A customer writes in Telegram at ten in the evening and wants to know the price and the timeline. If the reply comes in the morning, some of those people will have already bought from someone else. An AI assistant closes that gap: it replies right away, collects the details and hands the manager a prepared lead.
How it works, step by step
- Incoming message. The customer writes in Telegram, on your website or in another channel.
- Understanding the request. The assistant works out what the person wants: a price, placing an order, making a complaint.
- Answering from the knowledge base. The reply is built from your materials: price list, delivery terms, frequent questions. Not from the model's imagination.
- Collecting details. The assistant asks for whatever is missing: name, contact, task, timeline.
- Handing off to a person. If the question is complex or the customer asks for a manager, the conversation moves to a person together with its history.
- Writing to the CRM. A lead is created with tags and a status, so nothing gets lost.
What to prepare in advance
- A knowledge base. Current prices, terms, and answers to the 20-30 most frequent questions. The quality of the assistant depends almost entirely on the quality of these materials.
- Handoff rules. Which topics the assistant stays away from: discounts, complaints, non-standard orders.
- Tone of voice. How you talk to customers: formal or casual, short or detailed.
- Where leads go. A CRM, a spreadsheet or a managers' chat.
Common mistakes
- The assistant confidently makes things up. The fix is a restriction: answer only from the knowledge base, and when there is no answer, say so and bring in a person.
- No way out to a person. The customer gets stuck in the conversation. There must always be a clear handoff.
- No conversation memory. The assistant asks again for things the customer has already said.
- Launching without tests. Before launch, run 50-100 real questions from your past conversations through it.
How to tell it is working
| Metric | What it shows |
|---|---|
| First response time | How much faster customers get a reaction |
| Share of conversations handled without a person | How much routine has actually been taken off the team |
| Share of leads reaching a manager with complete data | Quality of collection and qualification |
| Lead-to-sale conversion | Whether things got worse after launch |
Where Telegram fits in
For many businesses, Telegram is already a working channel: customers and the team are there every day. That makes it convenient to build the assistant right into it, with no new apps. We have built both client and internal solutions on it, for example a platform of 142 bots, where each team works in its own isolated bot and the platform has held about 99.98% uptime over 90 days. And in Metra AI, the assistant handles leads in a business's direct messages and decides on its own when to bring in a person.
FAQ
How is an assistant different from a regular chatbot?
A regular bot works with buttons and pre-written branches. An assistant understands free text and answers from your knowledge base. There is a detailed comparison in the article "Chatbot or AI assistant".
What if a customer asks something that is not in the knowledge base?
A well-configured assistant does not invent an answer. It says it will check, and passes the conversation to a manager together with the context.
Can the assistant be connected to a CRM?
Yes. Leads and conversation history are written to your CRM or spreadsheet, so your team keeps working in the tools it knows.