You're Not Building a Wellness App. You're Building a Coaching System.

The first thing I did at Thrive Tribe, back in 2017, was ask if I could sit with the coaches.

Not to interview them. Just to watch. The brief was to take their public health programs digital, smoking cessation, weight management, lifestyle change across the UK, and I wanted to understand what I was actually taking digital before I made any product decisions.

What I expected: a structured protocol, delivered consistently, session to session.

What I found was harder to replicate.

Thrive Tribe ran evidence-based behavioral health programs built on cognitive behavioral therapy. Trigger identification, cognitive reframing, behavioral replacement, reinforcement. This wasn't wellness content. This was a clinical framework, developed and refined over years, with measurable outcomes. The product brief was, on the face of it, simple: take this and put it in an app.

What the Market Had Already Decided

The product response to "take a health program digital" is fairly well-established. Map the content into the app. Add a logging layer for behavior tracking. Build in reminders and progress indicators. Add a streak mechanic, a rewards system, some form of accountability.

This was what the market rewarded. Engagement metrics were measurable. Session length, daily active users, streak completion: these numbers told a clear story to investors, and they were not wrong as signals. A product people opened every day was better than one they abandoned after a week.

The problem was that opening an app every day and changing your behavior were different things. The model optimized for the first because it was legible. The second was slower, noisier, and harder to count.

I built the early wireframes around this model. It made sense, before I spent enough time with the coaches.

The Part That Doesn't Transfer

The coaches at Thrive Tribe were not working from scripts. They had the methodology, but the way they applied it looked almost nothing like a fixed sequence.

One session, a client was dealing with a family conflict. The trigger mapping that had worked the previous week wasn't relevant. The coach recalibrated: different exercise, different focus, different tone. Another client had stopped responding to the reframing work that had helped her in week two. The coach noticed, adjusted, tried something else.

I started asking them directly: how did they decide what to do next? The answers were consistent. You watched for what was not landing. You learned the pattern of this particular person, in this particular week, in this particular context. The program was a scaffold. What you built on top of it was specific to the human in front of you.

The coaches with the best outcomes were not the most protocol-faithful. They were the most adaptive.

That observation changed the product brief entirely.

If the value of the program was in the adaptive intelligence of the coach, then building an app that delivered the content without that intelligence was not taking the program digital. It was taking the least important part digital and leaving everything that actually worked behind.

Designing for the Already-Motivated

Most mental health apps were tracking tools dressed up as something more.

They logged moods, counted meditation minutes, recorded sleep patterns, and provided exercises from a content library. The implicit promise was that if you made behavior visible and content accessible, change would follow. For some users it did: the ones who were already motivated, already self-directing, already capable of translating a reflection prompt into an actual behavioral shift.

The users who needed the most support were precisely the ones least well served by this model.

An umbrella review in the Annals of Behavioral Medicine, synthesizing evidence from more than 865,000 participants across digital health interventions, found that the components most associated with positive outcomes were human coaching and personalization. Not content quality, not feature richness, not gamification. An exploratory retrospective study in Frontiers in Digital Health found that personalized human support increased adherence more than any automated engagement mechanism, including notifications, streaks, and progress feedback.

The pattern the research named was the same thing I watched in those sessions. The app equivalent of a skilled CBT coach was not a better content library. It was a system that learned what each person responded to and adjusted accordingly.

Most products were not built this way because building this way was harder to scope, harder to measure, and harder to explain in a product demo. So the industry kept optimizing for the metrics it could count and calling the result a coaching platform. This is a pattern I've seen across regulated platforms: teams build around the problems they can see, while the most expensive friction stays invisible until someone systematically maps it.

Three Products, Not One

Before I wrote the first feature on the backlog, I had to decide which product we were actually building.

A tracking system logged what users did. A content system gave them material to engage with. A coaching system guided them through a structured, adaptive journey toward a specific outcome. These were different architectures, not different feature sets. You could not turn a tracking system into a coaching system by adding more content. You could not turn a content system into a coaching system by adding an analytics dashboard.

When we built Lify, the machine learning layer was not a feature on the roadmap. It was the architecture.

The goal was to learn each user's triggers, identify which interventions were and were not working for them specifically, and adjust the program in response. Without that layer, we had a habit tracker with a CBT library, which was what most mental health apps already were.

The question was not whether digital products could help people change behavior. The evidence said they could. The question was whether we were building the thing that actually did it.

Further Reading

Direito, A., et al. (2023). Effective behavior change techniques in digital health interventions for the prevention or management of noncommunicable diseases: An umbrella review. Annals of Behavioral Medicine, 57(10), 817–836. https://academic.oup.com/abm/article/57/10/817/7251346

Moberg, C., et al. (2022). The impact of personalized human support on engagement with behavioral intervention technologies for employee mental health: An exploratory retrospective study. Frontiers in Digital Health, 4, 846375. https://www.frontiersin.org/articles/10.3389/fdgth.2022.846375/full