tl;dr

• We won the Bayes Impact Challenge with a bot that connects people to parks.
• We believe great, accessible applications that reach users where they are can encourage government to invest in open data.
• Talk to the bot!

Here at Trailhead Labs, we constantly push the limits of technology to connect people to the outdoors. Our mission is to guarantee everyone has a chance to experience the incredible benefits of getting outside.

Last month we entered the Y Combinator/Gates Foundation-funded Bayes Impact data science contest that paired federal agencies with teams of engineers to tackle problems with data. Our team (with our partners at Booz Allen Digital and NIC) worked with the Department of Interior on the problem of underutilization of parks by minority groups.

Reaching underserved communities can be a challenge, and poor user interfaces are an additional barrier. We decided to meet the audience where they are, and overwhelmingly they are on Facebook. The mission of Bayes Impact is to make a difference at scale, and it’s hard to ignore the scale of Facebook Messenger:

Fortunately, Facebook has just launched the Messenger Platform at its f8 conference. We chose to build a bot to help connect people to parks that have things they want. The diverse demographics of Facebook users is tantalizing because it helps us reach our goal of connecting underserved communities to parks. A recent study found African-Americans are six times more likely to check in to a location on social media than the general population while Hispanics are twice as likely to check in and three times more likely to message their friends on Facebook. Nothing rivals Facebook for engagement with these groups.

Data availability challenges

To help people get to the appropriate parks, we at least need some very basic information. Our main source of data was a federal database of approximately 4,000 recreational areas, 25 percent of which was not usable because the records lacked geographic information. In many other cases the metadata was incomplete so matching people to appropriate parks was impossible. However, by using some geocoding and data cleaning techniques we were able to provide quality results.

But our bot only knows about federally-managed parks — and that’s not enough. Many people do not live near them and may not have the means to get there. We want to get people to the right park regardless of who manages it.

The experience should be seamless. Imagine visiting Golden Gate Park which is managed by the city of San Francisco. We would like to be able to direct you to the nearest drinking fountain. Later, you stroll over to the nearby Presidio which is managed by the National Park Service. We would like to be able to direct you to a restroom there. When you are looking for a drinking fountain or restroom, you don’t care who manages the park.

To make this work, we need consistent and comprehensive park data across all governmental agencies.

We want to provide the best possible experience. To accomplish this, we need three levels of data:

1. Basic identity including names and park boundaries.
2. Metadata that includes facilities, activities, accessibility, hours, images etc.
3. Real-time trail conditions, facility availability and event information.

Today, even just goal 1 is still a challenge. Collecting the data from disparate governmental agencies on a case by case basis is painful and utterly impractical. Using existing open source material such as Open Street Map is promising, but normalizing and validating the data is nontrivial.

Achieving goals 2 and 3 requires collaboration among governmental agencies and between them and application developers. We hope our bot will provide an example to help encourage everyone to cooperate to make our goal of universal park access a reality.

And now if you like, say hello to Raibot.