A Telegram bot instead of a call center: what really gets automated

A bot doesn't replace a sales department. It covers the most expensive stretch — the first contact: a person writes at an inconvenient time, gets an answer right away, leaves their details in a clear form. After that a manager steps in, and it's already a conversation to the point, not "hello, how can I help".

What a bot does well

• ·Takes a request around the clock and never loses it: the message goes into a database, not into an employee's personal inbox.
• ·Asks two or three clarifying questions — task type, deadline, budget range — and filters out irrelevant enquiries.
• ·Answers common questions with the same correct wording every time: terms, deadlines, documents, address.
• ·Books an appointment for a service or consultation and reminds about the visit.
• ·Sends data onward: to CRM (amoCRM, Bitrix24), to 1C, or simply by email to the person responsible.
• ·Accepts payment via YuKassa — if the sales scenario calls for it.

What a bot does badly

• ·Tough objections and haggling: "too expensive", "the competition is cheaper" — that's a human's job.
• ·Non-standard requests: the bot either leads you into a dead end or guesses wrong.
• ·Emotions and conflicts: an annoyed client needs a live answer, not buttons.
• ·Agreements that need to be "felt out": the bot records them, but doesn't negotiate.

What the setup looks like

Client → Telegram bot (grammY)
   ├─ common questions           → answer from the scenario
   ├─ request + 3 clarifications → database
   ├─ urgent/complex case        → notify the manager
   └─ confirmed data             → CRM / 1C / email

A simple scenario — taking requests and notifications — comes together in a few days. Scenarios with payments, an accounting system and roles take longer: there it's important not to make a mistake in passing data, otherwise the bot starts "multiplying" duplicate deals.

Where to start

• ·Write down the 5 most common client questions — you already have ready answers for them.
• ·Describe what should happen to a request next: who answers, where the data goes, in what form.
• ·Define what counts as a "lost request" for you: without it you can't verify that the bot actually works.

If after that the question "wouldn't it be simpler to hire an operator" still remains — count how many enquiries come in outside working hours. Usually those are the ones that decide it.