The makers of wearable devices are now very awake to the importance of prompting positive behavioral change. Many customers are buying these wearables or their attendant apps with the expectation that they will lose weight, get fitter, or improve their health. Health insurers want the same things. But six to 12 weeks of high use before being consigned to a drawer is typical for a new wearable. Clearly these devices are disappointing both customers and insurers – but can their potential be realized?

Recent small scale, but meaningful, outcomes from the use of wearable-derived data include taking more steps, getting a bit more sleep, or drinking more water.

The Process diagram outlines the elements and outputs of the ARDA Coaching Engine.

Wearables can deliver much more than this. Prompting significant behavior change that leads to step changes in health and wellbeing can be achieved using a three-pronged strategy:

  • Apply deep domain expertise to big data sets to develop algorithms that link metrics with outcomes.
  • Provide advice that is highly customized to individual users in particular contexts.
  • Build trust by telegraphing changes in expected outcomes, which users go on to experience for themselves.

The results can be dramatic, as shown by a five year study conducted to build a business case to New Zealand’s largest health insurer. In the study, 1,500 non-physically-active participants received automated behavioral prescription to guide them toward a goal of completing a half marathon. Of the 1,500, 89 percent reached the start-line, with a 100 percent completion rate. By contrast, nine percent of the control group reached the start-line, with varying levels of completion. A number of healthcare indicators, such as BMI, heart rate, and cholesterol levels, also showed significant improvement as compared to the control group.The results are explained with reference to the strategy above.

The Alice image highlights the real-time feedback which differentiates the ARDA Coaching Engine’s capabilities from devices/apps currently on the market.1. Apply Deep Domain Expertise to Big Data Sets to Develop Algorithms that Link Metrics with Outcomes
The commercial teams developing self-driving cars did not start from a blank slate. That is, they didn’t provide a machine-learning system with all of the possible data inputs that a car can receive, and through trial and error let it figure out how to drive. Rather, they started with algorithms that dictated certain responses to certain situations (“Turn slightly left as the road begins to bend left”) and then modified them based on real world results. Google’s driverless cars have a detailed map of the local world, and sensors report back any ways in which the world differs from those maps.

Performance Lab has been gathering data on physical exercise for 20 years in order to create the ARDA Coaching Engine. In the same way that a car can operate without a driver, a coaching system can operate without a coach. And we also did not start with a blank slate. The engine has been guided by the disciplines of coaching, sports science and physiology to determine what data was relevant, how it might be connected to health and fitness outcomes, and how to best measure those outcomes. Stride rate and stride length are relevant data inputs for a runner, for example. And the power output that the runner is able to sustain is one useful measure of her fitness. Learning precisely how stride rate and stride length are connected to increases in power output over time allows the engine to provide coaching advice in this particular area. Of course, this is just one example of the relevant relationships between input and output data.

2. Provide Advice that is Highly Customized to Individual Users in Particular Contexts
In the running example above, one could imagine looking at patterns over the whole data set of tens of thousands of athletes with millions of exercise events, and developing ‘rules of thumb’ linking stride measures with fitness improvement. But that wouldn’t tell us anything beyond what a good coaching manual already explains. The power of data gathering and machine learning using wearables comes into play if we can determine, for a given individual, how his unique body works to maximize and improve outcome variables such as power output. And beyond that, we want to know how his body works in different contexts: on hills; on the flat; the day after a hard workout; after a rest day and so on.2

This is the level of contextual understanding on which the ARDA Coaching Engine is based. That contextual and individual knowledge allows the platform to provide sophisticated advice, confident in the knowledge that compliance will lead to improved outcomes.

iPhone picture provides an example of an ARDA Coaching Engine user interface. However, a user can select the feedback mode (could be audio, on phone, or silenced until you return from a workout) and the engine is technology agnostic.3. Build Trust by Telegraphing Changes in Expected Outcomes, Which Users Go On to Experience for Themselves
Motivation is often cited as the key to creating behavior change with wearables. Our particular view on what motivates people to engage in exercise, regardless of fitness level, hinges on trust. If there’s confidence a certain course of action will lead to a measurable performance improvement, we can tell the athlete what outcomes to expect after a certain time period. After complying with the advice, the athlete then both experiences and measures what was promised. She might find, for instance, that she runs faster yet with less perceived effort. As a result, she builds up a high degree of trust in the coach or system. This leads to continued compliance, continued performance improvement, and ever-increasing trust in a virtuous circle. It is by drawing non-active people into this virtuous circle of activity and improvement that we have achieved significant measurable health outcomes.

Getting to Market
The true potential of wearables in the health and wellness market will emerge once advice is customized to a segment of one – that is, to an individual in his unique circumstances. We can then help an exercise novice with high cholesterol find the best terrain for the low intensity cycling he needs to see improvements in lipid levels. We can warn someone seeking to lose weight to pack an extra wholegrain sandwich in week two of her workout program, when her first 20 minute run is going to trigger a hunger attack that might otherwise be relieved at the corner donut vendor.

As discussed above, automating this sort of advice requires sophisticated machine learning, boosted by deep domain knowledge. However, organizations that possess these competencies are not necessarily also leaders in designing delightful applications or devices. The reverse is also true: Witness the frantic hiring efforts of large device manufacturers as they move into the health space and seek expertise as they build platforms from scratch. A more likely recipe for success is the formation of partnerships. Increasingly domain experts, such as Performance Lab, will license either their proven algorithms, or provide APIs on standard formats, for easy integration with devices that already enjoy a level of market share.

Current licensees of the ARDA Coaching Engine begin working with the API and SDK three to six months before launch of a new version of their device or app. ARDA pairs easily with a range of third party sensors, and Java or Objective-C libraries can be used to optimize performance and battery life on a preferred platform, by focusing on defined activity types.3 Workout data can be viewed in real time or post-exercise by other parties, such as insurers, trainers or medical teams, based on permissions.

Given the current level of customer churn, wearable devices and brands that combine their strengths through a partnership approach stand to benefit greatly, as will those with a stake in improved population health.

1Performance Lab, 2005
2This differs from wearable platforms that seek to provide advice on so-called ‘training zones’ based on one parameter alone, such as heart rate. Heart rate without context (hills, temperature, workout history) is close to meaningless, and not reliable as a training guide.
3At a more micro level, commentary delivered to users by the ARDA Coaching Engine is triggered by ‘events’ that the system recognizes. Partners can use default commentary or customize it to fit unique customer types, use cases, or brand terminology.