29 Oct 2015

The future of activity data

Recently I was interviewed for ReadWrite in a post about the future of activity tracking and health data. Because it wasn't a post all about me, most of my answers weren't used, so I think it's worth publishing my full response here to give you an idea of what I'm excited about and what we're planning for the future.

In broad terms how can developers look to start creating something unique from this deluge of raw activity and tracking data? Where are the biggest opportunities?

I think there are so many opportunities to do great things with this data that nobody has really nailed yet. One of the big things that we're currently tackling is finding meaning in the data — taking raw numbers, aggregating and analysing them, and turning that into trends and correlations.

Fundamentally, activity trackers are stupid — they react to physical activity and return you a number. If you look at this as low-level data, you see there's so much opportunity to build on top of that to produce "high-level data". It's just like how higher-level programming languages are built on low-level languages, abstracting away lower-level concepts and allowing you to do more with less. We can do the same thing with activity trackers and their streams of raw numbers, turning things like step counts into more meaningful information about your relative progress against your average day, for example.

Another angle I'd love to see happen is using this insight into your activity and your habits to build intelligence into other services. When you know the user's wake time and bedtime, where they tend to check-in and when, events they have on for the day, and other "time and place" cues, you can start making pretty good guesses as to where they'll be at a certain time and what their intent may be. I'd love to see something like Google Now take advantage of this to silence your phone when it knows you're in a meeting, or switch to only priority notifications when you're out for a run, because it's useless to be notified of a new reply on Twitter when your phone's strapped to your arm. And this sort of thing isn't exclusively the domain of Google. Your music player could know what genre to suggest based on the time of day, your mood, and your current productivity levels, or your language learning app could go easy on you if it knows you had a really bad sleep last night. To some people, this sounds like a dystopian hell, I know! But I'm excited about all the opportunities here that people could be building right now.

Finally, I really want to see someone perfect food tracking. Manually tracking everything you eat by searching for it in a food database is so tedious that barely anyone can stick with it. There are opportunities to use photos to recognise food, check-ins and location to know where I'm eating, even natural language interfaces instead of "search and select". I'm looking forward to this becoming much smarter.

What's Exist's approach? How do you go about creating something that stands out?

We start by aggregating the raw numbers from activity trackers and other sources like social media, productivity trackers, calendar events, etc. We then take rolling averages for each of your metrics and use these as a base for predictions about your day. We have this idea of using your day of the week average as a goal to compete against, instead of a static goal of 10,000 steps or whatever you first set, like Fitbit or Jawbone UP would give you. So your goals vary depending on whether you're usually active on that day or not, and they change each week. The beauty of this approach is that as you hit your goal more, your average goes up, and your goal gets harder! The same thing applies if you have a particularly bad week. I don't know about everyone else, but this works really well on me. I love checking my goal in the morning to see how high it is and working out how I'll hit it.

On top of that, we look for particular events in your data and generate interesting insights. These can be things like streaks of hitting your goal, particularly good or bad days (like "Yesterday was your lowest steps day for a week"), and longer-term trends of increases or decreases in your stats.

Finally, we combine all this with correlations between your metrics, so we can find what things might go together. It's up to the user to determine the true cause of correlations like "You're less productive when you tweet more", but we've found it to be a great jumping-off point for users looking to better understand the relationships between things they do.

In future we're pushing towards using all this data to predict users' days and help them with actionable feedback. One thing we've just released is a real-time steps insight that gives feedback on how likely the user is to hit their steps goal. Because we know the time they got up and we can predict their bedtime, we can monitor their progress and point out if they're going to need a little help. And in future we'll be pulling in those correlations and other metrics to make this even smarter. If you're less likely to hit your goal when it's raining, we can warn you in the morning that today's going to be a hard one because it's raining all day. I'm really excited about all the things we can do in this area, but there's a long way to go.

What sorts of extra data would you like access to or might we see access to in the future? What more could Apple, Google, or Jawbone do?

I think as the IoT ecosystem matures we're going to see so much more data coming out of these devices, and that's going to spur an even greater need for tools that give that data meaning. Alongside the rise of cheap, ubiquitous Bluetooth location beacons, we'll see new experiences based around ultra-local location and interaction tracking — knowing where you are and what you're looking at, and even who you're with. I hope it isn't all just used to track and market to us! I'd love to have this feed into Exist's understanding of what you're up to.

Another big issue I see approaching is the ownership of data. There are plenty of startups trying to help users "take back ownership of their data" by playing intermediary between the user's data and services that request access to it, but I don't think anyone's figured out how to do it effectively. We've been thinking about this a lot with Exist, because essentially we're building another silo for your data, with all the connotations around that. We're really anti-advertising, but essentially users have to take us at our word when we say "trust us!". So we're looking into better ways of handling that, like open-sourcing Exist or decentralising the data.

Data ownership and control is an area where users will have to quickly become literate — the regular Facebook "privacy settings" uproars, and even the recent Madison Ashley breach, are forcing people to come to terms with what it really means to use a service and trust it with personal details. I think players like Facebook, Google, and Apple, if they really care, should be using their position to push better solutions, although I admit that's unlikely. I think even someone in activity-tracking like Jawbone or Fitbit could gain a lot of credibility in developer circles by committing more to open-source and easier access to user data.