How Integrated Data is Transforming Homelessness Response in San Francisco

In this episode, host Professor Stephen Goldsmith is joined by Amanda Ford from the Mayor's Office of Innovation in San Francisco to discuss how her team is successfully breaking down data silos to improve services for the city’s most vulnerable residents. Amanda provides insights into the Office's innovative approach, which includes leveraging integrated data systems, fostering trust and collaboration across agencies, and a willingness to take risks. She also shares how this work has successfully helped the "high utilizer" population.

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

Betsy Gardner:

This is Betsy Gardner, senior editor at Data Smart City Solutions at the Bloomberg Center for Cities at Harvard University. And you're listening to the Data-Smart City Pod where we bring on top innovators and experts to discuss the future of cities and how to become data smart.

Stephen Goldsmith:

Welcome back to the Data-Smart City Pod. This is your host, Stephen Goldsmith. I'm a professor at the Harvard Kennedy School and director of Data Smart City Solutions at the Bloomberg Center. Today we have a pleasure of talking with Amanda Ford, Data Science Program manager for the Mayor's Office of Innovation at San Francisco City Hall. We're going to discuss her work using data to address the persistent and complex issues of homelessness. Welcome, Amanda.

Amanda Ford:

Thank you. Great to be here.

Stephen Goldsmith:

So, before we get into the particulars of the project, tell our listeners a little bit about your background into city hall. You're an astrophysicist? I'm not sure how that exactly fits, maybe you can explain it to us.

Amanda Ford:

Yeah, so my academic background is in astronomy and physics. I graduated with an astronomy degree from Harvard back in 2001, and then I worked in the public sector in DC and did some science work and then made my way to graduate school at University of Arizona. So, my PhD is actually in astronomy, and I worked on galaxy evolution, and I worked a lot with large scale simulations. We were trying to understand “how does the universe work,” so we would make a bunch of these simulations with a bunch of different sort of physical mechanisms, and then we had the honor of comparing them with data from the Hubble Space Telescope. So, I found that really interesting, not just what can we observe, but how do we kind of understand the physical mechanisms behind them. So that's my academic background. And then in 2015, I was doing a postdoc overseas at the Max Planck Institute in Munich, and that was really exciting, but I started to want to be able to apply my skills in a more concrete way to help people more immediately.

And at that point, as big data was really starting to be on the rise, there was a lot of interest and excitement around data and data science, and I started to realize that the skills I'd built as a scientist, coding and working with simulations and visualizations and statistics and all of that. And I think importantly, drawing inferences from really limited data were things that I could apply in other settings. So, I came to Silicon Valley, I worked at a couple startups. I worked in agriculture, I worked in healthcare, kind of in that vein of big data, and I made my way over to Google. I was there for four years and became a little bit involved in some of Google's efforts in the civic space. So, I worked on Google Maps, they have tools for city planners. I worked a little bit on a covid health equity tracker, and so when I saw that the Mayor's Office of Innovation was starting up with the Bloomberg grant, I was really excited about that. I've been here since 2022.

Stephen Goldsmith:

Great. Your work, I think the Mayor's Office of Innovation itself in particular, your work is funded by Bloomberg Philanthropies, is that right?

Amanda Ford:

That is correct, yes.

Stephen Goldsmith:

We have an interest in things Bloomberg as well since we're in the Bloomberg Center at Harvard. Let's turn to the program we're talking about today. Your data science work, tackling issues of homelessness in particular. Before we get to the pieces, what is the problem that you were trying to solve with data specifically inside homelessness? What were you trying to accomplish?

Amanda Ford:

Yeah, so the problem we were trying to solve for is how to make the system work better as an integrated system. So, in San Francisco, we have many different teams that are specialized and go out and provide services to people who are experiencing homelessness. On the street. We have at least four different departments and over 12 different teams that go out and do that work. They do exceptional work, but the problem we were trying to solve for is they often kind of don't know what the other teams are doing, and it makes it very difficult for them to coordinate care and kind of provide a holistic best quality experience because sometimes one team will not know what the other teams have already done, and you kind of have this data silo problem that we were trying to solve for with the goal of providing better services and better care for people in San Francisco.

Stephen Goldsmith:

One of the things I've seen in some similar efforts is the perfect being the enemy of the good. How did you identify what data you wanted to bring together for this project?

Amanda Ford:

Yeah, so I share that perfect should not be the enemy of the good at all. And I think that was one of the advantages of our approach to this project is we were able to start really small. So, we identified the data that we wanted to start with, and we kind of knew that some data sets would be a lot harder to get others, so we just kind of worked off of momentum. So, we identified about six different data sets that we wanted. Two or three were pretty easy to get. We started with those and then we just tried to build some momentum and say, “hey, we have this three, this is what we've been able to do, let's go after that fourth” and so on.

Stephen Goldsmith:

Got it. Just before we get into a little bit more about the details, what were the jobs or names of the outreach workers or prevention workers or health workers? Who were your users in this project? Who were you trying to help?

Amanda Ford:

So, we had a couple different sort of end users or end clients in mind. So, the outreach workers you mentioned, those are a mix of emergency first responders, fire department workers, social workers, public health workers, people who go out and respond to overdoses. Also, people who coordinate long-term care. It was kind of a mix of those frontline outreach workers that was a very important audience for this work. We also had kind of a middle layer of program managers. These are people who are trying to figure out how do I empower my outreach worker team? How do I make sure that they're able to be as effective as they want to be?” And then we also had the audience of people like the mayor and the mayoral staff and the policy makers, people who were trying to figure out, hey, what is going on with our system and how do we make it better? And ultimately, of course, we were trying to serve the needs of unhoused people. So, what I mentioned where the city workers, all of us are trying to figure out how to help people better.

Stephen Goldsmith:

So Amanda, talk to us a little bit about the project itself, what you're trying to accomplish and how you manage the tension of the more information you have, the better the intervention will be for folks who need housing, but also the riskier the privacy questions.

Amanda Ford:

So, I'll talk a little bit about what our system is and then we can talk a little bit about the legal aspects and how we handle that privacy. So, what we have now is we have an integrated data system. As you mentioned, it includes encounter information for about 16,000 people over the last two years. That's over a hundred thousand encounters. So, this was information that already existed somewhere in the city, and our job was to sort of integrate it, make it easier to use, easier to access, and sort of empower people to go out and actually use it to help people. So, this was possible because of a new legal framework that existed in California. So, you might be aware that under HIPAA, we are permitted to disclose personal health information for treatment purposes without having to obtain authorization. This is referred to as the treatment exemption.

And in California in 2018, there was an assembly bill that kind of build on this understanding of housing as healthcare. So, it allowed us to start to share more information because housing was considered treatment. So, this AB 210 created a program and a protocol that we could use to implement and share with non-healthcare entities or non…sort of traditional healthcare entities if they were working on housing. So, this new kind of sharing network we came up with through the context of something we called the multidisciplinary team or MDT, this sort of allowed us to start to share this information. So, we started just by sharing it sort of in an analog way with case conferencing. So, if I was a fire department worker and I wanted to talk about a particular case with a department of public health worker, I could pick up the phone and call them and talk to them.

And then we were able to build on that to kind of automate this system to make it easier for people to share. I do think we were really mindful of privacy. There were certain things that we decided not to include in this data because they were sort of more sensitive. So things like HIV status, substance use status, domestic violence status, things like that, that are incredibly sensitive, we opted not to share, but our feeling was that that was the right balance because we really needed this comprehensive picture so that we could serve people really effectively. So for example, some of our shelter workers were saying, “hey, I'm trying to place this person in shelter, but they have a disability or a medical need that I'm not aware of because I don't have access to those systems, and so I can't put them in the right shelter so they're out on the street because I don't know that they have a particular medical need.” So that's the sort of problem and the balance we were trying to strike.

Stephen Goldsmith:

It's an interesting balance, particularly like the domestic violence one where folks may be without housing because they've been battered, right. It's an interesting set of tensions that you have to deal with. So how did the work of the street teams, outreach, medical emergency, how did their work change as a result of your data sharing efforts?

Amanda Ford:

Yes, there were a couple of changes. So, one of the first things that we noticed when we started to analyze this integrated data set there was kind of this…I'm going to say bimodality. In our population, there were people unhoused individuals who interacted with our system very, very infrequently. And there were people who interacted with our system very, very frequently. So just to give some numbers on that, we found that about half of the population had only been seen once, and about three quarters of the population had been seen three or fewer times. So really infrequently over about an 18-month period across these 12 different teams. And then on the other hand, we had this group of what we call high utilizers. So, they represented only about 9% of our total population, but about half of our overall encounters. So, these were people who over an 18-month period might be seeing 10, 20, 50, 70, one hundred times.

So, these are people who are really seeing a lot of street teams, but are still out on the street, obviously still needing care and still needing services. So when we were able to figure that out at scale with this data set, and that kind of matched the intuition of some of these outreach workers, but being able to put numbers on it, the first thing we did is we came up with a high utilizer list. So, we said, “hey, these are a list of people that are really needing a lot of services. Let's see if we can change our interventions.” And we were able to actually do that. So, the outreach workers went out, talked to people. There was one gentleman in particular who'd been seen by that point 67 times, and they were like, “hey man, how can we help you? We're kind of stuck here. What do you need?” And they were able to provide him better services and a more tailored experience. So, the last I heard, he has been living in shelter inside for quite a while, and he's on the path to a more stable situation. So those kinds of successes are really important and I'm happy to talk a little bit more about that. 

Stephen Goldsmith:

And how would that approach change – I mean, just in kind of imagining this from the street workers' perspective, they intervene. They found somebody who has been involved with the city before outreach efforts, and what do they do? They bring up the person's data on their mobile device and help decide what the right places for intervention, how does it actually work?

Amanda Ford:

Yeah. So, one of the things that we put together, we started the rollout in September of this year, was an outreach worker tool that was a mobile app. So, we'd been developing that over the summer, and it included a client history. So, you could pull that up on your phone if you were an outreach worker, you could say, “hey, I'm looking at Jane Doe right now. This is Jane Doe's encounter history and other relevant information about that individual.” So that might include things like if they have a history of psychiatric holds or a recent hospitalization or a recent overdose, things like that. And the feedback we got from the outreach workers is that was really helpful for them because if they were interacting with somebody and they didn't know that they had just come out of the hospital or they just have an overdose, they'd sort of approach them the wrong way.

So, they were able to tailor their approach and offer them different services or at least offer it in a different way. And we did do a pilot with people that we considered really, really high needs. These were people who were high utilizers and had recent overdoses and kind of met some other clinical criteria. So those people were kind of placed in a special pilot where we said, “hey, these people seem like they're really struggling, and they really need extra help.” And the outreach workers were able to go out and offer them kind of more tailored services, and we got some success from that pilot as well.

Stephen Goldsmith:

A lot of this I take, it's connected to supportive housing.

Amanda Ford:

So, we have a lot of different tools in our toolkit, which you might be familiar with. There's permanent supportive housing, there's shelter, there's different sort of options for people. But again, it kind of depends on where they're at. If someone has a severe mobility or physical disability, they're not going to be a match for all of the types of options that we have, which is why we have the different options and why it's so important to know somebody's history so we can kind of match them what they actually need

Stephen Goldsmith:

And does the care manager at the shelter have the same access to information?

Amanda Ford:

So that's what we're working on as well as making sure that we've done this rollout for everyone who needs it. So, we started with the outreach workers themselves because they're out on the street, and then we're in the process of figuring out who at the management level should also have access to this. As you mentioned, it's a little bit of that tension of expanding access while still keeping it very conservative from a privacy perspective. Our goal is not to have this information available to absolutely every employee in the city. So, we're trying to strike that balance.

Stephen Goldsmith:

Let's spend a minute or two before we close extrapolating the principles, right? If you were telling other cities how to do this or you were looking at other areas where integrated data will lead to better solutions, what are the principles? What are the lessons you learned? What were the problems you experienced that others might benefit from knowing about?

Amanda Ford:

I think, and maybe this will sound a little weird at first but stay with me and hopefully I can convince you. But I think one of the main principles that we had was the freedom to be wrong at first. So that was an advantage of having this external funding source through Bloomberg. We could afford to be wrong, at least in the beginning and sort of quickly iterate and move forward. So, what I mean by that is, in the beginning, we had some good buy-in from some of our stakeholders, but some of our stakeholders, we had some skepticism and some resistance towards data sharing for all the reasons you mentioned. Plus, it's hard to share data. Plus, they were worried their data would get misinterpreted or they would somehow get dinged for it. And so, we kind of took this approach of, "give us whatever you have, we will make our best guess on interpretation. We'll share that back with you. We'll give you an opportunity to give us feedback before we move it forward." And that really helped just kind of that freedom to be wrong, to say, this was my guess of what your data means and let me know where I'm wrong. That was really helpful. That helped build trust. That kind of helped us work very quickly.

So, for me in my office, if I'm wrong about a data interpretation for two or three weeks early on before we've deployed a product in the field, it's not a big deal. But if someone doing street outreach, who's out working with people every day is wrong for a month, that has a real human impact. So, I think that's one of the principles in government that I would like to see more of is, can we build spaces where it's okay to at least temporarily be wrong? It's very hard to be right a hundred percent of the time. So how do we innovate? How do we learn from whatever we didn't know before?

Stephen Goldsmith:

Well, if you're right one hundred percent of the time, you're not innovating, right? 

Amanda Ford:

Exactly. 

Stephen Goldsmith:

And I suspect having venture capital, innovation capital to try out the process helped a little bit internally in San Francisco City Hall.

Amanda Ford:

Yeah, it was definitely nice to just say, we're here. We're not taking up resources from anybody else. We're not carving anything out from any other daily government work. And so, we do have the space to innovate, and we do have the space to try some things that might be a little too risky for another department to feel comfortable taking on just due to the incentives and government otherwise.

Stephen Goldsmith:

If you think about innovation, obviously it's a continuing process where you learn, you challenge what you've learned, you build new things. How have you incorporated that process and what you've done?

Amanda Ford:

Yeah, so one of the things that I think was really successful for us on this project is as we were building this data set, we were really in touch with the data program leads, but we also wanted to make sure that once the data set was built and complete, at least in its initial version, we gave our stakeholders an opportunity to review it and get feedback before we launched the outreach worker tool. For example, before we built some of our other data products before we started publishing the results for policymakers, et cetera. 

So, one of the ways that we did that is we had what I call a hackathon/review-athon. So, we're all familiar with hackathons. We were able to give our data leads from our various partner departments access to our full data systems, but we also said, “this is a review opportunity as well. So here are the things that I'm going to bring forward. Here are the slides, here are the conclusions that I've drawn. Tell me where I'm wrong. Tell me if you agree, tell me if I've misinterpreted something along the way.” And that was really helpful in building trust because we were super transparent about what we were going to do with our data, and I think they really appreciated the opportunity to say, “hey, I gave you this data. It's really complicated. And actually, there was a misinterpretation here or there,” and that gave us a little bit of time to fix it. So, we actually did a sort of two week span of that hackathon review with on and then extended it a little bit. And I feel like that was really helpful in just getting that buy-in and then also making sure that we all felt comfortable that what we were deploying out in the field was accurate and had been interpreted correctly.

Stephen Goldsmith:

Well, it's really remarkable how much you've accomplished not just with the data, but how many people have been helped as a result of this process. Congratulations on your terrific work. This is Steven Goldsmith, professor at the Harvard Kennedy School and director of Data-Smart City Solutions. You've just heard from Amanda Ford, who is spearheading an incredible data informed approach to tackling homelessness in San Francisco. Amanda, thank you so much for your time today.

Amanda Ford:

Thank you so much for having me.

Betsy Gardner:

If you liked this podcast, please visit us at datasmartcities.org, find us on iTunes, Spotify or wherever you get your podcasts. This podcast was hosted by Stephen Goldsmith and produced by me, Betsy Gardner. Thanks for listening.