We Opened in the Worst Possible Location. All 5 Competitors Died. We Didn't.
When our founder opened the first ZWISERFIT location in 2019, she didn't pick a prime spot. She didn't pick a wealthy neighborhood. She picked Wanjiang — a working-class district in Dongguan, China, where the average gym-goer spends about ¥200/month ($28).
This was not a mistake. It was a deliberate stress test.
Here are the numbers.
The Data Points
Location: Wanjiang Street, Dongguan (not Shenzhen — this matters)
Online rating: 3.3 / 5.0 (Dianping)
Monthly membership: ¥299 ($42) — most expensive in the area
Floor space: 800 sqm (8,600 sqft) — largest in the area
Years operating: 7
Competitors who opened within 1km and closed: 5
Five gyms came. Five gyms went. Each had better funding, better locations, or lower prices. One was a chain with 200+ locations. All gone.
How a 3.3-Rated Gym Outlasted Everyone
Conventional wisdom says a 3.3 rating is a death sentence in the service industry. But here's what the rating doesn't capture:
1. The rating reflects the hardware, not the service. The 3.3 is for equipment age and facility wear — things that cost money to refresh. The actual customer retention numbers tell a different story.
2. Pricing paradox. ¥299/month was the most expensive in the area. Higher price, older equipment, lower rating — yet members stayed. Why? Because the value proposition was never "cheapest gym." It was "this gym will actually remember what you did last time."
3. The only 800 sqm space survived. Every smaller competitor closed. Scale created variety that kept members engaged across different training modalities.
4. Ownership longevity. The founder ran the front desk herself for 7 years. She knew every member who walked through the door. That continuity created a trust buffer that no app or discount could replicate.
Why This Matters for What We're Building
Most AI startups build in ideal conditions. Perfect data pipelines. Unlimited cloud credits. Early adopters who forgive bugs.
We built in Wanjiang because if the system works there, it works anywhere.
The 3.3-rated gym that outlasted 5 competitors is not an anecdote. It's proof that existing success metrics (ratings, price, location) don't predict survival when the operating system is different.
We didn't replace the human touch. We gave it tools. The system handles:
- Check-in, scheduling, payment tracking (repetitive ops)
- Equipment utilization monitoring (data-driven ops)
- Member retention pattern alerts (predictive ops)
The human handles:
- The conversation at the front desk
- The recognition of a member who hasn't come in 2 weeks
- The decision to help someone modify their form
This is the AI + human operating model — running in production for 7 years in the most difficult conditions we could find.
What's Next
We're packaging this operating system into two products:
- Saros (B2B): AI store manager for existing gyms
- Melody (B2C): AI health companion that learns you over years, not days
The foundation — KinTwin (the behavioral twin engine) — sits underneath both.
7 years of real-world data. 5 competitors proved wrong. 1 gym that refused to die.
And now we're open-sourcing the architecture.
This is the physical location where ZWISERFIT runs 9 AI agents on 2 CPU cores and 6GB of RAM: Wanjiang, Dongguan. The gym with a 3.3 rating that outlasted everyone else.
Built by Suzanne Mok — 7 years at the front desk, then 9 agents running the same desk.
More at github.com/ZWISERFIT
Top comments (0)