There's no wrong road in life
Zero to 1.2M views in 10 weeks, with AI as the production team
人生走彎路 · Weekly Mandarin Video Podcast
Role
Co-founder / Producer / Growth
Company
Wanlu (人生走彎路)
Timeline
Nov 2025 – Jan 2026 (10 weeks)
views across Meta on the final reel. It shipped the week I wrapped up, and it's still climbing.
"I emptied my family's group chat with one sentence, at Lunar New Year"
Context
Two hosts, zero audience, AI as the third teammate
Wanlu (人生走彎路, "Detours") is a weekly Mandarin video podcast built on one idea: there's no wrong road in life. It tells the stories young Asians trade over drinks. Dating disasters, nightmare neighbors, things you wish you could say in the family group chat. Two hosts, zero audience, zero budget, and AI as the third teammate. I ran it like a consumer product: every piece of content shipped with a hypothesis, a metric, and a next iteration.
What I built
01
An AI production pipeline, designed around what 2025 models couldn't do yet
AI generated the topic bank from our personalities, and scaffolded scripts. Since models couldn't cut video, I inverted the workflow: upload footage, let the LLM pick the Reels-worthy moments and write hooks as text, then execute the cuts by hand. AI judgment, human hands.
Script Assistant
topic bank · scripts · hooks
EP09.mp4
Uploaded
This episode needs a Reels cut — help me find the hook and script.
🎬 Cut: "He Listed His Ex as a Personality Trait" — confessional style
00–05s (Hook):on-screen text "the green flag on his profile was actually a red flag"
05–15s (Key moment):cut to "still figuring things out with my ex," said like a fun fact — hold on her reaction
15–20s (Close):jump cut to "Swipe left faster than this," then "New episode out now" title card
Title suggestion: He said it like it was normal. We said it was a red flag. 🚩
Before
After02
Retention-driven iteration
Reading the Reels Retention curve, I rewrote hooks at the exact drop-off timestamps.
Reading the Retention curve: before vs. after hook rewrites
Skip rate (%)
What happened
Instagram views, final reel
views across Meta
Spotify plays lift on that episode
lift on the next three episodes
My final reel, I emptied my family's group chat with one sentence at Lunar New Year, shipped the week I wrapped up. Engineered on a seasonal pain point, not luck. I kept tracking the data after I left. The system outlived its builder.
The viral reel: 0 to 700K in 4 days (cumulative views)
Every reel, in posting order: iteration compounds (log scale)
Why I left a growing product
I shut the channel down
Reels were my lane, and the same retention lens showed me the podcast's ceiling: episode listeners were dropping in the first 15–30 seconds, on the long-form content I didn't own. I shut the channel down. What I kept is the playbook. I now know how to take a content product from zero to a million views, and how to do it again. Happy to go into the details in an interview.
Episodes
Peak reel
1.26M
Watch time
14s → 24.6s
Spotify lift
17×