by Mathis Bitton, Elizabeth Haas and Peter Hirshberg

Summary. AI has the potential to revolutionize local government operations, but American cities face significant challenges in adopting these technologies. In interviews with more than 150 local leaders across dozens of U.S. cities, respondents consistently reported three problems: sclerotic and siloed bureaucracies, burdensome regulations, and risk aversion. AI, however, could be a turning point. Cities have successfully deployed AI to: automate tasks, make better decisions with data, and better engage with the community. The key factors in the cities that have done this successfully are: clear vision and strategy, systematic de-bottlenecking, public-private partnerships, and adherence to governance principles.

---

America leads the world in innovation. The United States has the highest-valued startups, the most prestigious universities, the most prolific researchers, the best AI companies, and the most venture capital funding. American cities, however, are the exception. In the latest ranking of smart cities by the World Competitiveness Centre, no American city made the top 30 — and only New York, Boston, and Washington, D.C. made the top 50.

There are a number of reasons why government — and local government in particular — operates differently than businesses. Three years ago, we started to work with the United States Conference of Mayors to understand which of these reasons matter most. In particular, we tried to find the main obstacles standing in the way of AI adoption at the local level.

In interviews with more than 150 local leaders across dozens of U.S. cities, respondents consistently reported three problems:

• Sclerotic and siloed bureaucracies (83%), or the way in which local departments tend to work in isolation from one another, with few incentives to reform themselves.
• Burdensome regulations (44%), or the web of rules that prevent cities from acting boldly, including purchasing requirements, building codes, and other regulations that make it practically impossible for cities to work with startups.
• Risk aversion (31%), or the sheer fear that often animates local leaders when it comes to new technologies — significantly more than in business leadership. Mayors know that technology can help their cities, but they remain afraid of experimenting with tools about which they know relatively little.

Combined, these three roadblocks have made local governments remarkably stagnant in an otherwise dynamic and entrepreneurial country.

The explosion of AI capabilities, however, could be a turning point. In this article, we look at the ways in which cities can use AI and provide a framework for local leaders looking to transform their cities.

How Cities Can Use AI

Broadly speaking, cities are pursuing AI projects in three broad categories: automating tasks (mentioned by 76% of respondents), making better decisions with data (41%), and engaging the community (23%). In what follows, we explore some of the main use cases in each category and explain how these capabilities have the potential to make cities both more efficient and more responsive to their residents’ needs.

Automating tasks

Today in Los Angeles, a small construction company needs an average of 14 procedures, 105 days, and $85,841 to obtain construction permits. In San Francisco, it needs an average of 19 procedures, 184 days, and $108,063 to obtain the same permits. Entrepreneurs looking to open a restaurant, salon, or shop in San Francisco must navigate over 25 different requirements spanning building codes, fire codes, zoning ordinances, ADA rules, and more. Just applying for these permits costs on average thousands of dollars and takes more than six months before they can break ground.

Tomorrow, AI could automate most of these processes away — and hundreds more. Every permit application could be filled or reviewed in a matter of minutes. Every Excel spreadsheet could be analyzed and updated in real time. Data could be shared from one agency to another without human supervision. In the cities we’ve studied, bureaucratic automation was by far the most common use of AI. And for good reasons: Of all capabilities, it remains the least expensive and easiest to implement.

For example, in Honolulu, Hawaii, the Department of Planning and Permitting has cut the time to complete residential permits by 70% with AI. The head of the department is now expanding the initiative into a new platform “not unlike TurboTax” that will “ensure permit applications meet all necessary requirements before being transferred for [automatic] completion.” If successful, the project could automate hundreds of tasks, save thousands of hours, and save millions of dollars in the long run. The initial project cost $200,000 — about one-tenth of the benefits expected in the first five years of implementation.

Some might fear that cities will replace their employees with algorithms, but in practice we’ve seen that AI makes day-to-day operations more efficient without removing the need for human input. Instead of filling out forms or spreadsheets, civil servants can dedicate their time to bigger-picture and larger-impact projects for their cities. As Reno’s mayor Hillary Shieve told us, “We’re not using AI to replace our people but to make them focus on what matters.”

Making better decisions with data

Cities already collect enormous amounts of data on the urban environment; with sensors and cameras, they monitor air quality, noise, utility consumption, traffic density, parking violations, construction activities, and environmental conditions. The problem is, they seldom take advantage of this wealth of information. The vast majority of our interview respondents (over 87%) admitted that their departments are sitting on data that they lack the skills or resources to process, let alone leverage in real time.

That’s what AI brings to the table. It can handle data on a much larger scale, at a much lower cost, than human beings. It can integrate diverse streams of data — cameras, sensors, surveys, language, etc. — and analyze, predict, simulate, and forecast trends in real time. It can also get better with time, detecting hidden patterns and adapting to changing circumstances. In short, it can help cities make smarter decisions about and provide enhanced services in every domain of urban life. These include:

Transportation

• Make traffic signals more responsive
• Implement dynamic pricing on roads and parking
• Re-route transit services in real time
• Communicate predicted bottlenecks to residents

Infrastructure

• Predict maintenance or upgrades
• Optimize resource allocation to extend asset lifetimes
• Forecast costs, timelines, and impacts of capital projects
• Optimize rollout plans to minimize disruptions

Crisis response

• Forecast potential crisis scenarios to take preemptive measures
• Model impact, timing, and scale to guide the allocation of supplies
• Automate communication strategies with at-risk populations

Social services

• Identify intervention points
• Target the provision of preemptive measures
• Simulate resource constraints against evolving demand
• Adjust eligibility requirements in real time

Urban planning

• Model decade-long scenarios
• Simulate the impact of potential policies and investments
• Evaluate alternative paths of development

Innovative cities are already launching pilot programs in each of these areas. In transportation, Seattle has partnered with Google Research’s Green Light initiative, which uses AI to manage signal timing to improve the flow of traffic. In a few months, the city achieved $10,000 of delay savings in eight locations. In infrastructure, Deloitte has found that AI-driven maintenance reduces infrastructure repair costs by 25% in more than a dozen cities. In crisis response, California has successfully used AI to monitor over 1,000 cameras to detect wildfires. Within the first four months alone, AI detected 77 wildfires and proved so successful that TIME magazine recognized the program as one of its “best inventions of the year” in 2023. In social services, the city of Allentown, Pennsylvania has saved an estimated $1 million in taxpayer dollars by streamlining incident investigations across 21 departments with AI. And in urban planning, researchers at Tsinghua University in China have just developed the first AI planning system capable of outperforming human architects.

In all of these examples, the city in question was already collecting the right data — and already had systems in place to do so — but could not process it. AI just gave them the opportunity to harness masses of data that they would otherwise leave untouched. And this kind of intervention has the potential to reinvent urban decision-making for the better.

Engaging the community

The final capability that AI brings to cities has to do with the local government’s relationship to residents. In recent years, the “smart city” paradigm has — rightly — been criticized for its technocratic bent. Advocates of smart cities often push for top-down innovation, neglecting the populations that new technologies are here to serve. Fortunately, AI can help make cities more, not less, democratic by simplifying access to both information and services. For example:

• Answering questions: In Raleigh, North Carolina, AI chatbots are able to manage 90% of calls to administrative agencies, which frees up time for operators to answer more complicated or time-sensitive inquiries.
• Filling out documents: In more than a dozen American cities, AI chatbots are helping residents fill out hundreds of documents. The bot asks simple questions and fills the form.
• Translation: AI translation companies like Unbabel can translate emails and web pages into more than 20 languages. With a team of human editors to verify for accuracy, they complete this service at $0.02 per chat, a much cheaper rate than traditional translation services.
• Interactive modeling: The city of New Rochelle, New York, has built a platform that combines visual computing and AI to model changes to the built environment. Citizens can evaluate proposed changes, make suggestions of their own, and see what their ideas would look like in practice.
• Dynamic services: The MIT Media Lab has worked on a platform that adapts zoning laws to the real-time preferences and needs of residents. The platform surveys locals on their preferences, collects data on living costs and other relevant variables, and updates zoning practices to evolving circumstances. The Lab has worked with Hamburg on an experimental project with this platform, which accelerated the construction of more than a thousand houses by more than a year.

The last two use cases are more experimental, but they have the potential to alter the way in which cities interact with their residents. In the not-so-distant future, we could imagine residents collaborating on all sorts of projects, with AI aggregating their contributions into coherent recommendations for the city. We could also envision dynamic public services — not just zoning laws, but also social programs, policing practices, or building codes — adapting to the real-time preferences and evolving needs of residents. If generalized, these capabilities would make cities altogether more responsive to popular input, providing a democratic counterpoint to the technocratic excesses of the “smart city” paradigm.

Key Success Factors

The question then becomes: How should cities think about integrating AI into their operations? Across our case studies, four success factors stand out:

Vision and strategy

First, a successful AI strategy fits into a broader vision for the city, with a clear set of priorities. In a lot of our interviews (about 70%), local leaders admitted that they experimented with AI without a clear goal. They launched pilot projects here and there, hoping that some would prove successful and scale. This approach costs more and delivers less.

Successful cities begin by identifying their most pressing needs and focus their use of new technologies accordingly. In other words, AI should not be implemented for its own sake — it is not a shiny object but a toolkit to solve specific problems. The question is not “How can we use AI?” but “What problem are we trying to solve, why, and how could AI help?” If AI is not the best means to tackle a given challenge, its implementation should not be forced. Conversely, if AI does prove useful, the development process should follow four simple steps:

1. Align on vision: Identify challenges and design a strategy with clear priorities.
2. Pilot: Test value and feasibility with limited users in a controlled environment.
3. Refine: Expand use to additional users, optimizing the technology along the way.
4. Scale: Roll out the technology to capture all the value.

The first step is often the most important. In every case, local leaders must begin by understanding their own context. What are the city’s needs and challenges? What will the city look like in 10 years? In New York City, for instance, former Sidewalk Labs CEO Dan Doctoroff and Robin Hood CEO Richard R. Buery, Jr. have advised the city to 1) automate citizen requests, 2) democratize access to information with chatbots, and 3) optimize traffic to tackle congestion because these were three consistent sources of frustration in citizen surveys, standing in the way of future-focused growth. In Las Vegas, the city created an “innovation district” for experimentation with autonomous vehicles to cement its position as a rising hub for transportation technology. Other cities that face a housing shortage might similarly prioritize AI-powered planning to accelerate construction and/or zoning.

Overall, developing an AI strategy is not about having a document called “an AI strategy.” It’s about establishing clear goals and priorities for the city, and then seeing how AI can help turn that picture into a reality.

Systematic de-bottlenecking

AI can improve city operations across every domain of urban life. But burdensome regulations and perverse incentives are often standing in the way. In our interviews, we consistently found that local bureaucracies resist transformation, either because current processes do not allow radical change or because the people in place have every reason to fight change. Cities must transform themselves before technology can transform them in turn.

In practice, this means:

• Including an innovation team in every city agency
• Creating incentives for bureaucracies to reform themselves
• Building an Office of Civic Innovation that rewards bold ideas with hackathons, prizes, and pilot programs
• Changing procurement processes to let startups compete for public contracts

In short, it means ensuring that the spirit of innovation thrives at every level of government.

For instance, until a few years ago, New York City could not partner with startups because its procurement process favored older, larger companies. Two years ago, the city changed purchasing requirements to encourage pilot programs, and the results speak for themselves: In 2023, over 600 companies applied to pilot products and over 50 pilots were deployed — 10 times more than in prior years. This is but one example, but it illustrates the kind of change that cities can bring to their operations. Across the board, local leaders should identify and remove the obstacles standing in the way of innovation.

Public-private partnerships (PPPs)

PPPs foster collaboration between city governments, private companies, academic institutions, and nonprofits. These partnerships facilitate the co-development of urban solutions, enable knowledge and resource sharing, and encourage experimentation with private initiative and public backing. For AI, a technology that researchers and startups understand a lot better than local governments, PPPs are crucial to success, because each side brings complementary qualities to the table.

In our case studies, we noticed that the most successful cities formed long-term partnerships rather than project-specific, short-term contracts, formalizing their arrangements in new institutions such as “advisory boards” or “smart city initiatives.” For instance, the city of Columbus, Ohio, has established the “Smart Columbus Initiative,” which involves the city, technology companies, universities, and community organizations. In five years, the initiative has developed AI-powered mobility solutions, logistics systems, and data-driven services.

Cities can also partner with venture capitalists to secure funding and support for their ventures. Along these lines, Toronto, Canada, has established the Toronto Innovation Acceleration Partners program, which connects startups with public-sector partners and university researchers to accelerate the development of civic-minded innovations. In the last few years, the program has not only turned Toronto into Canada’s new startup hub, but also led to dozens of AI projects in local government. Here as in most of our cases, all parties benefit from each other’s involvement.

Governance principles

Lastly, every step of the way, cities must enforce ethical AI principles.

First, municipalities should establish AI oversight boards to audit data sources and algorithms for accuracy and bias. Every agency will need clear processes to make sure that the datasets used to train AI are representative and don’t reinforce historical inequities related to demographics like race, gender, age, or ability status.

Second, every AI system interfacing with residents must be tested to ensure accessibility and inclusiveness across languages, dialects, and cultures. And any AI decision-support system must remain under human oversight for local leaders to retain full responsibility for their decisions.

Third, cities will need to obtain consent and protect the privacy of their citizens. Privacy protection should be engineered from the ground up rather than tacked on. Citizens deserve transparency into what data is collected, how it is secured, and how it is used.

Fourth, cities need to get citizen input along the way. At a minimum, they should organize grassroots campaigns to demystify AI, foster technological literacy, and understand popular concerns. At best, cities will also bring citizens into the development and deployment process – drawing on their feedback to ensure that the technology serves the community on its own terms.

With all this in place, cities can build sustainable models that inspire confidence rather than fear or skepticism. Only by putting AI ethics at the center of their approach can cities prove that they can be “smart” and humane at once.

A Call to Action

AI has the potential to reduce the size of bureaucracies and reinvent the way in which local governments make decisions, deliver services, and serve their citizens. But only with the right strategy can this potential translate into real solutions.

In some cities, the change will be evolutionary: Local leaders will seek incremental improvements, building AI on top of existing structures to deliver value sooner rather than later. In other cities, the change will be revolutionary: Local leaders will reimagine existing structures as if they started from scratch. Either way, AI can inject a culture of innovation within city hall that spills over far beyond city hall. If American cities heed this call, they will make their way back to the top of the smart cities rankings.

Ultimately, only with the right kind of leadership can cities harness the capabilities of AI. If technical acumen can open new doors, only local leadership can ensure that these doors lead to a better place. With more imagination, and less fear of innovation, cities can act as the architects of a new social contract between citizens and their communities.