The gym with the lowest rating (3.3/5.0) and the highest price (¥299/month) is the one that survived 7 years.
Meanwhile, every competitor with a 4.7+ rating and lower prices is now closed.
Here is the real story behind the numbers.
The market: Dongguan Wanjiang, one商圈 in Southern China.
Five competitors within the same radius. All rated higher. All priced lower.
One by one, they closed. The 4.7-rated gym? Gone. The one charging ¥199/month? Gone.
The 3.3-rated, ¥299-priced gym is still running — 7 years and counting.
Conventional wisdom says this is impossible.
Low rating + high price + largest floor space (800㎡) = textbook recipe for failure.
But conventional wisdom measures the wrong thing. It measures app store ratings. Not stored trust.
105+ active members paying ¥299/month for 7 years is not a rating problem. It is a retention story.
Why the 3.3 rating is actually the honest signal.
Real members rate honestly — "the equipment is old," "the AC doesn't work in summer," "it is crowded at peak hours."
Critical? Yes. Fatal? No.
Because the one thing members actually care about — "does anyone here know me?" — has a 5.0 rating. AI knows every member by name, face, and last session.
The cost structure that outlasts the competition.
Standard competitor: 3-5 trainers × ¥8K/month = ¥24K-40K/month in labor.
ZWISERFIT: 9 autonomous agents on 2 CPU cores. Zero marginal labor cost.
When your competitor pays ¥40K/month just to greet people at the door, you can afford to run an 800㎡ space with a 3.3 rating.
"How do you grow with a 3.3 rating?"
Same way you grow without VC money: by not needing growth to survive.
¥299 × 105 members × 12 months = ¥376,740/year in membership fees alone. Proven peak: ¥400,896 revenue with 237 members.
Not Unicorn numbers. Real numbers. From one gym in Wanjiang, Dongguan — the market everyone told us was a dead zone.
The lesson for physical business AI:
The market variable that matters most is NOT rating, price, or location.
It is trust. Measured in years. Stored in behavioral data. Verified by edge AI.
Seven years and one rating point lower than everyone who failed.
🔗 GitHub: https://github.com/ZWISERFIT
Top comments (0)