Overview
Personaizit is an AI customer agent product I designed and run in production. The agents answer customers on WhatsApp, Instagram, Messenger, TikTok and web chat. They qualify each contact and hand off to a person. The data below is from production as of October 9, 2026.
The problem
The same problem shows up in every client:
- Many messages arrive, and a small team cannot answer all of them on time.
- A large share arrives after hours. That is 48% for the hotel, 48% for Escapadita, 39% for Chimeneas Esqueda and 29% for Kenworth DAF Monterrey.
- A late reply can mean a lost sale or a lost booking.
Results
La Posada en el Potrero Chico
Hotel and tourism. Instagram, Messenger and WhatsApp. Since April 2025. Full case.
- 130,011 customer messages handled.
- 49 s median AI reply time.
- 5,303 contacts reached the booking stage.
- 84% of outgoing messages sent by the AI.
Escapadita
Vacation rentals. WhatsApp, Instagram, TikTok and Messenger. Since March 2025. Full case.
- 34 s median AI reply time.
- 96% of after-hours messages answered by the AI in under 5 min.
- 1,006 price questions handled by the AI.
- 322 contacts requested a quote.
Kenworth DAF Monterrey
Sales of new and used heavy trucks. WhatsApp and web chat. Since January 2026. Full case.
- 73 s median AI reply time (last 90 days).
- 93% of after-hours messages answered by the AI in under 5 min.
- 91% of outgoing messages sent by the AI (last 90 days).
- 17% of contacts are qualified and handed to an advisor, so advisors only see qualified leads.
Chimeneas Esqueda
Fireplace sales and maintenance. Mostly WhatsApp. Since September 2025. Full case.
- 40 s median AI reply time.
- 1,868 contacts (14%) reached the quote stage.
- 1,435 price questions handled by the AI.
- 93% of after-hours messages answered by the AI in under 5 min.
What I built
The system has three parts:
- Backend. A Python service built on FastAPI. It uses the OpenAI SDK for the agents, MongoDB for storage, Socket.IO for real-time updates, and Twilio.
- Web app. A Next.js and React app with TypeScript and Tailwind CSS.
- Mobile app. An Expo and React Native app, also with TypeScript and Tailwind CSS.
The agent replies right away, asks the qualification questions and classifies each contact. Qualified contacts go to a person on the client's team.
Lessons learned
- After-hours coverage matters. A large share of messages arrives when nobody is at work.
- Qualify before handoff. Advisors then see only qualified leads and spend their time on people who can buy.
- Fast replies across many channels keep the customer in the conversation. The people on the team keep the cases that need a personal touch.
All numbers from production data as of October 9, 2026. More at augevi.com/casos.