One month on the UP: Learnings from the Jawbone fitness band
As I write this in San Francisco International Airport, I realize it would be a good time to charge my Jawbone UP. No steps will be missed as I will be typing idly for an hour more at the airport and then some after boarding the plane to Latvia. Daily considerations about how this fitness band fits in my life and when exactly I should be interacting with it have permeated my thoughts to the extent that my brain now has a dedicated Jawbone push notification system.
The Jawbone UP
For those of you not familiar with the device, it is a wristband that logs steps, exercise and sleep solely through an accelerometer. It has only one button that is used to distinguish between three states: walking, sleeping (or napping) and exercising. The device plugs into your smartphone and is augmented with an app that provides visualizations of activity as well as support for additional manual data entry and import for further items, such as the food you consumed and detailed exercise types.
The Jawbone is by far the classiest fitness wristband out there. The signature pattern combined works really well with the almost metallic looking cap and button. I have heard negative comments about it, but I like its clean and unique geometry as well as the consistency across Jawbone products. There are only minor issues with the physical design: first, the shiny plastic cap is easily scratched. Second, the device is meant to be worn cap down, which I feel overexposes its thickness. I like to hide engineering bottlenecks, so I solve that by wearing the band the other way around (see below). I do think the UP would look even better if the design was influenced by this realization.
Wearing the UP
The app has a great, working design. Here is a view of my last night out:
Jawbone UP app, the Lifeline
The core feature of this app — the sensor — can be hit and miss. Compared to competing products in the market, the UP falls short. A further problem that impacts data quality is having to manually change device state. I’ve had a few quick ping pong matches which will never be logged. Similarly, I occasionally forget going into sleep mode, especially when going sleep late. The bracelet should be able to infer if the owner is sleeping or awake based on movement alone, I look forward to this feature as the results of forgetting to switch state have a bigger impact than any errors in noise removal.
Looking back at nights where I did record my sleep and comparing with friends prompted a reevaluation of my sleep schedule. When I was rowing at University, I started the habit of sleeping 8 hours every day. The reasoning was that professional athletes almost never sleep less than that and some even sleep up to 12 hours a day. As a rower who also had regular schoolwork, 8 hours felt like a necessary compromise. Now that I almost never have two workouts a day and have even more work to do, a cut was in order. A little research revealed that many well-known people in tech don’t sleep 8 hours. In fact, most sleep 7 hours or less. After reading this, I have been sleeping 7 hours ever since. The extra hour is very useful for getting things done at around midnight and also gives me more flexibility for going out.
While steps, sleep and exercise are the only three features that use the accelerometer input, the app includes a few more categories of data. Food is one of them and it goes hand-in-hand with the calorie counter. Entering my meals manually for a couple of weeks proved to be a challenge. The in-built food dictionary and flexible portion specification tools go a long way, but some lunches have been too diverse. Whenever I have a colorful fruit salad, it is infeasible to recognize contributions of individual types of fruit and this will inevitably introduce some error. Having said that, I did identify three big issues with my diet that were showing up day after day: I was consuming too little protein, too little fiber and too much cholesterol.
The corresponding corrective measures were obvious and I implemented straight away: I bought protein powder, started eating more vegetables, fruit, bread and oatmeal. I also stopped eating more than one egg per day and limited my sea-food intake. Real, observable change.
Apart from personalized insights, it was interesting to think about how the inferences and computations are performed. Problems that the UP faces include not counting steps when the band is in a moving vehicle, inferring sleep phases and periods of being awake, estimating calorie burn based on exercise type (e.g., cycling burns more calories than tennis for the same amount of band motion). These are tough problems in data science and might even require more data before we can engineer reliable solutions.
Returning to actionable insights, I think the device is best used periodically. Friends have spoken from experience that I would drop the device after a couple of months. While I haven’t made past that point yet, I can hardly see that happening, because the minimum commitment level is so negligible. I did stop logging food and I stopped syncing my band every 4 hours like I did in the beginning. I embraced the fact that a full-time investment is impractical and went on to adopt a more utilitarian, incremental and time aware approach to using the Jawbone UP. Now I look at the data only periodically and start logging food only after significant changes to my diet. This approach demands very little attention, captures much of the relevant data and I think provides 80% of the value.
Wearable devices will soon be in our lives, but not in the intrusive way phones are now. The best products ought to be responsive and mostly quiet, respecting your time and compiling only the most relevant insights.
