Founder · Singapore
Xavier Fok
Co-founder of Ally Media. Built Stickeebra to US$2.66M in sales. Now running a one-person business with AI.
I've been starting businesses since I was 18, in Singapore and Bangkok. These days it's a small VA team and AI for the boring parts, and I write down what it costs and what breaks.
- US$2.66M
- Stickeebra sales across 56,209 orders
- S$250k
- in sales from about S$8.9k of TikTok ads
- 240k
- TikTok fans for a client aged 74
- 18
- my age when I started my first business
Track record
Each one links to the proof.
Stickeebra
It started with affiliate ads in Bangkok: about US$500k in ads for about US$800k in revenue. Then we built our own bra brand and ran all of it, from product R&D to warehousing. Ads reached 50-60% of revenue, and when copycats arrived the profit dried up.
See the store numbers
Ally Media's TikTok ads
It worked because the service is high value and people had already been watching us post for 2 years. They weren't seeing us for the first time in the ad.
See the ad spend
Mr Yeo, 74
A 74-year-old shoe seller on TikTok. We handled the filming, scripts and editing. Zaobao wrote about him when his videos passed a million views.
See the Zaobao articleIn the press
Things I've built
- Ally Media
- Digital marketing firm I co-founded. Filming, scripts, editing and posting for clients.
- Stickeebra
- Our own bra brand, with product R&D, quality control, customer service, fulfilment and warehousing.
- My AI setup
- Local models, agents, automations and a small VA team. I publish what it costs and what breaks.
Writing
- I let a local AI sort and tag forty thousand photosA decade of unsorted photos, described and tagged by a vision model running on an old GPU, with nothing uploaded anywhere. How the pipeline works, what it misclassified, and what the whole thing costs to run. · 7 min read
- How I choose a local AI model: a method that takes an hourLeaderboards rank models on standardized tests, which says little about your specific task. The selection method I use for production work: define the job, narrow by hardware, then run a shortlist against a tiny test set built from your own real inputs. · 7 min read
- Why local models slow down and the levers that fix itLocal model slowdowns trace back to a handful of predictable causes: context length, the KV cache, models spilling out of VRAM, and uncapped output. Here are the levers that fix them, in the order I actually check. · 8 min read
Get in touch
Email me at [email protected], or find me on YouTube. The longer, messier version of all this is in my story.