Personalizing Government at Scale: Denver's AI Strategy

Host Stephen Goldsmith sits down with Suma Nallapati, Chief AI and Information Officer for the City and County of Denver, to explore how Denver is using generative AI to collapse bureaucracy and make government fundamentally more responsive to residents. Nallapati discusses Denver's Sunny AI platform, why combining the CIO and AI officer roles eliminates unhealthy friction between innovation and caution, and why the real opportunity of GenAI lies in freeing public servants from repetitive tasks so they can focus on the human connection that drew them to public service in the first place. Nallapati emphasizes that AI is a tool in government's toolbox—one that succeeds only when paired with ethical frameworks, transparency, and a relentless focus on resident outcomes rather than technology for its own sake.

Listen here, or wherever you get your podcasts. The following is a transcript of the conversation. 

Stephen Goldsmith:

Welcome back. This is Stephen Goldsmith, professor of urban policy at the Bloomberg Center for Cities at Harvard University, with another episode of our podcast, The Data Smart City Pod. Today we have a powerhouse guest who's driving incredible change in the city and county of Denver around not just the use of AI, but how do you use it responsibly and ethically and how it can change and improve services for more responsive government. So welcome to Chief AI and Information Officer for the city and county of Denver, Suma Nallapati.

Suma Nallapati:

Thank you, Steve. It's a pleasure and honor being on this podcast with you.

Stephen Goldsmith:

So lots of questions for you. I would first start with my bias. I think you have one of the best mayors in the country. Not only how he thinks about governance, but how he thinks about innovation. So he's lucky to have you and you're lucky to have him. So I'll start with my bias here.

Suma Nallapati:

Thank you, Steve. And I agree with you wholeheartedly. He has been one of the best leaders to work for. A sense of urgency for the city and county of Denver and it's residents has been truly inspiring.

Stephen Goldsmith:

Inspiring is the right word for the way he thinks about his job. Before we get to Denver, how did you get there? What was your background? Tell our listeners a little bit about you, please.

Suma Nallapati:

I've been in the technology industry now for 30 years. It was an accidental entry into technology because my background is actually nuclear physics with a specialization in radiation physics and isotope technology. I wanted to be doing more research in nuclear medicines, but when I moved to Denver in '96, there was not a whole lot of scope for that. I was very new to the country, did not know anything about how to find even a job. So I was like, I can't just sit around and wait. My bachelor's was actually in electronics and computer science, so that started my technology journey 30 years ago. Grew through the technology family, if you will. Started as a programmer back in the day, all the way to a technical lead, manager, a director, all the way to being the Chief Technology Officer for the state of Colorado.

And that was a huge leap for me in terms of my leadership, but it was absolutely phenomenal that I got to serve the residents of the state of Colorado through that role. That CTO role became a CIO, and then I went back to the private sector after Hickenlooper's term at the state. And I was the chief digital officer for a couple of companies. And somehow along the way, the appeal of public sector comes back to me. And that's how I came back to the city and county of Denver with Mayor Mike Johnston. So my journey in tech, I would say, is truly for helping the residents, the most vulnerable, through technology. That's what is very, very appealing to me.

Stephen Goldsmith:

Well, we'll leave out the astrophysics part of this podcast and concentrate on the latter. Before we get to what you're doing in Denver specifically, if you'd ask some of the leading AI advocates in US cities, privately, about the obstacles to broadening the use of AI to improve services, a good number of them will say because our CIO is too cautious. So the division between the AI office and the CIO office, on paper, they're partners, but one of them is driving change and the other is more cautious. So just a little bit about, this is kind of a weird job you've got AI and CIO. Whose idea was that? And is it in fact that better way to solve the problem I just mentioned?

Suma Nallapati:

Yeah, that's a great question. I felt strongly that the role needs to combine both the traditional CIO functions as well as this innovation. When it happens in two different disparate groups, there's always this unhealthy friction on what comes first. So, when I talked with the Mayor, he was so enthusiastic about combining both to get the maximum efficiencies. And we always start with the outcome, Steve, and work our way backwards. How does technology align with public sector goals? How do we ensure that government priorities like better services to our residents, helping the most vulnerable in the most transparent and efficient way is the goal. And so the role is in alignment with that. It's not IT for the sake of IT.

Stephen Goldsmith:

And how do you, since you have both roles, how do you think about creating an environment for innovation, but one that is properly constrained by democracy? What are the ethics? How do you think about fairness and choice of technologies, right? How do you think about the tension between innovation and regulation.

Suma Nallapati:

Again, when you start with the goal of how can we best serve our residents, it's very clear. AI is a tool in our toolbox, just like any other technology. Yes, the recursions are happening so much faster with AI. Yes, there's so much unknown for us to explore still, but with the right foundation of tools and policy frameworks, ethical use, data explainability, transparency, all of those are considerations for the use of call it AI, call it automation, call it efficiencies. It's the foundation that's very important. Starting with a very good guardrail policy in place on how this technology is going to be utilized, creating and designing systems that can actually be questioned rather than just taking AI for the sake of AI. All these are fundamentals on how we serve residents. Our city is demanding a lot more of us. There's very limited resources. How can we stretch that dollar to the maximum extent possible?

AI is again, a tool in that toolbox. So for us, having the ethical framework, having enough cybersecurity standards, having enough policy and governance people with cross-functional leadership at the table, questioning how this is being used, not just taking the vendor's word for how this AI is evolving. All these are foundational pieces and it can be used in different ways by different stakeholders and keeping that diversity and non-linear perspectives is what enriches the conversation around AI for me.

Stephen Goldsmith:

Well, I was deputy mayor of New York and in my portfolio 311, and this is like, call it a dozen years ago when we were experimenting with natural language prompts inside IVR bots, right? The old voice bots. Sixty percent of our calls to 311 in New York City were information calls. They weren't requests for service, they were request for information. So how smart is Sunny in Denver in terms of offloading work from 311 staff? But more importantly, I think given AI's capacity, talk to us about the continuum of a bot from taking the commodity call to personalization and problem solver.

Suma Nallapati:

For us, 311 is the way technology services interacts with our residents, right. Our team had to reduce quite a bit in terms of the budget reductions and the deficit that Denver was facing. However, Steve, we were not afraid of the reductions because we had Sunny as our buffer to augment our resources. Sunny is an AI platform that sits on top of our Denver gov website and that's integrated into our backend systems using very robust middle layer technology. It's available in 53 different languages, 24 by seven support, and we are constantly looking at the large language model for things like hallucinations and stuff. So it's very much curated within our own instance of our cloud environment, so it's not reading off of the internet. The subject matter experts, the domain experts of 311 are constantly looking at the answers. So with that, last year, around 400,000 interactions between Sunny and our humans happened.

That was the metrics with a great net promoter core. One example that I always give is if you're hungry and you're just a paycheck away from a homeless situation, we are finding people are interacting with Sunny because it's very embarrassing for a human to call another human and say, “I'm in this situation.” So what we are seeing is “I'm homeless, I need help,” are some of the interactions that Sunny is actually leading with a lot of empathy, a lot of non-judgment, and a lot of consideration and giving the responses. So we are finding that it's being actually very effective in these very sensitive cases. And you said informational and Sunny's doing a lot of that, right? When is the rec center open? When is the DMV open? Those kinds of information. You don't need to wait until 8:00 AM the next morning. You can get that at 11:00 PM. Sunny is allowing agents to be very human with our residents. We have a lot of elderly that like to call in and talk to someone, and they're able to spend that value add time with our residents in a very humane and human way.

Stephen Goldsmith:

That's interesting. Well, we have a professor in our center at Bloomberg Center at Harvard, Elizabeth Linos, who does a lot of work on how stigma deters people from receiving the benefits for which they're qualified. And your use of Sunny as a kind of an intermediary there is a really interesting example. We'll have to think a little bit more about that. What do you think the next generation of a bot is as the intermediary between residents and what they need for the city? Will residents be able to use something that looks agentic to solve their problems? Where do you want to go with Sunny?

Suma Nallapati:

What we are finding is Sunny is learning every single day. The more prompts, the more questions they're getting from residents, it's learning. And we are, again, double checking, cross-checking everything within our own knowledge databases so that it's truly within the domain of CCD and nothing off the internet. So I would say multimodal, proactive, personalized, even autonomous is the way I think chatbots are going to go. I would say with the amount of resources that are constrained for us in 2025, FY26, it's going to be much more of the resource from Sunny versus human agents because there's only so many resources to go. And thinking again with next gen bots, being able to handle text, voice, images, video, uploading a photo of a pothole and getting that right. The biggest thing for me is it's not just conversational, Sunny, but it's integrated to our backend systems, like Salesforce use cases, and then through a very robust MuleSoft integration layer.

So it's not just answering the question, but being able to generate a case and get it to the right agency and being able to triage that by itself and sending it so much faster to the people that are actually able to solve the problem. That contextual memory, the personalized conversations, all of those are going to lead to that autonomous AI agents. Again, with the human in the loop, I don't ever want to get away from that human in the loop, being privacy aware, secure, all those are foundational to all these bots. And I think it's also building emotional intelligence to a certain extent and sentiment analysis and things like that. The hyper-personalized real-time data that's so insightful and helpful when you have a real crisis on your hands.

Stephen Goldsmith:

Will we evolve soon to a place where Sunny reads the image, the photograph that's uploaded and automatically tags it to the right agency and creates the service ticket without the human in the loop?

Suma Nallapati:

That would be the hope to get to that level of maturity. We are still working on it, and we also want to be very careful with our users and how they are interacting with it and getting them that level of comfort on how their data security, privacy, all those considerations are being weaved into the threads. So at some point, yes, there is a possibility, but right now from a comfort level for our users, we are being very cautious.

Stephen Goldsmith:

I was talking to one of the, I'll not mention the name, one of the leading providers of permitting software in the country, and they pointed out that based on their exam, 40% of the delays in a permit process are because the applicant hasn't provided all of the necessary information and doesn't realize that they haven't done it. Each side is frustrated by the other. How can we think about an agentic front end that helps solve that problem by interacting so that the quality of compliance in the submitted data goes up?

Suma Nallapati:

Steve, as I mentioned, before I joined public sector, I was in the private sector, didn't even know who the governor was, let alone how I interact with government was not even a thing. So it opened up my eyes on when you think about public sector, people don't even know the difference between federal, state, and local. They don't even know which agencies to start with, right. So I want to start a cupcake business in Denver, say. Where do I start? That's the first question. And the websites can be extremely complex to navigate. So I would say this is where automation, this is where process improvements between the various agencies is very important. How do you streamline the process so that it's from a end user perspective rather than an agency perspective? This is where this group comes in, this is where this group's coming in. I don't want to know all that as an end user.

If I'm ordering something off of Amazon, I don't care which department does what, I just want my shoes on my porch by tomorrow afternoon. So that's the kind of expectation users have when they're interacting with government. We actually, when we held our GenAI summit, two years back, we had Reid Hoffman and Eric Smith as our speakers, and we were able to do a mini hackathon and was able to see the demos from public sector lens from startups and a company called CivCheck that we were able to bring and integrate into our technology roadmap to help us with streamlining this licensing and permitting process. And again, the Mayor is very, very interested in cutting down red tape, reducing bureaucracy where possible with enough guardrails. And we needed tools like this to help us with understanding how we can be better with our processes and not let the red tape prevent innovation, prevent businesses from thriving in Denver.

Stephen Goldsmith:

How are you thinking about AI in the use of permitting and inspection?

Suma Nallapati:

There's so much that can be automated, document reviews, smart forms, guidance that people can maybe fill at home and don't need to come into an office. Look at the first four steps maybe that can be automated and then sent to human officers rather than start with humans, and then that takes a really long time. And then being very transparent with the process. Right now it goes into a black box and people don't know what's happening. How do you keep the user updated throughout the process? Where's my permit? What's the next step? Where is this within the workflow? Giving those updates is very comforting. Amazon literally tells me when it's being shipped, when it's on my porch, right? Those updates are very comforting. That kind of ease of use is the goal. And again, multilingual is important. Compliance and fraud, lot of manual, repetitive tasks within that and room for error that can be prevented. And then the human staff is still very important. Again, handling the very complex use case scenarios. Final decisions are in the humans and can be seen as user-friendly.

Stephen Goldsmith:

So where are you on that curve, if a city wants to follow your example?

Suma Nallapati:

AI is one tool, as I mentioned, but let's start with the process right now, design thinking. Let's start with the whole workflow and see where the friction points are in your workflow. What step in the process is taking the longest? Is this taking 37 minutes and the next step is like four days and then the next one is 15 days? Get with that foundation and then apply automation. And maybe sometimes it's a business process change and it has nothing to do with AI. Look at all of that to create that cohesive holistic lens of where the friction points are within your workflow, and then AI can come at the top and expedite some of that with machine learning and all of those data that it's going through.

Stephen Goldsmith:

Let's go back to the consumer just for a second. When I was deputy mayor, if you wanted to open a restaurant in New York City, you could not apply for a restaurant license. There was no such thing as a restaurant license. There was license A, license B, fire permits and sanitary permits and building permits. And when you got up to a dozen, you could aggregate those and say, I got the certificate of occupancy. So I'm the cupcake guy in your example. I like cupcakes. I want to go to the cupcake business. In Denver, can I just say to whoever's the Sunny on the front end of the permit system, I want to open a cupcake restaurant, tell me what to do?

Suma Nallapati:

Yes. And Sunny is absolutely built for that. It'll at least get you the workflow and where do you need to start. And that's been the biggest problem for a lot of our residents. So it starts with that, I want to start a cupcake business, and it gives you the list of workflows. The backend systems, again, are legacy, and that's where the AI tools like CivCheck and others will help. The Mayor was very visionary in his thing and started with Peak Academy, which looks at process improvements in that Six Sigma way. And they outline all the friction points after these design thinking work sessions and CivCheck was brought in parallel. Sometimes it can be a process flow, it can be an automation or it can be AI purely. So we looked at it from that lens. So it's a combination of all of these that are resulting in faster issuance of permits, if you will.

Stephen Goldsmith:

Every one of your answers gives me an idea about 10 or 15 questions I want to ask. We've been covering the Peak Academy since it was invented and participating with people from Denver over the years. We recently had Brian Elms on to talk about Peak Academy and how it's evolved over the years. For the listeners who want to tune in, that's podcast [episode] 86. And Peak Academy, let's just for our listeners shorten it to teaching empowerment to public employees and what they can do. What does Peak teach about data literacy and generative AI today? How is Peak today with AI different than Peak Academy was seven, eight, nine years ago?

Suma Nallapati:

Great question. Peak is in the Department of Finance within the city and county of Denver. And when there's dollars associated with anything, it's very helpful because people want to see the hard dollar savings. They have partnered very, very closely with my office. We are like two peas in a pod when it comes to process improvements and efficiencies. We actually created a day of literacy for the non-technical people within the city and county of Denver. So Peak was right there along with us. And we had the vendor ecosystem come and present all these different possibilities. We were the facilitators, but we didn't lead the conversation because this is not about technology. This is about process outcomes, business outcomes, resident outcomes. So when we did that, I think it triggered a lot of cross-pollination of thoughts and ideas. And some of them are purely technical things, some of them are combinations, some of them are purely business process adjustments.

So going in this together has been very helpful. And Peak Academy is a true believer in AI after they've seen some of the possibilities. So we are doing this in conjunction. We are not doing this at the expense of each other, but just truly going in as true collaborators looking for the outcomes.

Stephen Goldsmith:

We have a grant from the Knight Foundation to look at how AI, generative AI can improve the relationship between community groups and the city as it relates to improving the quality of the neighborhood conditions, say streets or sewers. And one of the things we've been puzzling about is how generative AI will help explore causation. We have more potholes. Why do I have more potholes? Is it a drainage issue? Is it a maintenance issue? So I know you've been looking at these as well. So how do the generative AI tools help the city and community groups preempt problems, understand causation as contrasted to just where the problem is?

Suma Nallapati:

I don't think there's a direct AI tool that is responsible for all of the things you mentioned, Steve, but I think it helps with better resident experience overall. There are tools in the backend that can help with analyzing feedback from surveys and community and media gatherings, town halls that the Mayor actually is very, very strongly engaged on, and ensures that we take that feedback and work on it faster rather than it being like a three-month survey and then results and all of that. AI is helping us with even raw language and sentiment. So I think there's different ways I think AI can help. Resources are limited, we talked about it, right? Maybe once we get the feedback from these community groups, we can help with fairer resource allocation. Decisions are based on data rather than just emotion. How do you ensure that it's going to the right people that need it the most? How do you reduce bias in the services? Again, very important to design these systems carefully.

Stephen Goldsmith:

Suma, let's do this. When you come back in a year to talk to us again, which generative AI project that you have or will have undertaken will produce the most value to Denver residents? You have 25 seconds to come up with the answer.

Suma Nallapati:

I would say deeper personalization, meeting residents where they are.

Stephen Goldsmith:

Well, that was a great answer. It only took three seconds.

Suma Nallapati:

[laughing] I have a longer answer too.

Stephen Goldsmith:

Well, I want to close on that, but I also want to emphasize it because I've been thinking about it quite a bit, the fact that what generative AI really does is collapse the bureaucracy and makes it more responsive to the resident. And so your answer basically helps us think about how government, with humans in the loop, but can be transformed to be resident facing. And I think that's a really great way for you to quickly talk about the value you add.

Suma Nallapati:

AI helps us with the transformational work given to humans and the transactional, mundane, repetitive tasks to bots. People join public sector for the value of the heart, right? The heart of public sector is want to help people. When they're sucked into the repetitive mundane tasks, it takes away from that heart of public sector. And hopefully we give tools where the repetitive mundane tasks are taken out of the loop so that they can focus on the human touch, the heart of why they're in this job.

Stephen Goldsmith:

This is a really, really important point, right, that people join the public sector because they want to help people, but they often get put into these commodity paper processing routines where they're not actually able to help. So if we thought about what you just said, this is an enormous opportunity to motivate public servants by taking the work that is boring and tedious, outsourcing it to the bot, freeing them up to actually help people. So this is a great way to think about the purpose of your work and a great way to kind of celebrate the goals of the Mayor for whom you work. So let me just say to Suma Nallapati, thank you so much for your commitment to public service and your terrific ideas and your time with us today. Thank you so much.

Suma Nallapati:

Thank you, Steve. Thank you, Harvard Bloomberg, so much for this opportunity. Truly grateful.