The research brief showed that most coaching happened once or twice a week, while the days in between were largely disconnected from the process. The founders’ research found that coaches already assigned activities, but content and client history were scattered across notebooks, spreadsheets and saved links. Preparing another activity meant piecing that history together. After sending a link, coaches had little information about whether clients completed the activity. The opportunity was to give coaches a lightweight way to keep that momentum going between sessions.

When I joined, founders Alex, Rod and Pete had a Figma prototype: AI would help coaches make sense of client history, prepare sessions and suggest activities. An AI companion would help clients reflect between sessions.

Alex and Rod were both founders and experienced coaches, so our stakeholder interviews also gave me a firsthand view of their coaching practice. I also reviewed notes and recordings from their interviews with other coaches in their networks. We agreed to focus first on the assignment flow: how coaches chose activities, how clients completed them and reflected, and how that feedback helped coaches plan the next session. We saw that loop as the foundation for the rest of the product. Notes, session management and scheduling could wait.

I pulled together the coaches’ goals and concerns from those interviews. At first, what I knew about clients came from their coaches. Coaches wanted to: • Help clients make lasting progress beyond the sessions • Make AI a regular part of their coaching practice • Bring more meditation and breathwork into their work with clients • Give clients one place for materials, practices and reports instead of scattered emails AI was coming up in their professional circles. They saw how AI might help analyse client data and prepare sessions, but worried about clients’ reactions, privacy and data use. Notes and scheduling apps were familiar territory; some coaches had also used Kajabi or ChatGPT.

For the assignment flow, our hypothesis was that bringing session history, assignments and reflections together could help coaches plan what came next and support clients between sessions. That left us with three questions to explore: • How might we help coaches choose what fits a client without piecing their history together each time? • How might we make reflecting with AI feel useful to clients, while keeping them in control of what they share? • How might we help coaches turn what clients share between sessions into a meaningful next step? Before testing, we agreed on three success criteria: • Coaches could prepare an assignment suited to a client’s goals without gathering context again in other tools • Clients completed the activity without moderator help, understood they were talking to AI and who could see the conversation, and chose to leave a reflection • Coaches could identify a next step, such as a session topic or another assignment, and explain what informed it

The founders’ prototype covered the coaching goals, but I had doubts about its interaction framework, so we tested that version before making changes. Client records and coaching tools took up half the screen. AI chat took up the other half. In tests, coaches spent most of their time in the CRM, turning to AI for assignments, notes or content. Half the screen was a lot to give up for occasional help. They also moved between the assignment editor and client overview to check history. AI had that context, but coaches needed to check it themselves.

Client records became the main workspace, with AI available in a side panel. Quick views kept client context within reach. Coaches could open a full view when they needed more space.

The CRM stayed the default, but some coaches preferred conversation. An optional AI workspace let them start tasks in chat and revisit conversations from across their work. The side assistant helped with whatever they had open in the CRM.

Writing prompts didn’t need to become part of a coach’s job. We organised the Prompt Library around familiar tasks, including client progress reports. AI could bring together session notes, assignments and reflections into a draft the coach could review and edit before sharing.

In client tests, separating instructions and video from the AI conversation left people unsure where to start. We also couldn’t tell whether they’d watched the video or skipped to reflection. Some exercises called for closed eyes. A running video wasn’t doing much for them there.

An assignment could run daily until the next session, so the coach needed more than a completion status. I put an AI summary first to bring together patterns across the client’s practices and reflections, with the full conversations available below. Feedback from around 15 coaches and our team pointed to shorter summaries, links to the source conversations and help planning the next step. The summary shipped; connecting findings to client goals and suggesting another assignment or session topic remained planned.

Coaches said reflections and journal entries gave them more to work with when preparing sessions. AI helped them make sense of that material. They also said they rarely needed to work outside EvolveWell, apart from taking handwritten notes during sessions. This feedback came through Alex’s meetings with coaches and the Zig Ziglar coaching association.

Coaches also passed on feedback from their clients: they felt they were progressing faster, enjoyed talking with the AI companion, and felt more engaged with their coach. That was their feedback; we hadn’t measured time saved or changes in client progress.

Over time, we shifted our focus from individual coaches to coaching organisations such as Zig Ziglar, Co-Active and the International Coaching Federation (ICF). Our contact with Co-Active introduced us to Desiree Perez. After a year in coaching, EvolveWell pivoted towards aviation as she joined as COO. She knew the industry, had contacts there and saw a bigger opportunity for EvolveWell.

Doing breathwork and meditation with the team was one thing; deciding what to share afterwards was another. Taking part gave me a view of the product I wouldn’t have had from the coach’s side alone.

Coaches knew what they wanted to get done. Asking them to invent the right AI prompt added another task. I learned to start AI interactions with familiar coaching work, then let the conversation continue from there.