Building a Breathwork App Taught Us Something Unexpected About Human Behavior

When we started building OceanMind, we believed the biggest problem in wellness apps was the lack of structure. Most products in this space give users a large library of content and say something like: "Pick a meditation that feels right." But transformation rarely happens that way.

So we built something different. Instead of a content library, we designed a structured system of practice. It seemed obvious at the time. It also led us through a year of pivots, experiments, and hard lessons about engagement and human behavior.

The First Version: A Structured System

The original idea behind OceanMind was simple. Each session followed a clear progression:

• Movement — light physical activation
• Breathwork — controlled breathing patterns
• Meditation — integration and stillness

The logic came from observing how the nervous system works. Movement prepares the body. Breath changes the nervous system state. Meditation becomes easier afterward. When people practiced in this sequence, the effect was noticeably stronger.

But there was one problem. A full session took 30-45 minutes. And most people simply didn't do it.

The First Product Reality

From a practice perspective, the system worked extremely well. From a product perspective, it failed. Users didn't complete the sessions.

So we started simplifying.

Making the First Days Easier

Our first experiments focused on lowering the entry barrier. We shortened the early practices. We simplified instructions. We removed complexity from the first sessions.

This came with a trade-off. The first practices became less powerful, but they were much easier to start. And engagement improved. More people began completing the early sessions.

"Accessibility often matters more than theoretical effectiveness."

The Two-Level Course Structure

Eventually we redesigned the program into two stages:

Level 1 — Habit Building

Shorter practices designed to help users start and build consistency.

Level 2 — Deeper Practice

Unlocked after completing the first stage. These sessions returned to the full structure: movement, breathwork, and meditation.

When users commit to finishing both levels, the results can be remarkable. But we encountered another challenge. Communicating that value is harder than expected. People often want the benefits of deep practice without the commitment required to reach them.

Engagement Experiments (And What Didn't Work)

Like most early-stage products, we ran dozens of experiments trying to improve engagement and retention. Some ideas seemed promising but failed in practice.

Daily Streaks

We tested streak mechanics to encourage daily practice. Instead of increasing engagement, they sometimes had the opposite effect. Missing a day made users feel like they had failed. Many simply stopped practicing entirely. For a product meant to reduce stress, this mechanic introduced unnecessary pressure.

Gamification and Achievement Systems

We experimented with achievements, badges, and milestone rewards. The results were underwhelming. External rewards did not meaningfully increase long-term retention. In a category focused on internal states, extrinsic motivation appears to have limited impact.

Push Notifications

Push notifications were another major area of experimentation. Some types of reminders worked well. Others led users to disable notifications entirely. The tone and timing of notifications turned out to be more important than the message itself.

The Insight That Led to Our Pivot

As we analyzed user behavior, one insight kept appearing:

Giving everyone the same session suddenly felt inefficient. So we started asking a new question: What if every practice could adapt to the user instead?

Moving Toward AI-Generated Practices

This question led to our biggest product shift. Instead of delivering a fixed session, we began building a system that generates personalized practices in real time.

Before starting a session, users complete a short check-in describing their current state. The system also analyzes previous sessions, including:

• Which practices the user completed
• Which practices produced positive changes
• How the user's state changed after each session

Using this information, the system generates a practice designed specifically for that moment. Sometimes the session is five minutes. Sometimes longer. The goal is simple: Create the smallest intervention capable of shifting the user's state.

Measuring What Actually Works

One of the most important components of this system is feedback. Before and after each practice, users record their current state. This allows us to track which techniques consistently improve outcomes.

Over time the system learns:

• Which breathing patterns work best for a user
• Which techniques reduce stress most effectively
• Which practices increase focus or energy

Practices that consistently work appear more frequently. Practices that don't gradually disappear.

Research Behind the System

To build this system we analyzed about 200 breathing and regulation practices. These included both modern techniques and traditional systems. We looked closely at patterns such as:

• Inhale/exhale ratios
• Breath holds
• Activation vs calming cycles
• Transitions between physiological states

Different breathing patterns interact with the nervous system in very specific ways. Some increase alertness and cognitive clarity. Others activate parasympathetic relaxation. Understanding these patterns became the foundation for generating practices dynamically.

Early Internal Results

We've been running internal tests of the AI-generated practices. The early signals are promising. Users report noticeable state shifts even during shorter sessions. Instead of encouraging longer practices, the system focuses on precision: The right practice, at the right moment, for the right internal state.

What This Journey Taught Us

Building OceanMind forced us to rethink several assumptions. The most effective practice is not always the most adopted. Behavior change products require balancing three forces:

• Depth — meaningful transformation
• Simplicity — easy to start and continue
• Engagement — compelling enough to return to

Too complex, and users never begin. Too simple, and the results disappear. Personalization may be the only way to reconcile those two realities.

We're still learning. And that might be the most interesting part of building this product.