WhatsApp AI agent for your business: answers with your data, hands off to a person
A WhatsApp AI agent answers your customers 24/7 using your own information, asks the right questions to qualify a lead or book an appointment, and hands the conversation to a person when it should. The difference between one that works and one that embarrasses you is that it never makes up prices, stock or policies. I build them with that rule and keep them running.
Starting price for a first agent with one main use case. You get a fixed quote after describing your process.
Sound familiar?
- Your team answers the same questions on WhatsApp all day.
- Leads write at night or on weekends and nobody answers until Monday.
- You tried a chatbot and it made things up, looped or annoyed customers.
- You need exact answers from your data: prices, coverage, order status, availability.
Why it happens
- It answers from memory. A model without your data guesses. The agent has to answer from a knowledge base and from tables with exact data.
- No way out to a person. When the customer is upset or the question is outside the scope, the agent has to hand the chat to someone on your team.
- It loops. Without loop detection, two bots or one confused customer can keep the conversation going in circles.
- It was never tested with real chats. Real customers write with typos, voice notes and half sentences. The agent has to be tested with that before going live.
How I fix it
We define what the agent handles, what it never does and the tone of your brand.
I load your information into a knowledge base, plus tables for data that must be exact.
I connect WhatsApp Cloud API and an inbox like Chatwoot so your team can take over any chat.
We test with real conversations and adjust before going live.
After launch I review conversations and improve the answers.
What is included
- Intent design and the tone of your brand.
- Knowledge base plus tables for exact data.
- WhatsApp Cloud API and Chatwoot with human handoff.
- Loop and bot detection.
- Testing with real conversations before launch.
Mobile carrier in Colombia: WhatsApp customer service agent in production with a 16-intent classifier, loop and bot detection, a knowledge base and a 910-city coverage table, so it answers coverage questions by city without mistakes.
Frequently asked questions
Which AI model do you use?
Whichever works best for your case and budget: OpenAI, Gemini, Claude or open models. The design lets you switch models without rebuilding the agent.
What does it cost to run each month?
Two things: the AI usage, which depends on how many conversations you have, and Meta's WhatsApp charges for the messages that start conversations. I estimate both with your real volume before we start.
Can it book appointments or update my CRM?
Yes, if your calendar or CRM has an API. It can qualify the lead, book the slot and leave the record in your CRM.
Does it speak English and Spanish?
Yes. It answers in the customer's language, which helps if you serve Spanish-speaking customers in the US.
Other services
Something broken or worth automating?
Tell me what is happening. You get an answer the same day with what I see and a proposal.