Bloomberg had hired someone to look at exactly that problem. Michael Flowers was a key player in the administration’s smart data campaign. A former prosecutor and Justice Department lawyer in Iraq, Flowers joined New York City government in December 2009 as head of the city’s Financial Crimes Task Force. Among other tasks, he investigated mortgage fraud and learned, he says, that “the city knew a tremendous amount” about its people and businesses. ^[27] The task force’s responsibilities evolved as Flowers discovered that information used to track mortgage fraud could easily be adapted to identify other problems. By 2012, Flowers was director of analytics for the Office of Policy and Strategic Planning in the mayor’s office.

MODA . On February 14, 2013, Bloomberg announced the creation of the Mayor’s Office of Data Analytics (MODA), a small team within City Hall that would synthesize data from 40 different city agencies in an effort to solve problems that spanned the responsibilities of—and the information collected by—those agencies. In announcing its formation, Bloomberg said MODA would “launch a new platform that will improve the way all agencies share information.” ^[28] To lead this effort, he appointed Flowers as the city’s first-ever chief analytics officer.

Flowers and his team of about six looked for innovative ways to use data to solve problems. They relied on the insight that data relevant to the performance of one city agency might be housed in another. Flowers believes that the challenges to sharing data across agencies fall into four broad categories—“technical, cultural, political and legal, in no particular order.” Legal concerns included citizen privacy and statutory limits on the authority of certain agencies. More nebulous and potentially nettlesome were the cultural and political hurdles to changing the way city bureaucracies worked. Flowers explains why change is hard:

Bureaucracies are expressly designed to be resilient. That’s why they exist, because we want them to be able to handle the vicissitudes of elected government. It doesn’t matter if [the mayor] changes, because the trash still needs to be picked up... Moreover, tribal turf wars are real. They absolutely exist. Agencies get deeply invested in their subject matter areas… We want them to be deeply invested in their subject matter.

Michael Flowers: video.

Technology was, in Flowers’ view, the easy part—but that, too, was complicated. City agencies had developed indigenous systems for counting and categorizing the important features of their areas of responsibility. Getting a clear picture of the information housed in more than 40 city agencies was not simply a matter of combining databases. For example, different agencies had different ways to identify buildings. The Post Office used addresses. The Department of Buildings used unique building identification numbers. The Finance Department used the lot number for the land a given building sat on. Finally, emergency response agencies such the police and fire departments used latitude and longitude.

Flowers saw good reason for the differences in how each city agency handled its data—each method of categorizing a building, for example, at some point served a useful internal purpose. Not only would it be expensive and politically difficult to force each agency to move to a universal system; in Flowers’ view, it was not necessary. “The technology has advanced and the data science has advanced to the stage where the barrier to entry, to synthesizing these different systems, for purpose-driven reasons, can be effected rather simply,” he says.

Flowers worked as much as possible within existing systems at each agency. For example, if the Buildings Department had an inspection system, Flowers’ team would not try to revamp it. Rather, it would use information from other city agencies, such as the Finance Department, to help Buildings improve the order, rather than the manner, in which it conducted inspections. Says Flowers:

If I can demonstrate conclusively that if a property has a tax lien on it, and that the existence of a tax lien correlates with an order of magnitude increase in the likelihood of a catastrophic event at that location—one of the city’s most fundamental jobs is to prevent those catastrophic events from occurring in the first place, if they can… What this piece of information is telling me is that there’s a catastrophe more likely at this smaller subset of our one million buildings, and therefore I’m going to send my finite resources to that place first. That, without increasing the number of resources available to the Department of Buildings, dramatically increases their effectiveness.

There were some significant wins. Flowers’ City Hall group worked with Tobin and his team to jumpstart the Fire Department’s efforts to access and add data from other agencies to its risk model. In May 2013, FDNY’s digitized Risk-Based Assessment System launched as part of a massive, $26-million effort to improve fire prevention. The department saw the rate at which inspectors found serious violations jump from some 9-13 percent under the old system to 70-75 percent. New York City saw 47 fire fatalities in fiscal year 2013, a steep drop from 70 in fiscal year 2012. ^[29]

Flowers worked with other city agencies as well, on projects that ranged from identifying Medicaid fraud to tax violations. Even before MODA’s creation, in 2012 both the police and health departments had observed an increase in prescription drug abuse in Staten Island, specifically of the painkillers oxycontin and oxycodone. Flowers’ team thought it possible to identify which pharmacies were illegally distributing prescription drugs. It knew that the Human Resources Administration (HRA) was responsible for reimbursing Medicaid claims to pharmacies, and could audit pharmacies submitting Medicaid claims. But like any other city agency, HRA had limited resources—only a handful of auditors for some 2,600 pharmacies. Flowers and his team wanted to use data to maximize the likelihood that the auditors would find malfeasance. He recalls:

We did a basic analysis of the redemptions for those specific high concentration oxy, and were able to find that one percent—about 20 of the pharmacies—were responsible for about 80 percent, 90 percent of the [oxycontin and oxycodone] distribution, at least for Medicaid redemptions. Then, we further tested that by having HRA train their audit capacity on those pharmacies, and about 19 out of the 20 turned out to be up to no good.

The team used the same approach for other problems. In fall 2012, the city’s Department of Environmental Protection (DEP) enlisted Flowers’ team to help locate restaurants illegally dumping cooking oil into sewers—a practice responsible for the majority of the city’s clogged drains. It discovered that a city agency called the Business Integrity Commission was responsible for certifying that restaurants hired companies to dispose of grease. Using this information to identify which restaurants had not hired such services, and comparing it with data on sewers, the team advised DEP where to look for restaurants dumping grease illegally. By prioritizing the search for rogue restaurants, DEP found illegal dumping in 95 percent of the restaurants it inspected. “With nothing grander than public data,” a New York Times article on Flowers and his team later recounted, “the Case of the Grease-Clogged Sewers was solved.” ^[30]

Exit Bloomberg . As the Bloomberg administration drew to a close in late 2013, there were systems in place within some city agencies to use data for resource allocation based on informed predictions of where problems were most likely to occur. The approach had prevented crimes and fires, for example, at impressive rates. Meanwhile, data had been used to make city government as a whole more nimble, responsive, and efficient.

MODA, in particular, had helped overcome challenges to sharing information among agencies in order to solve cross-sectoral problems. But MODA did so outside the system, in an ad hoc manner and often on request. It was unclear whether the unit would survive. MODA had helped to build agency-level capacity to manage and analyze data, but it was a long-term process. Should it continue? Was this the best approach? What, ultimately, was the correct balance of centralization versus decentralization in the use of data for governance?