Digital Health & Wearables
Improving Long-Term Engagement & Getting Better Outcomes
These days, the digital health and wearables industry is embracing subscription models (paying a recurring fee at regular intervals). Companies like Apple, Oura, and Whoop are part of this trend. What this means for up-and-coming start-ups in this space is that keeping users engaged over the long-term is critical to building a successful business.
Long-Term Revenue Requires Long-Term Value Creation
As the business model shifts, so too must the value proposition for users. Can these products truly engage users over the long-haul? Increasingly, wearables offer to help users improve their mental and physical health. This begs the question: do they actually deliver better health outcomes, or merely offer vanity metrics?
Real Value is Achieved with Actionable Metrics
In a wearables context, vanity metrics help users feel that they are getting something intriguing (insights), but they do not inform the next steps or guide a user in the right direction. Products that don’t offer actionable metrics are a “want” not a “need” — a curiosity or status symbol at best.
For example, it is certainly interesting to know you’ve gotten 45 minutes of deep sleep, but how do you get more? Heart rate variability (HRV) is another example. If you track HRV each day, but without any context around the drivers in your life that impact HRV, then there’s not enough information to utilize the metrics for real guidance. (Are you emotionally stressed, not yet recovered from a hard cardio workout, dehydrated or hungover, or did you fidget too much, creating a false readout?) In contrast, digital HRV biofeedback does provide real-time guidance to let you know if you are training your nervous system in a beneficial way. Like a compass, an actionable metric will help you navigate through a treatment plan to reach a health outcome.
I have worked in hospital settings as a Health Psychologist, and I have analyzed engagement data at many well-regarded start-ups as a senior Data Scientist. Eventually, I moved up the ladder to a Director role, where I helped build and present sales decks that helped close deals with healthcare insurance providers, define new areas of product-market fit, and reach valuable customer segments without pricey marketing investment. This was all based on novel, built-and-validated, wearable metrics + Precision Care AI, and their promise to drive health outcomes in high-need populations. It’s clear to me that the future of health involves wearables, but what exactly that future looks like… is still under construction.
User Journeys Enhance Engagement & Outcomes
From what I have seen, behavioral interventions that really move the needle on a health outcome need to create a user journey (or a personalized patient treatment plan). Accordingly, in digital health products with the vision of achieving long-term engagement and actual health outcomes, the design focus is shifting from features (think sleep-tracking) to skills-training modules (think a week-long curriculum of psychoeducational videos, exercises, mini-meditations and wearable-based biofeedback). Another advantage to building an evolving curriculum is that it not only helps create real behavior change, but it offers novel content for the user. Generally speaking, novel content drives better engagement.
In my first data science job (at a wearables company), I led many A/B tests utilizing behavioral principles. At one point, we contrasted the impact of a 3-day user journey versus daily notifications on user engagement. The user journey promised that if users logged calories every day for 3 days (with a minimum calorie count met), we would give them a personalized summary with nutritional and weight-loss tips at the end. Users in the “control condition” received a notification each day with a catchy message. We learned that users got used to receiving notifications and stopped responding over time. In contrast, the user journey with a promise of personalized information significantly increased engagement with food-logging over time, (which I can tell you is a high-burden task that’s hard to get people to do).
Modular Curricula: Architectures that Enable Behavior Change
Many digital behavioral programs based on the Diabetes Prevention Program are structured in this way — they offer an evolving curriculum, with a different topic each week. Some mental health programs like Meru Health and Lyra also have a modular curriculum. This is important because higher-level skills often build upon more foundational skills. For example, mindfulness is often one of the first skills we teach, because it is hard to change a pattern you are not aware of.
Behavior Change Milestones and Metrics
In the figure above, I have suggested a series of behavior change milestones a user would move through in order to reach behavior change. It is not meant to be applied rigidly, but as a conceptual tool to aid in designing the user journey. Moreover, each stage in this journey will have its own unique “success metrics.”
For example, Noom includes quick quizzes to assess knowledge and skill-acquisition, and an exercise to determine your “ultimate why” to help users clarify their motivation. When users begin to disengage, the app directs the user back to the motivational exercises. Compare this design to a feature set that always provides the same repetitive metrics and visual interface. Which do you think will achieve better long-term engagement and health-outcomes?
Premium Offerings: Therapist or Behavior Coach Support
Moreover, such programs often offer a premium version, in which users pay for remote support from a therapist or behavioral coach, in conjunction with an app + wearable. From a business perspective, this can significantly increase the market size and revenue potential. Human touch also helps with stickiness (engagement).
However, therapists or coaches then become another layer of “users” in your ecosystem. They need an internal provider dashboard, which is essentially a second software product that adds significant technical complexity. As human providers are more expensive and less scalable than software, companies need algorithms to optimize provider time and focus, and training materials to ensure that the algorithms are used effectively.
If wearables are included, users will expect their providers to be experts in the device metrics and trouble-shooting — even though most therapists and coaches are not trained to understand why someone’s bluetooth connection failed or whether a given HRV score is low.
Beware Perverse Effects with Human-in-the-loop AI/ML Systems
As we combine providers and AI/ML algorithms to make health care scalable, we run into unexpected problems or “perverse effects.” For example, a provider dashboard might ask therapists to reach out to users, which a churn algorithm identifies as likely to disengage. However, by the time users have begun to disengage, they may have already made up their minds that they aren’t interested anymore. So, the provider might spend a lot of time crafting personalized text messages to re-engage users, which tend to fail. In this example, the AI/ML churn algorithm could have the perverse effect of driving costs up instead of down. On the other hand, users often react very negatively to messages they feel are “canned” or machine-generated, especially if they paid for a one-on-one provider.
This example illustrates some of the complex problems organizations face, which lie at the intersection of psychology, business strategy, and data science.
Why You Will Save Time and Money by Hiring an Interdisciplinary Advisor
Behavior change principles can seem easy compared to building an AI model, but as someone who has studied both deeply, let me say that this ease is deceptive. Ideas can be easy to understand but very hard to implement! Simply providing the knowledge (i.e., an instructional video) is not enough; it must be combined with skills training and barrier reduction. One of the biggest mistakes I see over and over again is that — because behavior change principles seem simple — product teams don’t believe that they need a specialist embedded with the team.
If you are building a digital health start-up, you may be asking yourself, how do I know what is the right thing to build? You can hold a long series of meetings with many experts in different areas — clinical, product, user experience, data science, sales, and business strategy — to identify what modules and features make the most sense. That will take a very long time. The more people who have to reach consensus, the longer it will take. This is why having an interdisciplinary advisor or consultant can end up saving you a lot of time and money.
Maybe the product team has a lot of ideas they are excited to try out. But you also need to show results to raise funding in 9–12 months. If you can architect the app with the ability to learn what works from the data, you can empower your team while guiding the process toward a desired outcome. What I offer as a consultant is someone with deep experience in many areas of this puzzle, who can help connect the dots for your team.
A Sneak Preview: An AI/ML Approach to Understanding Long-Term Engagement
Say your company has a modular app-based program to improve digital health. Now you want to understand whether the data can help you show evidence that the program works, and learn how to take the intervention to the next level.
In an upcoming blog, I will also share the published research findings from a new machine learning analytics tool I built to learn about long-term user engagement and outcomes. It is designed to be used with a digital health platform that utilizes a “modular” approach to behavior change, like we discussed in this article. We will walk through a case study example of how this tool helped provide insights that inform the strategy to improve long-term engagement and build a Precision Care treatment program.
Here are the kinds of questions this tool can help you answer:
Do users who engage more consistently with the program get better outcomes?
Which modules contribute most to user-churn?
What kinds of users churn at different points in the user journey?
Are there certain subgroups of users who might benefit from putting these modules in a different order?
Can we identify the characteristics of super-users, in order to better target them in our marketing campaigns?
In conclusion, subscription models appear to be here to stay, which means long-term engagement matters more than ever. This has broad implications for modular product design, behavioral design strategy, using data to drive success, personalizing the user-journey with AI/ML, and integrating remote providers.
Originally posted on Medium.
