How Cities Can Measure What Actually Matters

What does a city government owe its residents? Host Stephen Goldsmith speaks with Eyal Feder-Levy, CEO of Zencity, to explore how GenAI is fundamentally transforming the way cities measure, understand, and respond to resident needs. For decades, performance management in government has relied on operational metrics like crime numbers, pothole repairs, traffic flow. But what happens when the data looks good, yet residents feel less safe? When efficiency improves, but trust declines? 

In this episode, Feder-Levy argues that citizen satisfaction and perception should be the true North Star for city government. Using social sentiment analysis, AI-powered data agents, and real-world examples, he explores how GenAI is cutting response times, revealing hidden patterns, and closing the gap between statistics and lived experience.

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

Stephen Goldsmith:

Welcome back. This is Steve Goldsmith from the Bloomberg Center for Cities with another one of our podcasts. We've spent a fair amount of time, several years, more importantly, recently, thinking about trust and responsiveness.

How does a city encourage more trust from its residents? How is it more responsive in a way that encourages the trust? How does it respond to the needs of its citizens in a way that encourages their participation?

And over the last year or so, with eight or 10 of the country's largest cities, we've been asking the question, does generative AI fundamentally change this equation? Does the ability of a city employee or resident to use new tools enhance the opportunities to learn of problems earlier and respond to them more quickly?

So in that context, I reached out to a friend, Eyal Feder-Levy, who's the CEO and co-founder of Zencity, to ask him both to give us some ideas for our group of eight or 10 cities, and to talk a little bit about how his work in social sentiment mining connects.

And so welcome, Eyal. Thanks for joining us.

Eyal Feder-Levy:

Thanks for having me. It's really great to be here.

Stephen Goldsmith:

First - we'll come back to Zencity - are you a computer scientist? What are you by profession?

Eyal Feder-Levy:

Well, my mom asks that as well many times, but by training, I both have a background in computer science and in urban planning. I spent years in the technology space and then pivoted to the local government space and built most of my career in local government and in urban planning, which is where my passion is.

And coming to this world in the beginning of the previous decade when smart cities and concepts like that were just starting up, having a technology background boxed me into questions of how cities are using technology and data in their work. So I've been dealing mostly with that throughout my career.

Stephen Goldsmith:

Thanks for the background. And before we get deeper in stat programs, which we'll call performance management, how have you helped cities with evaluating sentiment in a way that gives them insights into performance?

Eyal Feder-Levy:

Back when I was working in local government, a lot of my role, a lot of my responsibilities were about how we use data, how we take data to make decisions. And I would always ask, "What are the questions that we're looking to answer?" And I know that that was a lot of the early motivation for the stat program, the rest of the stat program that came out of it.

And when I asked any leaders that I was working with, city managers and mayors, the answers almost always came back to the what they were trying to learn and hear is, "What does our community want? What are the things that are important to them? How happy or satisfied are they with their services?"

If we take an issue like traffic. Somebody would say, "We want to reduce traffic." We'd ask, "Why? Why do they want to reduce traffic?" "So people will spend more time at home." "Why?" "So that they will be happier with life in their community and with the services they're getting."

So, we started Zencity a few years back with the goal of being able to provide a data-driven answer to that question. We saw again and again that cities were answering this really important question based on gut feeling. A town hall meeting where only a handful of people show up. The STPs, the same 10 people that always show up.

We saw that that provides a lot of these leaders with a very skewed view of their, quote unquote, "community." Even when there's hundreds of thousands of people that live in a jurisdiction, they hear from just a handful of them.

So we became very passionate about, can we use technology to give a data-driven answer to that question? And as you said, we started by analyzing social media data. And over time we expanded to get that input in a lot of other ways.

But what we found is that hearing from many, many people and putting numbers behind them, making this a quantifiable thing really drove a change in the usage of this to drive decisions around policy and budgeting and messaging.

Stephen Goldsmith:

When I was deputy mayor of New York for Mike Bloomberg, the mayor was into measuring everything really. And I look at the mayor's performance scorecard, and I was responsible for that part of the operation and presenting it to him.

And we had hundreds of things we measured, but I was never sure that those measurements produced outcomes anybody cared about. How does listening to the sentiment, grading the anonymized sentiment in the social media world compare to the way a city would measure its own performance?

Eyal Feder-Levy:

So, I think, measuring the online sentiment is just one way to bring this to life. But the more interesting question to me is, how does measuring sentiment or perception generally be an outcome metric or an outcome perspective?

I would go as far as saying that measuring people's satisfaction or happiness is the North Star metric, is the target equation that local government is working for. Think about it. If at the end of the day a city saves money or collects more money, it doesn't give out dividends to council members, hopefully.

It uses those resources to serve their community better, to invest in the areas that people care about the most with the goal of making people happy with life in their community.

In my experience with most public servants I've met, there is a real passion to making sure that people's lived experience in the city is a good one, and that is what organizations are optimizing for.

So the question that I've been obsessed with is, can we measure that instead of trying to measure leading indicators that might lead us to it? And I'm very confident to say that it's something we can do.

Stephen Goldsmith:

You've watched our group of stat performance cities look at what stat would be, what performance management would be in a generative AI world.

I'm thinking about generative AI as the way a city or a community would discover the causes of things that annoy them or discover a solution to a problem. Why is that pothole here every six months? Or why are so many people slipping on this part of the sidewalk?

So when you watch these cities experiment with building their own agents, what do you predict the future is for generative AI in performance management in behalf of a city?

Eyal Feder-Levy:

Thinking back to the goal of stat programs, of performance programs, they don't exist to give somebody a report card. We're not doing this to clap and say, "Hey, we're doing great." We're doing this to actually improve services.

To recognize what are things that could work better and how they can work better and then make those solutions and measure and see if we are actually doing better.

And the current way or the historic way we did stat has a very, very long lead time and it's a very resource intensive process. There's a lot of people that need to do a lot of work and spend a lot of time in order to get that done.

So I've been extremely inspired by the group in seeing how we can crunch those capabilities to make that cycle of stat a lot faster. Basically, I think there's four levels of improving stat that I personally am excited about.

Number one is, let's do that with less resources. We don't need so many people in the room and so much staff time and so much expensive technology to get all this data done. That's the basic level. Do the same thing with less resources.

I think the next level, which already starts to make this better is do it in less time. One of the lowest hanging fruit that I already saw, the ages that our group members have done is that we shortened the cycle between I have a question, I get the root cause answer to it, and I can take action on that root cause.

Before that, you'd see a number, you'd ask the staff, "Hey, could you tell me what the reason is? " They'll say, "Great. Yeah, in two weeks, we'll circle back to you after we've investigated and then we'll schedule that meeting two weeks after, et cetera, et cetera."

Now we can, using these agents, I've seen that, for example, with the 311 Analyzer that Steve from Seattle has built, we're able to get to the root cause immediately while we're in the meeting without knowing the question ahead of time, and that already crunches time and that's value.

And then the next levels of value is we can recognize patterns and root causes that we didn't even know to ask for. That's, I think, the next thing we're aiming for. And then eventually we can open up those capabilities to everybody even beyond the city, to community members and to other folks to ask those questions.

But right now, the thing that we're already starting to see from that group is we can crunch the time and the resources that it requires to answer those questions.

Stephen Goldsmith:

Let's say that hypothetically, the deputy transportation director wants to explore the data to determine how to be better at their jobs. How they could be more responsive. How do you imagine six months from now or a year from now, they could use generative AI to ask questions which would help them do their work better?

Eyal Feder-Levy:

The simple version of this is when we have a good definition of what success looks like. So for example, for a transportation director, that could be time in traffic.

Or that could be amount of people that are using public transportation or amount of people that are using alternative mobility solutions, or of course, as I would say, the satisfaction of people with ability to move around the city.

Once we have those target equations well-defined, we first can use GenAI to wrangle all the data easily to see how we're doing. With tools like MCP servers, we can now easily build agents to pull data from a lot of different siloed sources and present them easily and ask questions over them easily.

I think the next level would be that we can ask, "Why?" Like, "Hey, average time in traffic or average trip time. Let's dive into, is there specific roads or intersections where travel time is longer or are there neighborhoods where we're seeing a lot of people convert trips to micro mobility? Show me that specific example."

And I can ask those questions in a human language in the same way that I would ask a knowledgeable staff member. So in a case like this, we would have an MCP server and the data would flow and then we'd have agents that give us an easy presentable overview of the data and we can ask follow-up questions on them in a human language.

Stephen Goldsmith:

So now let's bring the two parts together. And many mayors across the country are frustrated by the fact that crime is down, but their public feels like they're less safe, right? They feel like crime is up.

How would a mayor or city official investigate the difference between the data on safety? It could be any subject called safety and the sense of safety.

Eyal Feder-Levy:

First, I'll say it's one of the biggest challenges that local government is facing and specifically in public safety, it's emerged as a real challenge.

And in a lot of places, the perception of a problem is sometimes different than the state of the problem itself, but the perception is really important. A, because as we said, that's the North Star.

But also if we take safety specifically, a parent will decide if their kid can walk to school, not based on recent homicide numbers. They would decide that question based on do they feel safe in their neighborhood and do they feel like it's safe to walk to school? So that perception is reality and is an essence that we need to solve.

I was just meeting yesterday with one of my favorite law enforcement partners. We run recurring monthly surveys for them on sense of safety across the city and trust in law enforcement. One of the interesting things we saw there was that over the last year, across almost all demographics, sense of safety and trust in law enforcement has improved.

There was one specific community where sense of safety went down dramatically, which is the Hispanic Latino community. But at the same time, their trust in law enforcement, in the city's law enforcement, went way up.

We were able to use our generative AI agent there to ask a few questions and follow-ups. That the reason sense of safety is down for that Hispanic Latino community is because of things that are happening on the federal level.

And that actually shines a really positive light on the city's police department who the residents now trust more because they have less trust in the federal government.

So I think step one, to answer your question directly here, is to be able to understand that this is something that you need to measure separately from measuring crime data, and then track in a stat process, track both of them together.

In a lot of our law enforcement customers in LAPD, for example, is one of our leading examples. This data's embedded into the comp stand process, which is the law enforcement version of a city stat program, and they're looking at perceptions and sense of safety alongside the crime numbers.

And a lot of times they're moving together because people feel more safe if there's less crime and if the police is doing a great job. But in some cases, we see a divergence between them and there we can create interventions that are different, that are messaging and communication interventions to allow change of perception and not just policing interventions.

Stephen Goldsmith:

Let's imagine a year from now, you're talking to the deputy mayor of New York or mayor of Indianapolis or any other kind of random choice, and that person wants to increase trust in government by being more responsive, by fulfilling the expectations of residents better.

How would you set that up? What changes would you make? How would you arrange the flow of information so that you could actually produce more satisfaction and trust?

Eyal Feder-Levy:

It's a great question. And I know to some extent every mayor and city manager and police chief are asking themselves this question. The first part of my response would be that you can't manage what you can't measure. So the first step would be to measure it.

Our approach to measure this, and today I'm proud to say we serve about 200 million people in the 400 cities we work with, is that you need to put a few capabilities in place of listening and measuring people's perception and satisfaction.

Both representative community-wide surveys, analyzing of what people are sharing on social media, post-interaction surveys that ask people that got services from the city directly, how was their experience on that specific service? Putting those data collection infrastructure in place and bringing all that data together, I think is step one.

GenAI is already playing a really big role in that. But a year from now, I'm thinking about, hey, we can capture voice data a lot more effectively. Hey, we can survey people in much more interesting ways by mimicking a video interviewer to get more information from people. So there's a lot more exciting stuff coming down the pipe there.

But after we've collected that information, the next step is to analyze it and review it on a regular basis. And to your question before, to cross-reference it with other metrics that we're looking at that are operational metrics.

And the thing that we can do there, hopefully in a year from now, is very easily ask our agent, are we seeing a parallel trend between perception data and satisfaction data and the operational data? And if not, where do we see the gap and what could be driving it?

Stephen Goldsmith:

Well, that sounds compelling. Maybe I should sign up for another stint as a deputy mayor in some city.

Eyal Feder-Levy:

I'm sure every city will be lucky to have you.

Stephen Goldsmith:

I don't know. Thank you. Well, this is Steve Goldsmith with Eyal Feder-Levy from Zencity. An interesting conversation about the future of responsiveness and the definitions of it. Thanks so much for your insights, for your volunteer work with our cities and for telling us today how cities can develop more trust. Thanks so much.

Eyal Feder-Levy:

Thank you. Talk soon.