Use data to pay closer attention to delivering the services New Yorkers care most about.
New Yorkers are increasingly dissatisfied with the quality of life in their city. According to the Citizens Budget Commission’s 2025 resident survey, only 34% of New Yorkers rated the city’s overall quality of life as excellent or good, down from 51% in 2017. Just 22% said they felt safe on the subway at night, compared with 46% in 2017, while only 27% gave city government’s services a positive rating, down from 44%.
This dissatisfaction cannot be explained by a lack of resources. New York state and its local governments collect more taxes per capita than any other state and rank second nationally in per capita spending. In fiscal year 2024, they collected more than $12,000 per person — 22% more than second-place California — and spent at levels far above the national average.
Yet the city continues to struggle with basic quality-of-life problems. The FY 2026 Mayor’s Management Report, for example, records a 6% increase in 311 total service requests from the previous year and a 26% increase from FY22.
The problem, then, is not simply how much New York spends. It is whether the City is turning those resources into the services, safety and quality of life that residents expect.
As I recently wrote in a report for Manhattan Institute, artificial intelligence, together with New York City’s preexisting data, offers major pieces of the solution; the question is whether we will take advantage of it.
At the core, every city has an identical difficulty to solve: How should we properly allocate scant resources? If we have a limited number of police officers, firefighters, inspectors, teachers, social workers and funds, where should we distribute them most effectively?
In answering that question, governments can choose to act either proactively or reactively. The most effective governments are preemptive, but of course, it is far easier to put out a fire than to anticipate where the fire is going to be in the first place.
The history of policing in New York City provides a perfect example of this tension. Decades ago, the methodology behind policing was quite different from what it is today. Prior to the CompStat era, policing was mostly guesswork and intuition. A crime would happen, someone would report it to the police, and the police would respond. In effect, the NYPD saw crime as single instances to respond to on a case-by-case basis. In that context, CompStat — which required police to collect and map crime data on a weekly basis, holding precinct commanders accountable for results — was not only a technological innovation but a methodological breakthrough. Rather than a series of single occurrences, crime became a series of patterns that could be measured and studied. With that change, the NYPD was much better able to forecast hot spots and trends proactively.
The reason forecasting trends is crucial for both safety and satisfaction is that it means that resources are likelier to be used properly.
This principle is true for almost any city service. Where should we send fire inspectors? Garbage trucks? Night patrols? Reading tutors? The list goes on.
This is where artificial intelligence truly presents a new way of allocating resources. Many city departments already use regressions to predict patterns. A regression is a statistical model that measures the relationship between an independent and dependent variable while still taking into account other influential factors. But regressions rely on clean, formatted data and many manual steps along the way. The best artificial intelligence models, though, are multimodal. This means that they can process different types of data in a single program. Therefore, rather than looking at a spreadsheet with millions of rows, we can use that same spreadsheet, but also old PDFs, audio and video, and have AI process all of it. This tool greatly expands pattern recognition.
Consider housing inspections. AI could analyze not only spreadsheets of past violations but also photographs, building records, tenant complaints and 311 calls to identify buildings at elevated risk of serious code violations or infrastructure failures. By synthesizing these disparate sources of information, the City could direct inspectors toward the buildings most in need of intervention, rather than relying primarily on historical violation data.
AI offers a way to search that existing information for relationships that human analysts might never have the time — or even the ability — to identify.
The beauty of this is that it doesn’t rely on new surveillance tools to implement. We already have millions of data entries logged and publicized on NYC Open Data. While New York City Open Data has thousands of datasets on almost every aspect of government services, it is not without its problems. First, it is difficult to access the data in a meaningful way. For most users, downloading gigabytes’ worth of spreadsheets without any sort of tool to measure changes, sort and filter is unhelpful at best. Second, many datasets remain unautomated and therefore outdated. Additionally, as David Tussey, former executive director of the Department of Information Technology and Telecommunications, has found, government data often contains hidden errors, such as failing to account for daylight saving time, and lacks uniform standards across departments.
Utilizing AI can actually help fix many of these issues. First, it can automate and clean data. Second, it can help to create more sophisticated regressions with visual dashboards from the existing spreadsheets while incorporating previously unusable data. Third, we can make sure that AI helps us keep these data up to date, as many of the datasets are unautomated.
A perfect example of applying this methodology to improve outcomes occurred in 2024 in the medical world. Joseph Coates had a rare blood disorder called POEMS syndrome, which impacted his ability to move and stay conscious. He was given the option to die either at home or in the hospital. A doctor named David Fajgenbaum, however, at the behest of Coates’ girlfriend, suggested a novel mixture of chemotherapy, immunotherapy and steroids. In four months, Coates was healthy enough for a stem cell transplant, and his disease was in remission soon thereafter.
Fajgenbaum’s breakthrough came from an artificial intelligence program he created that repurposes drugs for diseases they were never intended to heal. Fajgenbaum created a platform that cross-references roughly 4,000 drugs against 18,500 diseases before it issues a probability score based on efficacy.
As Donald C. Lo, a scientific lead for Remedi4All, told The New York Times, “[there is] a treasure trove of medicine that could be used for so many other diseases. We just didn’t have a systematic way of looking at it.” Analogous to the case of city data, Lo reported that “these drugs are already approved. You can already buy them at the pharmacy.”
The idea behind these medical breakthroughs is really just another instance of pattern recognition. Every single drug has a side effect, but a side effect in one case is a cure in another.
The significance of this case is not simply that AI can discover things humans cannot; it is that the relevant information was already sitting there. What was missing was a practical way to examine the enormous number of possible relationships within it.
City government does not need to collect an entirely new universe of information. It already has an enormous archive of information about its streets, buildings, complaints, inspections, crashes, crime, sanitation, transportation and public services. The problem is that much of this information remains fragmented, inconsistently formatted and difficult to connect. AI offers a way to search that existing information for relationships that human analysts might never have the time — or even the ability — to identify.





