Jawbone's Head of Data Science and Analytics on the Future of Wearables and Analytics Insights
Jawbone is one of the most well known wearables companies. Their latest product, the Jawbone UP3, is one of the most advanced fitness trackers on the market. Evolving from that, Jawbone’s CEO recently announced that the company is thinking of itself as not only a hardware company but also a software and data company. Brian Wilt, Head of Data Science and Analytics at Jawbone, recently gave a presentation at QCon SF about how data science and machine learning are helping shape a successful product. Following up from the presentation, we asked for his opinion around the future of wearables and Jawbone.
InfoQ: Fitness trackers have gone a long way from simple step tracking to multiple metrics. Do you see the future in more sensors or better understanding of the existing data?
Brian Wilt: That’s a great question. At Jawbone we focus on the experiences we’re trying to build. For example, understanding cardiac health and fitness, instead of the “hammer in search of a nail” approach. New sensors and models are both important and complementary pieces which serve these experiences.
InfoQ: Adding to that, would you see any benefit in tracking device’s raw data instead of compressing and transmitting them? i.e. is there loss of information when you compress the UP3’s data?
Brian: Yes, having raw uncompressed data is helpful for things like understanding data pipes during development and open exploration. But in a power-constrained production environment, you want to minimize data transfer to save on battery and improve latency in service of the user experience. So we deploy data products in all parts of the stack -- firmware, app, and server -- which operate on just the right amount of data, from raw voltages from the sensors to “digitized” steps. There are several other considerations when deciding where to deploy data products in the stack (for example, development/deployment time). That’s how we think about the problem.
InfoQ: Fitness tracking has been great in giving us data and numbers back but not so great in giving actionable insights. Do you see a future in fitness trackers that will integrate with other aspects of our lives, giving advice about eating patterns or work habits based on bioimpedance sensors feedback?
Brian: It’s true, wearables need to do so much more than tracking. Our Smart Coach software gives actionable insights, for example, “Your heart rate was high last night, implying you’re dehydrated -- drink more water!” Additionally, with goal setting mechanics and pledges in the app (we call them “Today I Will”s at Jawbone), we’ve seen significant improvements to our users’ health. In one experiment we ran, a group that was prompted in the morning to commit to going to bed early slept 23 minutes more and was 72% more likely to hit their goal. We’re using the same techniques that Google and Facebook use to get people to click on ads to improve their health. That’s the power of data.
InfoQ: Sleep research at such massive scale, but also down to individual level coaching like you mentioned in your talk about Golden State’s player Andre Igoudala is something that would be unthinkable of in the past years. What should we expect to see around sleep data in the future?
Brian: The Andre Igoudala study was illuminating because the statistics and correlations were so easily quantified. We believe strongly in the importance of understanding sleep in service to building a complete picture of our health. Tens of millions of Americans suffer from sleep apnea, with the majority of these cases going undiagnosed. There’s clearly value in wearables being able to reveal patterns of poor sleep or, perhaps in the future, being able to diagnose and manage chronic diseases.
InfoQ: Do you believe that there is room for doctor-interpretable fitness trackers data? Currently smart patients present fitness data to their doctors but most of the times doctors don’t care about them. Is there a point at which we will be able to summarize data into insights to a point that doctors will find it useful?
Brian: One day, yes. Part of our goal is to estimate higher-level measures of health -- for example, not just counting steps, but computing caloric burn. Not just bedtime and waketime, but amount of deep and REM sleep. Of course, the holy grail is to one day predict and manage conditions like heart disease, diabetes, and mental issues, closing the loop between measurement and behavior change. And there’s massive financial incentives: fitness and weight loss are billion dollar markets, health care is a trillion dollar opportunity.
InfoQ: Data Science as you said, means different things to different companies at different scale. Do you believe that interpreting and understanding data will converge with predictive analytics to the point that small and large teams will be essentially working on the same problems, at different scale?
Brian: The tools from data science are useful at many scales. An article I really like describes the evolution of data science at Twitter into specialized roles (analysts + builders) and adapting to different company scale. Statistical and machine learning techniques have been known to the predictive analytics community for decades. But what has been so exciting in data science is that combining these techniques with big data directly from your product yields game-changers like Google’s PageRank, Facebook’s News Feed, and LinkedIn’s People You May Know.
InfoQ: What other cool projects are you currently working on in the Machine Learning space at Jawbone? :)
Brian: At Jawbone we have two principal use cases for machine learning, the “analog to digital conversion” from raw sensor data to health measures like steps, and understanding relationships between health metrics (e.g., how quality of sleep and heart rate are related, or sleep and energy level). We’re always looking for great people up for the challenge, drop us a line if you’re interested!
About the Interviewee
Brian Wilt leads data science and analytics at Jawbone, making data human. He drinks his coffee black but loves lattes, which doesn't really make sense. He studied physics and neuroscience at MIT and Stanford (go Card). Follow him @brianwilt.
Big Data vs Small Data
Re: Big Data vs Small Data
I think this is amazing.