Rolling out an AI agent on WhatsApp starts at R$ 14,000 setup with a monthly fee from R$ 1,490, and goes live in about 14 days. For an operation losing sales to slow replies, the return usually shows up between the third and the fifth month. International clients are quoted in USD or EUR after the scoping call.
The difference between the chatbot everybody hates and an agent that actually resolves the conversation is not WhatsApp. It is what sits behind it.
A menu bot and a conversational agent are not the same thing
The menu bot is a decision tree. Press 1 for quotes, 2 for support. It works as long as the customer types exactly what the tree expects. Since nobody types like that, and people send things like “hi, do you still have anything Thursday morning?”, the conversation stalls and the customer asks for a human by the third message.
A conversational agent reads the intent, checks the actual calendar, answers with a real slot and books it. It is not guessing: it classifies what the person wants, looks the information up in your systems and replies with something true.
The practical distinction is simple. A menu bot informs. An agent executes: books, reschedules, logs an issue, opens a ticket, confirms attendance.
Before talking about cost, measure the leak
An operation handling 300 enquiries a month at an average ticket of R$ 250 moves R$ 75,000. If 25% of those contacts are lost to slow replies, a figure that shows up often in businesses that only answer during office hours, that is roughly R$ 18,000 walking out every month.
You do not need to recover all of it. Winning back five percentage points is already R$ 3,750 a month. In higher-ticket operations, and that is where the arithmetic gets genuinely interesting, the same percentage is worth a great deal more.
On the other side, one person dedicated to answering messages costs between R$ 4,000 and R$ 6,000 a month with payroll charges, works office hours and takes holidays. The full setup is worth less than three months of that salary. The monthly fee is about 30% of it, and it never takes a holiday.
What the investment covers
| Item | Amount |
|---|---|
| Setup, one-off, half upfront and half on delivery | from R$ 14,000 |
| Monthly: hosting, API, AI usage, adjustments and support | from R$ 1,490 |
| Time to go live | around 14 days |
Setup covers WhatsApp Business configuration, the custom workflow, training the AI on your company’s tone, integration with calendar, CRM or ERP, the web dashboard, automatic appointment confirmations and training for your team.
It moves into the R$ 25,000 to R$ 60,000 range when there are multiple locations, unusual integrations, larger teams to train or white-label requirements. Operations below roughly 150 enquiries a month usually do not add up, and I say so on the first call.
What tends to change in the operation
- Answers in seconds, at any hour. Most out-of-hours messages are about availability and price, which the agent handles on its own.
- Fewer no-shows. Automatic confirmation the day before, with rescheduling in a single message. People who cannot come say so, and the slot gets reused.
- A free front desk. Whoever was answering “how much is a consultation?” thirty times a day goes back to looking after the person standing in front of them.
- An organised history. Every conversation logged, searchable, with a report of what gets asked most.
What does not change, and it matters to say so
AI does not replace the team. It resolves the repetitive majority and hands the rest to a person with the context already organised. Delicate cases, negotiation and exceptions stay human. The difference is that the human reaches them rested.
And the agent does not invent information. It answers from the knowledge base and from your systems. When it does not know, it says so and calls someone. That is configuration, not luck.
Two systems in production
Residential building. AI service desk sorting messages into six categories: information, complaint, maintenance, booking, request and human handover. Dashboard with SLA triage and the option for staff to take over at any point. Running for more than two years.
Construction company with a vehicle fleet. Six modular WhatsApp workflows, including refuelling control with a photo of the pump and the odometer, a hiring process with over fifteen steps, and offboarding with a document checklist. More than 200 messages a day, integrated with the company’s SharePoint, live for over eight months.
How to tell whether it fits your company
It fits if WhatsApp is your main contact channel, if the same five questions repeat every day, if there is a calendar to fill, and if someone on the team has answered a customer at night or over the weekend.
It does not fit if volume is low, if every enquiry is genuinely unique, or if the process is still undefined. In that last case, automating the confusion only makes it faster.
Need this solved at your company?
A twenty minute call is usually enough to tell whether it makes sense, and you talk straight to the engineer who builds it.