When you launch a new nonprofit, regardless of its goals, it’s still rare today that the first two hires are a chief technology officer and a chief data scientist. But that was the smartest move possible when DoSomething.org CEO Nancy Lublin created Crisis Text Line, a service that counsels troubled and suicidal teens by text message.
Today, more and more nonprofits have data that they can use to improve their services and advance knowledge in their field. And Crisis Text Line is a good example of how to go about using it.
The Crisis Text Line team knew from the beginning that, if it was successful, it would be collecting more data than ever before about teens in trouble. Research has shown that it’s easier for teens to open up via the anonymous texting platform, rather than a more intimate phone call. But beyond helping individual teens, Lublin’s idea was that Crisis Text Line could pool all of its data to advance the crisis counseling field, at a time, today, when suicide was the third-leading cause of death among youth age 10 to 24.
Crisis Text Line is, indeed, swimming in data. Since its launch in 2013, even with minimal marketing, it has had more than 87,000 conversations and almost 7 million anonymous text message exchanges with teens. Every day, Crisis Text Line counselors deal with 20,000 incoming text messages. Conversations can last almost an hour.
“We see an opportunity to provide people in the crisis space with data they’ve never had access to before about how crises are happening in the moment,” says chief data scientist Bob Filbin. “They’re been survey data, but it’s a much smaller sample size.”
When the nonprofit launched, Filbin spent the first six months traveling the country to visit 12 crisis centers and interview more than 100 counselors. “One of my earliest insights is that the crisis counselor is actually our primary user. The texter, they’re just one their phone,” he says.
Much of the initial focus has been improving the experience of the counselors, who are all trained volunteers who work from home through Crisis Text Line’s web interface. Expanding the service from 20,000 messages per day to the goal of 100,000 a day by the end of the year will require using their resources even more efficiently.
For help, Filbin recently turned to the organization DataKind, a nonprofit that connects volunteer data scientists with nonprofits through hackathons and longer-term projects. Through its DataCorps program, DataKind paired Crisis Text Line with Noelle Sio Saldana, principal data scientist at Pivotal, a big data software company in Silicon Valley. She was going to be the first fellow in the company’s new “Pivotal For Good” program, which sends employees on a three-month sabbatical to work with a social good or nonprofit organizations, pro bono. The partnership with DataKind made sense for Pivotal’s first project.
“There’s all these things, anecdotally, that we can characterize about people when they are in an emotionally dark place,” Saldana says. “Is there a way to automate that? Is there a way to make that not just anecdotes, but make it data-driven?”
The challenge Saldana took on was an interesting one: It’s a version of what some in the business world call the 80-20 rule, or the Pareto Principle, where the majority of effects (80%) come from a minority of causes (20%). For Crisis Text Line, their version was that 3% of their texters were super users. They continually came back to the service, using it more as a regular counseling session than truly for the crisis moments the service is meant for. These small number of people were using 34% of the counselors’ conversation minutes.
“It’s called circling–they are not progressing through a problem,” says Filbin. “Clearly, we’re not the right service for them. A lot of these people need to be in longer-term services.”
Working mostly remotely over three months, Saldana wrote code to dig through the data. The goal was to identify repeat texters as early as possible, so they could be referred to other services, like therapy, at a sooner point in the process. In the end, her algorithm could pick out repeat texters as early as their fifth conversation rather than their 20th. Since each conversation averages around 40 to 50 minutes, that’s a lot of time saved. After her work was implemented, 3% of “repeat” texters were only using 8% of conversation minutes.
Over time, the goal of Crisis Text Line is to open up more of its data to researchers, journalists, and policymakers in a safe, private, and ethical way, in order to gain insights about teens in crisis. It has already mapped its data by state, day of week, and time. Eventually, when it has had enough users that it doesn’t risk violating any one person’s privacy, it might further break down its data by area code. It plans to create an independent ethics committee to advise it on how to best share conversation-level data with trusted researchers or organizations like the Pew Research Institute or the CDC.
For Saldana, it was a new experience to work for a social good cause.
“The vast majority of my clients are large enterprise companies. I have no shame in saying that what I do is to help optimize ad clicks and revenues for these companies. It’s exciting work that we get to do in terms of doing data science,” she says. “But it was so refreshing [to work with Crisis Text Line.] There’s a big feel-good part of it, and there’s amazing things that you can do from the data science side. The other part of it was realizing that they, too, are data-driven.”
Aside from Pivotal, a few other Silicon Valley companies are starting to contribute to data science-based volunteer efforts, such as LinkedIn and Palantir. DataKind wants to make these efforts easier, and it now has chapters around the world. Days DataKind founder Jake Porway: “I want pro bono to be a part of data science, like it’s a part of law,” .
The organization is still working to set the right tone, especially for nonprofits that are less data-savvy than Crisis Text Line. “I would be mortified if people though we were data science for nonprofits. It’s easy to think than, and it’s not that far off,” says Porway. “But to me that’s an image of exactly the problem that plagued international development: let’s swoop in when you need us. DataKind, to me, is much more about building collaborations and saying ‘Here are the skills we have to contribute. Let’s humbly team up with others to say where might we be of service.'”