![Podcast microphone against a blank background](/sites/g/files/omnuum10826/files/styles/hwp_21_9__1920x825/public/datasmart/files/podcast_microphone.jpg?itok=3Q9qYvou) # How GenAI Can Actually Boost Public Sector Creativity Episode Eighty-Two October 29, 2025 [ Betsy Gardner ](/people/betsy-gardner) This episode features Mike Sarasti, former Chief Innovation Officer and Director of Innovation and Technology in Miami and a leading advocate for government transformation, in conversation with host Stephen Goldsmith. They unpack how GenAI and rapid process mapping are revolutionizing public sector efficiency, not by shaving seconds off legacy workflows, but by making space for human creativity and curiosity. Mike shares real-world examples and explains how city leaders can democratize AI tools and clear bureaucratic tedium while guarding against hype and automation overreach. Embed *Listen here, or wherever you get your podcasts. The following is a transcript of their conversation.* **Stephen Goldsmith:** This is Stephen Goldsmith, Professor of Urban Policy at the Bloomberg Center for Cities at Harvard University, with another one of our podcasts. Welcome back. Thanks for joining us again. Today, we have an old friend in the world of data and analytics and cities, Mike Sarasti, who is currently an entrepreneur. He calls himself a musician, and I call him an entrepreneur working with cities, but I know him from his work in Miami-Dade and City of Miami. Hi, Mike, nice to have you back. **Mike Sarasti:** How are you doing, good sir? Good to see you. This is fun. **Stephen Goldsmith:** You have so many different personalities. **Mike Sarasti:** \[laughing\] Try to keep everybody on their toes. **Stephen Goldsmith:** This is a podcast so people can't really see you, but I do see your guitar over your right shoulder. **Mike Sarasti:** Yup my guitar, I'm over here in my studio space, yeah, which I've been spending a lot of time in after my city life, which is great. **Stephen Goldsmith:** Well, I want to go back to your city life before we start, but while we're talking about guitars, okay… **Mike Sarasti:** Yeah? **Stephen Goldsmith:** What do you think the role of creativity is to AI business process transformation? **Mike Sarasti:** Oof, we're going right into it. Yeah, I'm constantly having this conversation in my AI chat group as they're sending me AI-generated songs and telling me how they made things. So while I appreciate that more accessible, this idea that creativity and tools that were historically really expensive, like having to go into a studio before was quite inaccessible for a lot of musicians, and I do enjoy the idea that people can go in and do more now just from the tools that are available on their desktop. I think it is a real...it's really important for us to continue to celebrate craftsmanship in these things that we are doing. And I think that applies to if we suck out the creativity of government work, it's something I used to talk about a lot, I think fundamentally we're all creative beings and problem-solving is an act of creation. So even as the AI overlords give us a lot of these tools that say, "Oh, we're just going to do this for you," preserving that weight on craft and the spark that comes from creation when you're solving a problem from a different way is really important to preserve. **Stephen Goldsmith:** Before we go back to the beginning and start the podcast the way we're supposed to, I have one more question based on the guitar. And I know we gave you questions in advance. None of these questions did we give you in advance, so just to- **Mike Sarasti:** Let's do it. **Stephen Goldsmith:** Back when I was the mayor of Indianapolis, I read this, as I recollect anyway, Wall Street Journal article about Southwest Airlines, and the story was essentially that in order to turn around planes at the gate more quickly, they didn't go look at United or American. They went to look at pit stops at NASCAR or Indy 500 to see how it is they change the tires and oil in 29 seconds, right? You are an expert. We're going to talk about the connection between AI and business process for engineering, but in terms of creativity, the guitar in the background, I think too often in government the definition of efficiency is making an old process slightly less worse as contrasted to breakthrough, imaginative, creative thinking that lets you understand and imagine a whole new process. How do we generate the creativity to think about the new processes as contrasted to just making the old processes slightly more efficient? **Mike Sarasti:** Yeah. I mean, I think so much comes from play, right? I mean, it's called playing music, playing a guitar. That is not accidental. And I think way too often in the process efficiency space, it's all like, "Let's just make it faster, shorter, faster, anything you can do, and move on." The part that does give me some excitement now, if we can clear out some of the tedium and create more space for play, more space for actual human collaboration because you're offloading all of these tedious things, a lot of great things are going to come out of that, right? If it's actually clearing out space for human interaction, for the jamming part of things, and you're set up with these great tools, that is the exciting part. Shaving off a few seconds is a little bit of a dopamine hit off a process, but the breakthrough stuff does come out when you have some freedom to do it, and historically, maybe you had to take a whole retreat and people are mad because they've got a bunch of work on their desk. Well, what if they're not mad that they have a bunch of work on your desk because you've been able to clear some of that stuff out? Now they can really immerse themselves in the play part and the discovery that comes from the stuff humans are good at. Tell the stories about what we're doing, and what does that turn into, and the sharing. So hopefully that's the direction we're moving where we have more time to do that. I worry that we're just, if we're not careful, we're just like, "More optimization, more optimization." I'd say, "All right, well, let's shift from that. Let's make it a little bit more fun." **Stephen Goldsmith:** So now that you've answered the podcast questions, let's go back to the beginning and start. So who are you? Give us a couple minutes about your Miami, Miami-Dade experience, and then we'll talk a little bit about your current work. **Mike Sarasti:** Great. So I've been in government, in or around government, for about 20 years now. I started at Miami-Dade County, did about 10 years there. I came in initially in a survey research capacity because I had done that towards the end of my grad school, but it was in the department that was running 311 and miamidade.gov. So a lot of UX, customer focus. And towards the end of that time period, started getting really involved, the open data movement was happening and immersed myself in that world. That was my entry point into data. We were also building a 311 CRM internally. So I was around a lot of technology teams, building, customer-centric focus, not in an IT department, but definitely a technology-enabled team thinking about user experience and customer interactions. And then towards the end of that, I got asked to join city of Miami as the city's first chief innovation officer, which was really cool. I love saying that it wasn't important that it was me, but it was important that the city had designated a role for innovation. About two years into that, they said, "Well, do you want to do the IT department director role as well?" So I got a chief innovation and information officer added on, and through that got to work on everything under the sun around technology from data to obviously the building of tech and the website, and did a lot of work around that, but got to keep the innovation portion of it, which was largely grounded in a lot of analog things. We lifted Denver's Peak Academy program and set up an instance of the program at the city of Miami, which was a crash course in a lot of the process work and tools and techniques to do process mapping and process re-engineering, and always tried to make a connection to how we were building better technology and better systems through the lens of process. And all of those things together brought me to what I've been doing the last three years, which is using some of the newer tools at our disposal, including AI, to do rapid process mapping and discovery. So all the things that we were doing that might've taken us weeks and years earlier, trying to leverage the new tech to do it in hours and days, hopefully, in some cases. **Stephen Goldsmith:** Mike, let's pretend for a second I'm the chief administrative officer of a major jurisdiction. Give me your rvrwrk pitch. What should I be doing that would allow me to transform the way I engage my residents? What is it about AI and business process that's particularly appealing? **Mike Sarasti:** Well, I'll connect the origin story a little bit to what I mentioned, which is a decade or so ago, along with some of our mutual friends, we were thinking a lot about open data and the concept of machine-readable process came up. We were trying to get people to release data that are like, "Oh, you already had the data. It's in the PDFs." And the shift was like, "All right, no, that's fine that you're publishing a big, giant PDF with a bunch of tables and numbers in it, but how do we start to make this data machine-readable so that we can do a lot more stuff with it and make it open and available?" That was great. Somewhere along that journey, I started to get frustrated because, yeah, we had the outputs of the data, but we didn't really have all the information about how things worked. So the shift for me started going from machine-readable data to machine-readable process. We were doing these Post-it Note-based process maps. I'm starting to see these Post-it Notes up on a wall, and those started to look like cells on a spreadsheet for me. Each step has all this metadata around it, et cetera, et cetera. So how things work, I think, is fundamental to being able to fix the process, and a lot of times we skip that, not because we don't want to do it, because it's been very, very tedious. So what we did with rvrwrk was we became obsessed about how we make that discovery, and being able to map out all those processes, not tedious. And it turns out that a lot of the things that GenAI is good at, pattern recognition among them, is that we can actually source processes very, very quickly. So whereas it used to take you months, sometimes years to map out a series of processes, in many cases we're doing that in days. Where does that help you? Obviously the simple is just have SOPs documented. That has all kinds of benefit when you've got turnover, someone's leaving, someone's being onboarded. But it can also be very, very powerful if you're mixing it up from a performance management perspective with other data points because if you are also taking those processes and you are creating a structured version of that, it turns out that it makes the data a lot richer. **Stephen Goldsmith:** Take a breath here. You said a lot of stuff. Without naming a client, give us an example of where you've done this, permitting or just name the subject where you've process mapped with this much velocity. **Mike Sarasti:** Yeah, Finance. We have a big place that has mapped out...we were able just recently to map out a hundred processes in a little over 30 days. If you would've asked me to map out a hundred processes as a CIO, I would've told you that we've got a three-year project on our hands, and we did it in a little over 30 days. What did that do? That particular department was in the middle of a software implementation. So that is going to help with their transition, but in addition to that, we're doing conversational... The way we do it is we do conversational interviews with individual employees. This is not a select group of people that we're bringing into a room and the loudest person in the room gets to determine the outcome. We interview everybody in the department because everyone gets a conversational AI interview. We built a conversational agent that asks you about your process. So you on your own time, Steve, can make a phone call, you describe a particular process. Everybody could do that at the same time if they want. So you could have 100 contributors, 300 contributors. You also get multiple perspectives. So if you've got three contributors on a process, it's not driven by one individual. You can create a composite of that process where it's crowdsourced. So that's a great example. Another example is a municipality that wanted to actually implement more GenAI, but they didn't really know where to start because they didn't have their processes documented. It's a little meta. We're using AI to launch an AI project, but we were able to rapidly source, "All right. Here are the actual steps in the process. Now can we apply a different prompt that analyzes those steps and looks for opportunities to implement AI?" **Stephen Goldsmith:** Let me see if I can make the question more complicated. Your answer was really too good. Let's see if we can make it more obtuse. When I was deputy mayor of New York, licensing and permitting reported to me. It was in my portfolio. **Mike Sarasti:** This is my favorite, one of my favorite topics, Steve. So let's go. I'm ready. **Stephen Goldsmith:** We worked on efficiency, which made each of the, I'll make up the number, 15 agencies that touched a restaurant permit would try to be a little better or a building permit, but of course that wasn't really all that helpful to the poor New Yorker actually applying for the permit. So as you talk about the processes and improving them, let's talk about where there's interconnectivity across the agencies, all of which are using different systems, and how do you process map 10 agencies or six agencies into a combined customer-facing solution? **Mike Sarasti:** Yeah, that's great. It turns out that those connections tend to be a little bit more difficult to get at. Obviously, if you get in with somebody in a room, those things start to emerge. The agent is designed to just be curious and to put you in a position to just tell stories about your work. So what turns out is that individuals are describing a process to you, and that the questions are not as structured as they usually are, they tell you, "Oh yeah, this is how a process usually works, but then I've got to send this document over to Stacy in this department and then Bob's over in this department." As they're in this conversation they’re telling you the software that's problematic and their pain points, you can let them speak freely, and then after the fact in processing that transcript using GenAI, you can establish those relationships. So you might ask a question across 10 transcripts, how many times does Steve show up as a node across these transcripts? Well, it turns out Steve is very important to the process across all these areas. So that's one of the prompt engineering after the fact on the transcript. Now that you've got these transcripts, it turns out you can ask a lot of questions that I hadn't even thought about asking at the beginning. You don't have to think about all the possibilities, just do the interview and now it's like the ability to go back, "I wish I would've asked that question." Well, it turns out that because we've done it this way, you can go back, ask that question, you can establish the nodes, and we're storing and processing the data to put it in a standardized format. So the interview, we actually converted into a JSON file that you can establish those relationships and make some of those connections. **Stephen Goldsmith:** So since you're in the consulting business, how about some free advice? **Mike Sarasti:** Yeah, let's do it. **Stephen Goldsmith:** I've got a group of major cities, like seven or eight of the largest cities in the US, and we're working on modernizing performance management, what you would know as stat management. And one of the goals is to help mid-managers be able to use AI tools to discover process problems or causation problems, something other than just how to more quickly respond to a problem. So how do we partially democratize the use of generative AI to access data? So what should I be telling these cities in terms of data literacy training? Let's say you've done the process mapping, but we want to have more, I love that word, curiosity. We want to have more curiosity by mid-managers about why they're filling the same pothole 10 times. How do we take the next step? **Mike Sarasti:** If we're asking about specific skill sets that I think people should be developing in this GenAI world and leveraging some of this tech and things that are possible today that weren't possible before, we should be getting very good at prompt engineering. And every time I think about, "Is this a good use of GenAI or not," I always parse it into two different categories in the way I think about it. One is that AI in general is very, very good at pattern recognition. So if it's something that is being able to detect a pattern, generally that makes me feel better. If we're in a spot where GenAI is making a bunch of choices for you, it's doing all the writing, it's writing the entirety of the song, it will produce those kinds of results with very, very short prompts. I see these things that are like one sentence, "Oh, it spit it out, this looks great." That universe of AI worries me quite a bit. I could probably do a whole podcast on my critiques of that side of it. But if you understand that it's really good at detecting patterns and you can get really in depth with the prompt, you don't have a one-sentence prompt, you can have a three-page prompt that effectively acts as filtering on large volumes of data, that is something that wasn't available to a manager before. Your ability to go in and have a dialogue with data is something that is quite possible these days. You're not waiting for the database engineer to get freed up, and I think that, as being something new, is a skill set that people should lean into because it's very, very accessible. You can practice on your own with your own stuff. You can take a document that's a PDF and you can have a conversation with it and beef up your ability to prompt it. I think those skills are really, really important and very valuable for a performance management. I did a little experiment in the lead up to this where I took 10 different transcripts on a process, and I hadn't done this before from a performance management lens, but started designing a bunch of metrics based on what employees had said about their business process to start forming something that could be a little bit programmatic. Where are all the instances of rework in this process? Where are all the handoffs, which is probably a good indication where you might have some rework? At what point does it hit the customer? How many times is it coming back? Were there any indications of things like that? That can help you design a good number of metrics and it's not something you need a high degree of technical skill to do. **Stephen Goldsmith:** A couple more questions. One, with respect to these issues you've mentioned that are entangled among agencies and people, what do you want from a leader to set the right stage? Whether it's the CIO or the mayor or the CAO, what structures or what messages would be most empowering? **Mike Sarasti:** I'll start with what I don't want, what I think is dangerous. There are a lot, and I'm not in it anymore, but I can only imagine, and obviously what I've heard, how many AI companies are coming into the mix and saying something like, "Oh, just give us all your data and we're going to have...AI is going to automagically do this, this, and that." I think that's quite dangerous. One, because as most of us in this know, a lot of these data systems weren't...it's a sliver of what's happening. The systems themselves weren't designed to be able to give you all the nuance. The data isn't digitized. I think that's very important to know, which means you're still coming back to humans. The knowledge is still trapped in people's brains. So I would say definitely continue to focus on the human experience. You have new tools at your disposal to make that side of the shop richer. And again, the data you have, it's only...we've been working, Steve, as you know, in our realm, we've been working with what we got. It's usually the data that's easier to get to. Those things tend to float to the top of a performance management program because it's what we've got. And this moment, if anything, is the potential to enrich those datasets with natural language nuance and think about that as structure, I think is very, very, very powerful. That's the part of it that doesn't replace the human experience. It's actually helping amplify the human experience. So in summary, be cautious of the people that are promising you all this stuff that AI can do, and lean into the things that are really, again, going back to some of the things we talked about earlier, more opportunity for human creative stuff. You're going to reduce the tedious stuff that is generally clogging up the system from human creativity and elevating the potential for all these other things to emerge. That, I think, is an exciting stance for leadership to take. **Stephen Goldsmith:** Have you worked with any cities or counties that are concurrently involving their communities in the redesign process? **Mike Sarasti:** No, I think we've had some inquiries, and honestly, maybe this is partially on us because I think there's a push or a pull to do the chatbot, get people involved in the public. I'm not sure that technology is quite there yet for the same reason. The information you need for those things to work well is not often digitized. We haven't gotten to that point yet. So I'm often pushing people back like, "Let's make sure you understand what the organization looks like." But I do think the potential for conversational AI, again, the same way that we might be asking employees to describe their experience, can transform what's happening in terms of customer feedback. We've often had these kind of crude tools that we've had for a while. The survey, just the nature that it was very difficult to comb through open-ended responses, meaning we often weren't doing a lot of open-ended responses, that limitation is not really a problem anymore. You can put people in the stands to tell those stories fully and then use some of this tech to look for patterns in that and shape and develop programs. So I hope we're on our way there. I do think the tools are there, as long as we make sure it's the right ones. **Stephen Goldsmith:** Let me ask you this in closing. I've been around long enough, I'm so old I was at the beginning of the e-government movement. I was at the beginning of the digital analytics center movement in terms of changing the way cities operate, and I feel like this is more powerful than either of those. Our listeners are city and county and state leaders. If we want to accelerate change, take advantage of these tools, we'll put a caveat, understanding the risks and the privacy questions and the algorithmic bias, but if we really want to accelerate transformation, what should we do? **Mike Sarasti:** I think you need to use the tools. That's always the first thing that I say. There are some cities that we see that are just like, "We're going to cut off access to this." I think that's a mistake. I think it's important that we develop a certain amount of intuition using these tools so that you understand the choices that are being made in some of these responses. There's a whole lot to discuss around sycophancy of the prompt and always telling you you're doing great. You can start to pick up on those things only if you're using the tools. Always when you get the first response, I'm like, "Oh my God, that first response was amazing." If you talk to these tools long enough, that starts to deteriorate a little bit and then you're like, "Okay, I know what's happening here." So I think to accelerate change and also to accelerate the kind of change we want to see, we need to start putting these tools in people's hands and encouraging people to not let go of their agency on these things. You're not looking for that first response from this thing. You want people to think about it, again, with curiosity and a process of discovery. **Stephen Goldsmith:** I thought your answers today were great until the last one because I've been looking to my AI for confidence building because it more often complements me than my own colleagues, and I thought that was personal. I thought it was personal. **Mike Sarasti:** \[laughing\] Well depends on which part of the day. Sometimes you just need that. As long as you know that that's what you're going in for, it's all good. **Stephen Goldsmith:** All right. Well, Michael Sarasti, this has been a pleasure. Steve Goldsmith here from the Harvard Bloomberg Center for Cities. You've got a lot of experience and it shows, and I hope you can help your clients provide better quality services for their citizens. Thank you so much for your time. **Mike Sarasti:** Thanks so much for having me, Steve. Great to see you. ## About the Author ### Betsy Gardner ![Headshot of Betsy Gardner](/sites/g/files/omnuum10826/files/styles/hwp_1_1__100x100_scale/public/2025-05/Betsy%20Headshot%20resize.jpg?itok=k2OsSp1g) Betsy Gardner is the editor of Data-Smart City Solutions and the producer of the Data-Smart City Pod. Prior to this, Betsy worked in a variety of roles in higher education, focusing on deconstructing racial and gender inequality through research, writing, and facilitation. She also researched government spending and transparency at the Lincoln Institute of Land Policy. Betsy holds a master’s degree in Urban and Regional Policy from Northeastern University, a bachelor’s degree in Art History from Boston University, and a graduate certificate in Digital Storytelling from the Harvard Extension School. See also:- [ 311 ](/topics/311) - [ Artificial Intelligence ](/topics/artificial-intelligence) - [ Civic Analytics Network ](/topics/civic-analytics-network) - [ Innovation ](/topics/innovation) - [ Open Data ](/topics/open-data) - [ Performance Measurement ](/topics/performance-measurement) Share on:- [ Facebook ](#) - [ Twitter ](#) - [ Linkedin ](#)