Skip to content
All work
AIBotsIntegrations2025 · ~3 min read

A WhatsApp assistant that books the client itself

The client arranges a consultation in plain WhatsApp text, and the bot finds a free hour, files the lead in Bitrix24 against a specific manager and sends a reminder 15 minutes before the call.

# The problem

The company handles visas and travel to specific countries. Clients write on WhatsApp, and before me all of it was done by hand: a manager read the message, checked who was busy, named a time and filed the lead in Bitrix24 themselves. That is how most places work, and the arrangement has standing costs. A message that arrives in the evening or at the weekend waits until morning. Settling on a time stretches into a conversation: when suits you, that hour is taken, let us try another day. And the booking exists only until somebody forgets to move it into the CRM — at which point nobody is expecting the client at all.

# The solution

Messages arrive through the official WhatsApp Business API, delivered over a webhook by Wazzup. An OpenAI model writes the replies, so the client types in plain language: “book me for tomorrow”, “what is available”, “changed my mind, make it 11”, “delete the booking”. There is no button menu at all — the bot works out the intent itself and holds the context of the whole conversation. The system has no database of its own, and that is deliberate: the schedule, manager availability and the bookings themselves live in Bitrix24, and the bot works with them through the API. The client sees everything in the CRM they already use, and there are no two sources of truth to keep in sync. Distribution turned out to be the hard part. A slot is one hour, and the system allows up to five bookings per manager per hour. On top of that sits the roster: a manager works today, not tomorrow, is already busy at this hour. All of it is stored in Bitrix24, and the bot has to check before it names a time — otherwise it promises an hour that does not exist and a person has to untangle it. Every booking becomes a lead marked as coming from the AI, plus a task for a specific manager with the exact call time. Fifteen minutes before the consultation the client gets a reminder. A manager can switch the bot off in one particular conversation and answer personally — for when the discussion goes somewhere an automaton has no business being.

# The outcome

I have no metrics: the client collected no statistics and I had no access to their data. What can be checked is in the recording above — the bot holds the context of the whole conversation, refuses to book a time that does not exist, and handles rescheduling and cancellation correctly. There will be no public demo, and that is not caution. Every conversation means paid requests to the model, and an open link would turn into free AI for passers-by at the client's expense. I show it on a call instead. Development took a month, including the pauses when the client went quiet.

# How it works

  1. ClientWrites in plain text: book me for tomorrow, move it to 11, delete the booking.
  2. WazzupDelivers messages from the official WhatsApp Business API over a webhook.
  3. Bot · PythonWorks out what the message means, holds the context of the conversation and decides what happens next.
  4. OpenAIUnderstands free text and writes the reply. There is no button menu in the system at all.
  5. Bitrix24The only store: schedule, manager availability, leads and tasks. The system has no database of its own.
  6. ReminderA message goes out 15 minutes before the consultation.

Dashed: not my work — platforms and the client’s own systems

Next project

Ozon and Wildberries product analysis from a single article number

Need something similar?

Discuss the task