How an AI manager cut the first reply from 5 minutes to 5 seconds
One assistant across three accounts — Telegram, Instagram Direct and a numbered Viber: answers in the brand’s voice, understands voice notes, qualifies the interest, chases chats that went quiet and hands warm leads to a manager.

First reply = 5 min
It took up to five minutes for a manager to reach the chat. In that time the buyer had already written to the shop next door and been answered there — the lead went cold before the first conversation.
One assistant: first contact, follow-up and looking after the customer

One bot across every account
Telegram, Instagram Direct and a numbered Viber are run by a single assistant: one logic, one memory, one tone of reply. However many accounts a buyer writes from, they pick up where they left off — the context is shared across all of them.

Follow-up on its own
Returns to chats that went quiet, keeps the accounts running and gets back in touch without a manager.
Tyvek braceletgreen · 19 mm · no printQuantity60 pcsPrice3.75 UAH / pcTotal225 UAHOrderships todayText, voice and cards
Understands voice notes and text, and sends photos, links and product cards straight into the thread.
Brand voice and memory
Answers in the shop’s own voice and holds the context of a conversation — it never re-asks what it already knows.
Qualify and hand over
Reads the interest and sends warm leads into the “Orders” group with everything it has collected.
A human at any moment
One command and a manager takes the chat over — the bot steps aside and goes quiet in that thread only.
Five real conversations and what happens in them
One person across Instagram, Telegram and Viber, one memory
The same customer writes from Instagram today and Viber next week. The bot recognises them as one person and carries the memory across. It does not re-ask what they already told it in the other channel, and it does not start the pitch over.
Answers around the clock, honest about opening hours
The bot is on around the clock and always answers a question. But when a customer is waiting on something that only happens during work, a quote, a call back, a dispatch, it says up front what the hours are and which days are off. It does not promise something that will not happen until morning.
Five messages in a row, one answer
People type the way they think. A line, another line, then a correction. The bot waits for the burst to finish, reads all of it together, and sends one composed reply to the whole thought, instead of five half-answers tripping over each other.
It sees an attachment the way a manager would
The logo off a screenshot is only the start. A customer sends a payment screenshot, and the bot understands the order is paid. Sends an invoice, and the bot reads the line items and the total. Shares a photo of the bands they liked, and the bot recognises the model and asks about the size and the run. It reads the attachment in the context of the sale, like a manager, not just as an image.
A request from a chat becomes an order in the CRM
The bots are wired into the client's accounting system. The request collected in the conversation is written into CRM as a ready customer order: line items, quantity, price, delivery and the channel the customer came from. The manager only has to post the document. In parallel a qualification card lands in the technical group, so the team sees a new customer without opening the accounting system.
The result, in numbers
How reply speed, conversion, throughput and handoff quality changed after the assistant went live.
From the first bot to ongoing support




Launch on Telegram
An AI assistant across three accounts. It answers in seconds in the brand voice, runs the first exchange, qualifies the lead and hands the warm ones to the "Orders" admin group. Manual takeover and a stop command from day one.
Instagram Direct
The same bot moves into Instagram through ManyChat and n8n, on one prompt and one memory. Leads from Telegram and Instagram meet in a single group for the team.
One memory across channels
Forwarding and context work between any channels. The bot recognises one person wherever they write and does not start the conversation over.
Product catalogue
The bot knows the range, the materials and the runs, and assembles the order right in the chat.
Following up the leads who went quiet
The bot comes back to the people who asked a price and disappeared. It waits its interval and writes first, with the request already in hand, so the conversation does not die.
A third channel — Viber
The bot answers from a phone number on Viber, in the same voice and with the same memory as Telegram and Instagram.
Control panel and working hours
A kill-switch with a per-day schedule and auto-restore, file and payment recognition, follow-up. Operations built for the managers.
CRM for the sales team
The bots are wired into the accounting system: an order from a chat is written into CRM as a document, with line items, delivery and the source channel, while a qualification card lands in the technical group. Sales and accounting in one place.
On support and development
The project is under ongoing support: we watch the conversations, fix the edge cases, and every month add scenarios, channels and integrations for the sales work.
The technologies we used
Three accounts and the “Orders” admin group in one bot.
Direct as a full sales channel.
Answers from a phone number, no official API.
Official Meta app, fully verified.
The fast model for simple, high-volume chats.
The engine wiring every channel together.
Memory, schedule and the bots’ stop state.
One gateway for WhatsApp, Instagram and Telegram.
KeyCRM
The client's CRM: orders from chats are created without manual entry.
The same approach, a different niche
AI assistant for a beauty clinic
Appointment booking, Q&A and visit reminders for a cosmetic clinic — plus a cosmetics catalogue with a cart and checkout, an online store inside Telegram.

Want the same for your business?
Let’s talk through your case and show how the assistant comes together for your sales.
