Before talking to anyone, I ran a Service Safari, walking through the experience as a member would. Arrival. First-minute onboarding. Wi-Fi setup. Picking a zone. Trying to take a call. Asking for help when something broke.

This wasn't research yet, it was orientation. But it gave me a baseline of the actual service moments before any staff or member could tell me what they thought happened. Designing solutions from secondhand descriptions of a service is one of the fastest ways to design the wrong thing.

The feed version of this life is a laptop on a balcony. I had lived the other version too, the month where the novelty has worn off, the logistics are solved, and nobody knows your name yet.

So I started where I could be honest, with my own route. 2018 in Sri Lanka and Australia, remote design work before anyone called it a boom. 2019 at Dojo Bali, a small place that held a real social safety net. 2025 in Da Nang and Bali, mega coworkings where the infrastructure is perfect and the intimacy is gone.

Then I checked whether that was just me. Desk research and netnography said it was not. Hypermobility tracks with perceived loneliness, and the local reporting on foreigners in crisis in Bali is far bleaker than anything on the hashtag.

The most useful failure of the term was a piece of data I threw away.

I had a WhatsApp exchange from January, transcribed, verbatim, exactly the kind of material that makes a case study feel real. Then I read it back against my own methodology, which promised that names were removed and respondents coded P1 to P5. In a space that small, the messages plus the name of the venue identify the person completely. Coding them changes nothing.

So it came out. All of it.

The same rule removed a second thing. My research board held a wall of news cards about individual deaths in Bali. All real, all named, and easy to turn into something that reads as sensation rather than research. That context now sits in the work as figures and studies, with a content note and a helpline line at the top.

Consent is not a form you collect at the start. It is a constraint on what you are allowed to publish afterwards.

You cannot walk up to someone in a coworking space and ask how their mental health is. The question is intrusive, the setting is public, and the answer would be worthless anyway.

So I moved the question one step back. Instead of asking members how they feel, I asked the people who watch them every day. Five staff across two spaces in Bali, coded P1 to P5, from a team member on the floor to the general manager.

Their answers did not line up, and that was the finding. One manager said members tend to be open, so difficulties rarely stay hidden. Another said the opposite, that people say nothing and the team works it out for themselves. The general manager was specific: work pressure, fatigue, and loneliness from working alone for too long.

Every one of them helps. None of them has anywhere to send someone.

Strain does not build evenly. It peaks around months three and four. The novelty has faded, the logistics are solved, and deeper relationships have not formed yet.

At that point there is no door to knock on. Community support stops at small talk. Clinical therapy starts far above where the person actually is. Between the two there is nothing, which is why staff who genuinely care still end up doing nothing beyond a kind conversation.

That gap is the design problem, and a wellbeing feature does not close it. It needs a route: someone notices, and there is a named professional at the other end. The space is the navigator, never the therapist.

wo things, written down because I want to remember them.

Sensitive research is a protocol, not good intentions. Consent, the right to stop at any point, no names, respondents coded. I wrote all of that before the first interview and it changed the questions I asked.

And design for the trigger, not the symptom. Designing for the symptom produces another wellbeing tab nobody opens. Designing for the trigger, the month three drop and the missing referral route, produces a service with roles in it and someone accountable when it fails.