AI Agents and Peak Academy: Brian Elms on Empowering Government Workers

Host Stephen Goldsmith sits down with Brian Elms, former director of Denver's groundbreaking Peak Academy and founder of Change Agents Training, to explore how generative AI is transforming government's most successful employee empowerment model. Elms explains how Peak Academy has saved governments over $50 million by teaching frontline workers to become problem solvers in their own services, and why unlocking employee potential matters for everyone in a government organization. They also discuss how AI agents augment this work, with Elms recommending eliminating useless work first, then layering on performance management and AI tools to help subject matter experts — not just executives — drive continuous improvement from the ground up. 

Take the survey at bit.ly/datasmartpod.

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

Stephen Goldsmith:

This is Stephen Goldsmith, professor of urban policy at the Bloomberg Center for Cities at Harvard. Another episode of our podcast, Data-Smart City Pod. I've asked today's guest to join me because he's nationally recognized as a leader in government innovation, in part because he's done really cool things, in part because he's been at it for a really long time. He was the director of Denver's Peak Academy, which I think I've probably written about a half a dozen times and the CEO of Founders of Change Agents Training. Welcome, Brian Elms.

Brian Elms:

Thanks, Stephen. It's so good to see you.

Stephen Goldsmith:

Now, just before we get started, I got a bunch of questions for you, but this says you're the founder of Change Agents Training. So we've been developing agents here, but they're not people. Are you changing generative AI agents or are you changing people as agents?

Brian Elms:

Well, funny you should ask. I think it's a twofold. Yes, we are building agents to help us with process discovery, and we do a lot of our work changing how employees interact with their services to try to improve them. And then we layer on some agents that are AI.

Stephen Goldsmith:

So let me reflect and ask you a question. I do not think that very many initiatives of government have had as much effect in US cities as what you did with the Peak Academy in Denver. It's been replicated everywhere and has produced enormous change. So I want to ask you mostly today about why it's not outdated, but let's start with, what was Peak Academy and how did it unlock so much value?

Brian Elms:

Well, first of all, I have to collect myself. Hearing you say that we were so impactful, yeah, I'm kind of floored. So let me collect myself first. Peak Academy started in 2011 in the city and county of Denver. It started because we were out of money. We were in a recession and we needed to figure out how to keep helping people with less money, how to keep helping people with not enough people to provide the services. It was designed to help improve our service delivery by investing in employees, teaching employees problem solving techniques, coaching them and supporting them when they ran into barriers or challenges for changing their service deliveries. So our entire focus was how do we train all of the employees in the city and county to improve their services one service at a time? And we learned very early on that a lot of employees have the ability to do really cool things if we can unlock their potential.

And so that's what Change Agents does in the same way that Peak Academy still operates. So Peak Academy just celebrated their $50 million worth of savings in government. They help provide training and support for employees who provide day-to-day services to residents, and they try to figure out ways to do it better. And they ask the subject matter experts to take a step back, look at their process and try to improve that service. It is highly replicable. I mean, we have over 60 governments that run some form of Peak academies today. It is a lot of fun. I think that's one of the big aspects to it because you get to have agency in the services that you're providing. And when an employee who does work for a living and they're doing their day-to-day job, when they feel like they have the power to make change, you can unlock their potential really quickly. And that's what it does. That's what Change Agents Training does, and that's what Peak Academy is still doing today.

Stephen Goldsmith:

So I'm going to stay with the original theme, finding out a little bit of how you implemented it. But I have this photo, my favorite picture I've carried with me everywhere. And it's three fellows who worked for the Indianapolis AFSCME local who were representative of that local, not the only ones, and they're out on the street patching holes in the roads. And these fellows were asked, as were their colleagues, how would you unlock productivity if you could change any of the rules, if you could make any suggestions? And when you talk about subject matter experts, it was not the mayor or the supervisor. These guys are the subject matter experts in asphalt, right? And I won't go through all their suggestions, but how did Peak Academy institutionalize my anecdote?

Brian Elms:

I think all cadence, Stephen. The amount of training that they do, the amount of follow-up that they do, the amount of things that they're doing to try to create that flywheel effect that you're describing. How do you get the guy who fills a pothole on a daily basis, look at it differently and continue to do it over and over and over again? Provide them the right tools, provide them the right setting, and they'll flourish. We see it all the time. I mean, I'm doing a project in Alabama right now where we're looking at the DMV and reestablishing how they do renewals. And because of what they're doing, they went from 90-minute wait times down to about 10, maybe even less than that in more instances. They figured out how to improve their business licensing process so much so we have actually helped them generate more revenue than they ever have.

It is very simple, practical ideas that snowball over and over and over. And when one employee makes a change, it affects another employee and they want to make a change. It is absolutely one of those things where it's like innovation is caught, it's not necessarily taught, where excitement is caught, it's not taught. And when people feel like they can make change in a process, they just keep going. They don't stop.

Stephen Goldsmith:

So a lot of the business literature suggests that it's the middle managers who are most resistant. I found that to be a little true in Indianapolis and really true in New York City where there were lots and lots of managers. So discuss a little bit around how the original Peak model dealt with not just empowerment of the men and women on the street, but the people who are watching them and supervising them.

Brian Elms:

Unlocking potential in both areas is a real key. So how do we help mid-level managers get what they need to get done without getting into the morass of report writing? So how do we unlock them to enable their team members? And a lot of that has to do with giving them the proper tools to be able to do so. So instead of like checking up on them in a micromanagement fashion, how do you change your philosophy to create scoreboards for them or create incentives for them to keep going after and improving their work? The mid-level management place, I would say is one of the hardest jobs to be in, and you get pressure from the people below you and you get a ton of pressure from the people above you. And trying to figure out how to calm those two pressure valves is a tremendous asset.

So if a mid-level manager can calm the people at the top and calm the people below them, they can actually thrive. So what we try to show them is there are ways, there are skill sets that you need as a manager that can do that, create standards, create scoreboards, learn how to coach, command, and counsel your team, move away from this catch them all sort of place where I'm just trying to catch up all the time, I'm just trying to catch my team doing something. Move away from that theory and into the coach command council theory where you're coaching your team, you're commanding your team when we're in a crisis and you're counseling their team when you're having challenges. Give them things to pursue. How do we make this a little bit better than we did yesterday? What do we need to do to fill one more pothole? What do we need to do to help one more client today and give them parameters to do so?

Super fun. But yeah, I would agree with you. Mid-level manager is the hardest place to be in government and it also can be the place that stifles a lot of innovation.

Stephen Goldsmith:

So let me ask you one more Peak 101 question. Tell me a little bit how it was structured. Did people get time off? How many sessions did they go to over what period of time? What did it look like?

Brian Elms:

There are multiple different training platforms that they offer. So they offer an introductory training that does about three or four hours of training. They have a very intense training, which they call their black belt, similar to a lean process improvement program. When I was doing it, it was about 40 hours. They've dropped that down to three days and then they do follow-up sessions now.

Stephen Goldsmith:

Let me try to get some free advice from you. Generally what professors do is kind of learn from whoever they're talking to and tell the next person. We don't actually do anything. We just got to talk about it. So let me ask you for some advice. We've been working with half a dozen of the largest cities in the country on how to modernize performance management or stat programs. And the theory is that the original model had a good leader and a smart data analyst wonk in the background and the leader would cross-examine his or her management, and then they would be supported by the analysts. The iteration of the questions and answers, part of innovation is iteration, and the innovation would be two weeks later, they'd come back. So we've been trying to figure out with generative AI two things: how do you speed the iteration and how do you broaden with natural language the number of people who have access to the data? So let's take Peak and apply it to unleashing the power of generative AI across the city enterprise. How would you do that?

Brian Elms:

The biggest challenge I had, Stephen, was I did not know what was going on in some of the departments. And so we're tasked to figure out the performance in these departments, but we don't really know how they do what they do. We're sort of guessing how they do what they do. And back then, I had no database to tell me what technology are they using, what's the hierarchy structure? What are the processes that they're going about? And I think now you can do that with generative AI. You can get that database fulfilled where you have the DNA of the organization. If you have the DNA of the organization, your ability to create performance understanding is going to skyrocket. So if I knew how they did what they did, what systems they were using, my ability to help them find better ways to do it or empower their team members to find those spaces would be insanity.

You can do that now and you can do that quickly. We just did a project in Australia where we helped them understand about 48 processes, documented high level information, task level how they do it. We did 48 processes in three days. If I were to do that back then, back in 2012, 2013, 2014, that project would've taken me a year to get those 40 processes. I can now do those 40 processes discovery-wise and provide them actionable suggestions based all on their feedback to us and our feedback to them using generative AI.

Stephen Goldsmith:

So you're making some interesting points on business process re-engineering and process mapping and your ability to do that quickly, which is really quite fascinating actually. How would you get the deputy sanitation manager of Denver to be able to use today the same analysis that you and your team used 10 years ago?

Brian Elms:

Well, we wouldn't have to do what we were doing, which is Excel 101, how to read your performance dashboard. That should have changed. I also feel like there are parts of the performance management structure, Stephen, that were really heavy touch. We had multiple analysts looking at multiple services all at the same time. And like you said, it would take about two weeks for us to get feedback. I think that feedback loop slows down tremendously. I also think that you need a really good analyst who can sit with that director and show them this. I don't think you have to teach them how to do the data entry anymore or even the data analysis. You can use AI to do that with you and for you. So I think that would change. The other piece, Stephen, we did a lot of analysis on things that we didn't have a lot of power and control over, and we spent weeks, if not months, if not years, trying to unpack those things.

We now can do a deep dive in, okay, what's happening on this intersection that is causing these potholes to appear rapidly? That's something we would not have been able to do in the old version because I would've just been looking at how many potholes did you fill? How long did it take you to fill them? Were we able to get a better pavement standard going on? But now I could say, "Hey, that intersection is causing problems both at 911, at 311, and in all these other areas." And I would be able to do a deeper dive with that team. And I don't know that I would have to sit with the director of that department and explain how we did the analysis. The analysis would be a little more obvious because it's more transparent.

Stephen Goldsmith:

What would data literacy training look like in a Peak Academy?

Brian Elms:

What we found, Stephen, when we first started Peak is we were doing Excel 101. We then brought people into how do you use Excel a little bit better and how do you use this data? Because I think there isn't as much data literacy needed when it's going to spit it out to you.

Stephen Goldsmith:

Well, we need to teach them how to be semi-literate so they can use those tools.

Brian Elms:

And they have to also be able to recognize when generative AI is making things up. That's the real trick. And that's a challenge that I have in my work is where is it hallucinating and where is it making stuff up?

Stephen Goldsmith:

I, just for fun in my class last year, went over to the Boston open data site and I grabbed street data and I played kind of a game using an agent in the Harvard sandbox. I said, "I'm the transportation director on the north side of Boston. How does the number of potholes I have compared to the south side?" And then I said, "How does my demographics compare?" And it went on for a while until I got over my head and I had to stop. But going back to the beginning, you're the CEO and founder of Change Agents Training. So how would you train agents to use agents to solve city problems?

Brian Elms:

Well, we've built our own agents on problems that we've seen. So we're not necessarily training our clients how to do that. We're still training our clients on how to visualize their processes, how to show people where opportunities are available, how to help them get to the place where they can see a problem. One of the challenges that Peak Academy does really well and what we do in change agents is we try to change the language of change. We try to change the language of problems. How do you see it, how do you say it, and how do you solve it? And what we do is we try to get people on the same vernacular for attacking a similar challenge. If I can do that, our ability to move faster through the problem solving exercise is exponential. If they're all using different language for how they see something, if they're all using different languages for how to solve something, we're going to be talking through each other.

I've never thought about creating a product where I teach people how to make agents. That's a cool idea. I'm hoping I can steal it from you someday and let you know how successful it is, but I still look at how do I make humans who mow grass for a living better at their jobs? How can I help someone who works in a park and every day goes out to fix a backflow preventer, but they can't find it. They don't know where it is. They don't know why it's broken. That's who I work with the most, is helping those agents feel empowered enough to make change. That's my goal.

Stephen Goldsmith:

Just to follow up quickly, let's talk about that grass cutter in the park. One way I hear your answer is you're going to work with her so that she can do her job better. Another way to think about this is to ask the question this way. What can you teach her about how she can use the data and tools available, or maybe her boss, to be the coach that you are, right? How do we use the generative AI to unlock the potential to do things better? How do we get there?

Brian Elms:

Getting that agent to understand, one, that they can make improvements is going to be the biggest barrier. So I don't just cut grass for a living. I am my own person and I can actually solve these problems. That's the first barrier. Not all grass cutters believe that that's their job. There is a thought process that they have that you're going to have to unlock that. There aren't all those people in the world who care that there's the ability to do this. And one of the challenges that we have in both Change Agents Training and our other company River Work is not everyone wants this much information. Not everyone wants to act on it. So what we try to do is meet them where they are and get them to believe that there's a different way of doing it. That's the first piece. The second piece would be to sit down, show them what their data would look like.

You're totally right. You could use AI to measure how much rain they've gotten, how much nitrogen they need to put in the soil, the downtime of the mower. You could totally do that and improve that service. I think you would want to challenge them to see how far they would go without saying, "Oh, you're not going to go this far because you don't know how to use an AI agent." I want you to go as far as you possibly can using a marker and a Post-it note. And then if you're excited by that, let me pull you to this next realm that you're talking about. And can I get a individual who always thought of themselves as a mechanic to become the leader of the mechanics, the manager of the mechanics, the department director? That's a challenge that you want to grab them where they are and see if you can pull them through.

There are so many anecdotes where I can tell you, "I started working with a mower and now they're the head of Parks."

Stephen Goldsmith:

That was a great answer. I'm from the school of, we have good people working in bad systems, not bad people. So a whole approach in your dedication has been towards changing the systems to help people work better. We've got an audience of state and local officials listening. Name one place they should start in order to unlock the productivity of their workforces in order to be more responsive to their citizens.

Brian Elms:

There are three things that I tell every leader and supervisor. So one, learn, be very curious about how you do things. Constantly try to learn. Two, let your team members fail, let them make mistakes, and help them when they fall apart. Try to keep pulling them through this world. Three, live in this space where you're constantly improving. Steve, if I asked you, how do you get a travel approval when you were mayor? You would say, "Hey, I asked my assistant and then a couple weeks later I'm traveling." Or you don't even know there's a travel approval process. But if you asked me how did I get travel approved through Peak, it would take me about 17 to 18 steps to do it. Most managers, most leaders have no idea how long and complicated it is to do a simple service, so constantly improving.

It's not just good enough that there's 17 steps right now. Let's make this 15 steps next year. Let's make this 12 steps in a week. The challenge that we have, Stephen, there's so much useless work that we perform and we are in this constant cycle of that useless work. I love doing performance. I love helping people get to that space, but if your team is doing useless work, they don't want to do performance management. So your challenge is teach them where to find the useless work, rid the useless work, and then we can start doing how are we performing.

Stephen Goldsmith:

That was a great answer. You didn't just call me useless, did you? That was... I just want to make sure.

Brian Elms:

I mean, useless work sucks the joy out of your job.

Stephen Goldsmith:

Yeah, no. I'll close with another anecdote. I was sitting in New York City Hall working for Mike Bloomberg when somebody decided because we didn't have enough money that there'd be no out of town travel unless approved by me. Me! We have 300,000 employees. I thought this is like, what is this process and how possibly would I do it? I was not very good at it. But any rate, on that closing note, let me just back up and say, I have long admired what you, Brian, have done. And it's because it liberates publicly minded officials to do a better job for the residents. And the literature shows that what motivates most public employees is not the money, it's the ability to serve. And so what you've done is made it possible for people to serve better. And for that, we thank you and for your time today on the podcast. Thank you very much, Brian Elms.

Brian Elms:

Floored again, Stephen. Thank you so much. Coming from you, that means the world.