing with our friends at Google to see how much this is going to cost to roll it out.

[32:10](https://youtu.be/WLlYO13upOY?t=1930) And also, because agencies pay the costs of IT, the agencies would be responsible for paying for any licenses they want. We don't expect that agencies will provide the paid version of Gemini for Google Workspace for all their employees. It will be a subset. One of the things that we learned by conducting this pilot is that not everybody can benefit the same way from using these tools. Some will benefit more than others, and some will take to using those tools more quickly than others. And so for the profiles of users who could gain significantly in their efficiency and quality, we expect that there's going to be... We know that there's significant interest from the agency side because I have agencies contact me all the time saying, hey, when is it going to be available for me to get more licenses?

[33:12](https://youtu.be/WLlYO13upOY?t=1992) And so that's basically, in a nutshell, what we did to roll out Gemini to production. So currently, Gemini app is available to roughly, I think, 15,000 state employees who now have access to it. The remaining, we've got some big agencies like Health and Human Services and Workforce Services who are still working on their communications, and once those come on board, then the vast majority of our 22,000, 23,000 employees will be using the free version of Gemini. And agencies, within the next few weeks, will have the option to purchase additional paid Gemini for Google Workspace licenses, should they so desire. So that's a lot. We've crammed into 40 minutes of presentation. Very grateful for you patiently letting us share a bit of our story.

[34:14](https://youtu.be/WLlYO13upOY?t=2054) We're super excited about Gemini, its capabilities. We're also realistic about the rates of adoption and the efficiency and productivity gains that we're going to see moving forward. So it is still a work in progress, and happy to... That concludes our presentation. We're happy to answer any questions that the group might have, and, Agueda, we turn it back over to you.

* * *

Original page: https://innovate-us.org/scaling-ai-adoption-in-the-public-sector-lessons-from-utah-s-journey-from-pilot-to-implementation · Licence: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org