We already have a five-minute quickstart for connecting a Telegram bot. This is the other guide — the one for when you want Telegram to be your actual support desk, not a weekend experiment. By the end you'll have a bot that answers on its own, a team group where every conversation shows up in real time, and a clean way for a human to take over mid-chat without the customer noticing a seam.
Budget about an hour. Most of it goes into your knowledge base, not the plumbing.
Why a bot and not a "support" username
You could publish a personal Telegram account and let customers DM it. Plenty of small teams do, and it works right up until the person holding that phone goes on vacation. A bot gets you four things a personal account never will:
- A shared inbox. Every conversation lands in one dashboard the whole team can see.
- History that survives turnover. The chat log belongs to the company, not to somebody's SIM card.
- An AI first line. Sixty-something percent of support questions are the same twelve questions wearing different hats. The bot handles those.
- Takeover. When a conversation gets hard, a human steps into the same thread. The customer never gets bounced to a new contact.
Step 1: Create the bot with BotFather
In Telegram, find @BotFather (the one with the verified checkmark) and send /newbot. It asks two questions: a display name (anything you like, "Acme Support" is fine) and a username, which must end in bot and be globally unique. Have two or three candidates ready, because the obvious ones are taken.
BotFather replies with a token that looks like 1234567890:ABC-DEF.... That token is the bot. Anyone holding it can read and send messages as your support desk, so it goes in a password manager, not in the team group chat.
While you're in BotFather, two optional commands worth the extra minute:
/setdescription— the blurb people see before tapping Start. One sentence about what the bot helps with./setuserpic— your logo. A bot with the default gray avatar reads as abandoned.
Step 2: Connect it to Lane.Chat
In the dashboard, go to Channels → Telegram → Add new bot, paste the token, and click Verify. Lane.Chat checks it against the Telegram API, pulls in the bot's name and avatar, and sets the webhook for you. No curl, no server config, nothing to deploy.
Now message your bot from your own Telegram account. The conversation should appear under Conversations within a couple of seconds. If it doesn't, the token was pasted with a stray space on the end — that is the cause roughly nine times out of ten.
Step 3: Give it something to say
A bot that replies "I'm not sure about that" to everything will burn goodwill faster than no bot at all. Before you announce anything, load the knowledge base:
- Shipping times and costs, broken out by region
- Return and refund policy — the current one, not the promo version from two years ago
- Product FAQ: sizing, compatibility, care instructions
- How to reach a human, and what hours one is around
Then interrogate your own bot with twenty test questions. Phrase them the way a mildly annoyed customer would ("where's my stuff"), not the way your documents are titled ("Shipping Policy Overview"). Every miss is a gap in the docs, and it's much cheaper to find them now.
Step 4: Mirror conversations into a team group
This is the part most teams don't know they can have. Create a private Telegram group for your support staff and add your bot to it. Then, in Channels → Telegram → Group Mirror, link that group.
From then on, every customer conversation is mirrored into the group as its own thread. Your team watches questions and AI answers flow past in real time — inside the app they already keep open all day. And when someone replies inside a group thread, that reply goes out to the customer from the support desk. No new tab. No separate agent app to force on anyone during week one.
The rhythm that works: let the AI's answers scroll by, and only open a thread when something looks off. Reviewing the bot stops being a scheduled chore and becomes ambient awareness.
Step 5: Configure human takeover
Decide explicitly when a human enters. Three triggers to configure before launch:
- The customer asks. "Human", "agent", "real person" — hand off immediately. Do not make them argue with a robot about whether they need a person.
- The AI whiffs twice. Two low-confidence answers in a row means escalate. Don't let the bot flounder in front of a paying customer.
- Money or anger. Refund requests and visible frustration skip the queue and go straight to a person.
When takeover triggers, the conversation gets flagged in the dashboard and in the group thread. Whoever picks it up mutes the AI for that conversation and types as themselves; when they're done, they hand back and the AI resumes watch.
One small copy decision that matters more than it looks: have the bot say something honest at the handoff. "Getting a teammate — one moment" beats silence, and it definitely beats pretending the bot and the human are the same entity.
Step 6: Dry run before you announce
Grab two colleagues for fifteen minutes. One plays the confused customer, one plays the angry one. Verify:
- The top twelve questions come back correct, each citing the right source doc
- "I want a human" flags the conversation within seconds
- A reply typed in the group thread reaches the customer — from the desk, not from the agent's personal account
- Takeover and hand-back both work without anything weird showing on the customer side
Then put your t.me/YourBot link where customers already look: order confirmation emails, the site footer, your Instagram bio, the packing slip.
Week one, in numbers
Watch three numbers for the first week: how many conversations the AI closed alone, how often takeover fired, and which questions produced bad answers. That third list is your knowledge-base to-do list. Expect to fix five to ten docs in week one. By week three it slows to a trickle, and the desk mostly runs itself.