Key findings
- The opening assessment charged a high energy cost before giving much back.
- Repeated check-in prompts could turn absence into guilt.
- Notification wording also exposed private context outside the app.
- The motivational language did not have one consistent meaning.
- Opening assessmentmost effort, first
- Check-ins
- Coming back
Focus groups on early drop-off.
People quit at the exact moment the app asked the most of them.
At a glance
The app combined an opening assessment, daily check-ins and a structured wellbeing programme for people experiencing fatigue and low mood. The client wanted to understand why people left early. We led the research programme from the initial question and participant safeguards through analysis and testing of changes.
This was a product-research engagement with clinical input. We needed to understand the experience without asking people to disclose more personal information than the research required. The product's relationship with someone on a difficult day was as important as its behaviour during a planned research session.
The brief
The initial request was to run focus groups about early drop-off. A group discussion could help explore language and ideas, but it was a poor fit for understanding private moments when someone did not feel able to continue. Talking in front of strangers could introduce pressure to explain an experience neatly or to withhold it entirely.
The real problem was the effort involved in starting, continuing and coming back. We also challenged the retention target behind the brief. Rather than allowing a single headline activity measure to dominate, we focused the research questions on completing an early part of the programme and coming back after a lapse. Those were behaviours we could investigate in relation to the experience.
Research with a clinical lead
Before recruitment, we agreed how sessions would be run with the client's clinical lead. The protocol addressed screening, distress, the moderator's ability to pause or end a session and the availability of support. Participant payment did not depend on finishing a difficult conversation.
We limited what we collected. The research did not need a parallel record of diagnoses, medication or mood scores. We agreed recording and deletion arrangements, pseudonymised notes and the responsibilities for handling data before fieldwork began. These decisions affected the research design, not just its paperwork.
Product claims were another boundary. We flagged the implications of the language being considered and referred decisions about clinical claims to the client's regulatory adviser. Our role was to make the product and research questions clearer, not to make a medical classification decision.
Why we combined methods
Moderated individual sessions let us see how current and new users interpreted the live app. We could examine the opening assessment, the meaning of its questions and the expectations created before the first useful activity.
Diary research addressed a limitation of a scheduled session: people may have enough energy to attend research while struggling to use the product at another time. Short entries and optional voice notes gave us a view of those everyday decisions without making participation another demanding task.
A small co-design group focused on language and concepts without requiring anyone to share personal history. We then compared the qualitative work with cohort analytics. The analytics helped locate patterns; the sessions and diaries helped explain why apparently small steps became barriers.
What we found
The opening assessment charged a high energy cost before giving much back. On a good day it could seem manageable. On a difficult day, seeing the task was enough to close the app. The relevant question was not whether a motivated person could complete it in a test, but whether it was a reasonable first step in the context the product served.
Repeated check-in prompts could turn absence into guilt. Participants interpreted the sequence as the app keeping score. Increasing reminders risked intensifying the reason they were avoiding the product. A missed session needed a different response from an unfinished purchase.
Notification wording also exposed private context outside the app. A message visible on a locked phone could be read by a partner or colleague. Some users had disabled notifications because of that exposure, so an engagement problem also contained a privacy problem.
The motivational language did not have one consistent meaning. Language about breaking a cycle encouraged some people while sounding like blame to others. Meanwhile, the assessment's clinical style could make people feel labelled before they had received help. These were interpretation problems, not simply readability scores.
What we tested
We tested a shorter start with more profile information collected later, after someone had reached useful content. This treated onboarding as something the service could earn over time rather than a complete information request at the door.
We tested a lower-effort check-in for days when a longer activity was unrealistic. We also tested neutral notification wording and user control over timing, so the prompt did not disclose the subject of the programme to someone nearby.
For returning users, we explored a welcome-back flow that let them resume instead of confronting them with a broken streak. The point was to support another attempt, rather than to make the record of absence the first thing they saw.
Different changes needed different evaluation designs. We used comparative experiments where appropriate and a before-and-after comparison for people coming back to the app. The tests were tied to specific behaviours, rather than a general claim that the app had become more engaging.
The principle
A product for people with limited energy must work on their difficult days. A smaller action can be more useful than another reminder, and returning after a break can matter more than keeping an unbroken record. The research changed what the team considered a successful interaction.