With citizen demands rising, America’s urbanization and growth continues to upend the traditional flow of government, a fact leaving officials disoriented but determined to uncover modern tools to expand and diversify services.
Data science is an essential ingredient in this pursuit. Its benefits read like a miracle drug: It’s an accelerant for decision making, it automates time-consuming tasks, and has potential to articulate and improve even the most nebulous of processes. Yet while data science is universally embraced in theory, California Chief Data Officer Joy Bonaguro said most organizations struggle to effectively identify and design projects.
She calls this challenge the “Truffle Pig Problem.”
Speaking via video chat at the California Department of Technology’s A.I. Community of Practice event last month, Bonaguro said truffle pigs are famous for their keen snouts that sniff out truffles, the mushroom delicacy that can be buried as deep as three feet into the ground. When the menu calls for data science, she said the Truffle Pig Problem occurs when an organization can’t sniff out useful projects.
“Sniffing out a good data science project is the single greatest barrier to data science adoption,” Bonaguro said.
If government agencies feel overwhelmed, Bonaguro said it’s a common problem and there is ample of evidence of poor data usage across sectors. Highlighting the 2022 NewVantage Partners Survey, which measures private sector adoption of big data and A.I., Bonaguro said the survey showed two thirds of its respondents investing in data science yet only a small number, 26.5 percent, reporting they actually use data to make business decisions. Further, she said nearly 92 percent of respondents said the challenge lies in people, processes, and culture and not technology.
“They characterize it as a technology push rather than a [technology] pull from humans who want to make more data-based decisions,” Bonaguro said.
To harness data in cities and states and avoid a Truffle Pig Problem, Bonaguro looked back at her experience as San Francisco’s chief data officer to provide a few pieces of tactical advice for governments.
1. Set realistic expectations
When she proposed her first data science projects in San Francisco Bonaguro said that internally, they weren’t always met with standing ovations. Some viewed the projects as resource hogs, others saw them as redundant efforts, others dismissed them as plays to be trendy. There was also a small set of data enthusiasts that saw them as the solution for everything.
What helped get projects moving, she said, was setting realistic expectations and defining the data science project. Bonaguro said she made sure departments understood the projects were not designed to be an overhaul of department operations, but would likely be tied to small changes. Projects would not seek to find new data, but look at data already available. And for success, projects required the support of a department.
“In order to counterbalance this range of reactions what we needed to do was define what we meant by data science and how it fit into the construct of other practices that were already in play,” Bonaguro said.
2. Make it about service change and business goals
To have a real impact that delivers lasting results, Bonaguro said it’s crucial data projects aren’t exercises in academic research. They need to influence real services, workflows, and operational tasks.
Yet to hit this target, she said a data team or specialist should be prepared to meet regularly with the departments they hope to help to learn how and why processes exist, and also to verify what the data actually means. In San Francisco, Bonaguro would frequently present findings and data models to departments to gain feedback and to understand how her data models could be improved.
“You can’t just have a model and then it just sits there,” Bonaguro said. “You need to embed it into the business process.”
3. Don’t go to them, let them come to you
Equally important for real service change, Bonaguro said it’s important that data science be seen as an opportunity for departments instead of a burden. In fact, for data science to work, she said departments in San Francisco needed to understand data science was a privilege not an obligation. This mindset was for the department’s benefit as much as it was for the city’s data program.
She explained that successful data projects need committed clients – in this case department leadership and staff – to champion the changes in workflows and processes that would follow. If a department started out feeling like the data project was a hassle forced on them, or that they were doing the data science team a favor, there was a high likelihood the end result, no matter how beneficial, wouldn’t last. To change this dynamic, Bonaguro and her team created an application process for data science projects that required interested departments to apply for help.
“This is really about leveraging concepts like scarcity and selectivity. All of that led to what we ended up calling the Anatomy of a Committed Client,” Bonaguro said. “Because again, we’re aiming for service change…and you need a committed client to actually realize the insights of your data science project.”
Data Science in Action
As example, Bonaguro called up a success story when she assisted San Francisco’s Assessor-Recorder to eliminate a backlog of property assessments. Over time, the city’s growth and development had overwhelmed assessors with a glut of requests. Bonaguro said full property value assessments were time intensive and the assessors—who had to assess property value to estimate property taxes after every sale — just couldn’t keep up. Worse still, property tax revenues represented a third of San Francisco’s general fund.
“This was a huge financial issue for the city,” Bonaguro said, stressing that the backlog had tied up millions of the city’s tax dollars.
At the request of the Assessor-Recorder, Bonaguro and her team went to work, first by meeting with key staff to understand the assessment process in depth, and then analyzing the city’s property assessment data to generate a model to flag properties that likely needed a full assessment from those that just needed a short review.
What resulted from the effort was an algorithm that could evaluate a property based on its type and characteristics and generate its estimated value.
“We actually put this model into production,” Bonaguro said. “So, it was a living, breathing algorithm. And what it did is generate a predicted price and the service change was to say, ‘Hey, does that sale price fall within a reasonable range.’”
If it fell within a reasonable range the property would only require a quick review. If the property did not the city did a full assessment of the property to determine its exact value and the resulting property taxes a new owner would pay. In the end, Bonaguro said this focused analysis completely eliminated San Francisco’s backlog while breaking down funding constraints.
“The end result was that the very first run of the model…reduced the backlog by 10 percent, which allowed the city to immediately access $239 million in roll value (the combined value of properties) which translated to 2.8 million in property tax revenue.”
While the data project was a big win for the city, Bonaguro stressed the factors that enabled success were not big ambitions, but rather small, specific and collaborative efforts. Bonaguro’s team did not overhaul the city’s assessment process, nor did it hunt for new and illusive data, nor did her team require super computers or advanced A.I. Instead, she said her team looked at the fundamental components of the problem and then worked with the data tools they had and within the systems and processes already in place.
“It’s often about small changes. We’re not necessarily making a major overhaul or disruption…,” she said. “It’s about service science. It’s about service change.”