Filipino small businesses get pitched chatbots that cannot actually speak to a Filipino customer. The voice is wrong, the script breaks the moment someone code-switches into Tagalog, and the flow assumes a US e-commerce funnel that nobody around here uses. I started the micro-agency to fix that, one shop at a time.
The pitch is plain. I build a Botpress flow that handles inquiries the way the owner’s nephew would on Messenger: friendly, fast, knows the menu, hands off to a human when someone is actually buying. Behind the flow sits a Claude or GPT model handling the open-ended turns, with the structured nodes catching the repeat questions so we are not paying token costs on “magkano po?” a thousand times a day.
So far I have built test rigs for a kakanin seller, a small print shop, and a barbershop that wanted online booking. None of them are paying clients yet. That is on purpose — I am treating the next few months as a portfolio sprint, not a sales sprint. The work I am shipping is meant to show up well on a landing page and survive a real owner stress-testing it for an afternoon.
The technical lessons have been about restraint. The temptation is to make the AI do everything. The right answer almost always involves more structured nodes, fewer LLM calls, and an explicit “transfer to human” path. Owners want predictability more than they want cleverness.
The business question I am still working through is pricing. Setup-plus-retainer is the obvious model. Whether SMEs around here will pay a recurring fee for a chatbot they cannot see — versus a one-time build they can point at — is the experiment.
If this turns into real revenue, it funds the games and the next product. If it does not, I will have learned a lot about Botpress, prompt design, and how Filipinos actually buy small things from small shops. Both outcomes are useful.