# An AI sales assistant on Telegram that knows when to hand off

> Source: <https://dev.to/gfirik/an-ai-sales-assistant-on-telegram-that-knows-when-to-hand-off-1pe1>
> Published: 2026-11-19 20:09:56+00:00

**Recommendation:** Ground the assistant's answers in your team's live data source and define explicit handoff triggers—booking intent, decisions needing a person, or a request for a human—then pass the full chat history to agents so customers never repeat themselves.
More in the playbook "Designing agent systems": <https://wpnews.pro/playbooks/agent-architecture>

Aviago is a tour operator that sells through its own Telegram bot. Its sales team had the problem every chat-based business knows: they were answering the same questions over and over. How much is a flight to this city? Is there anything cheaper? What dates work? Every one of those answers was typed by a person.

So I built an AI sales assistant into the bot. It now answers most of the common questions on its own, in the same tone as Aviago's main sales assistant, and hands the buyer to a person when a person is needed.

This post is how it works, and the design decision that matters most: when the AI stops talking.

The first reply decides a lot. A well-known [Harvard Business Review study of 2,241 US companies](https://hbr.org/2011/03/the-short-life-of-online-sales-leads) found that firms that contacted a web lead within an hour were nearly seven times more likely to qualify it than firms that waited even one hour longer.

Chat makes this sharper. On Telegram the buyer is already in a conversation, so they expect a chat-speed answer.

But speed alone is not the goal. A [Gartner survey published in August 2026](https://www.gartner.com/en/newsroom/press-releases/2026-08-04-gartner-survey-finds-87-percent-of-customers-say-companies-using-genai-for-customer-service-must-provide-access-to-a-human-agent0) found that 87% of customers say companies using generative AI for customer service must still give them access to a human. People want the fast answer, and they want to know a real person is there when it matters.

That is the brief for a sales assistant: answer fast, answer correctly, and get out of the way at the right moment.

Aviago's Telegram bot already existed. I didn't replace it. I split the conversation into parts and put the AI between them, so it handles the parts it does well and leaves the rest to people.

The model changed along the way. The first version ran on Groq; it now runs on Claude. Because the workflow is separate from the model, switching was a change in one place, not a rebuild.

The fastest way to lose a buyer's trust is an AI that invents a price. So the assistant doesn't "know" Aviago's prices. It reads them from the same sheet the team updates.

That also means nobody has to retrain the AI when a price changes. The team edits the sheet, as they always did, and the next answer is correct.

Buyers were used to talking to Aviago's main sales assistant. A generic chatbot voice would have felt like a downgrade. So I tuned the assistant to answer the way the team already does: same tone, same kind of phrasing. For the buyer, the conversation feels like the same shop.

How far the assistant goes on its own is not a technical limit. It's a choice the business makes, and it can change.

The assistant hands off when the buyer is ready to book, has a request that needs a person's decision, or simply asks for a person. At that moment the sales team gets the full chat history at once in Discord. Nobody asks the buyer to repeat themselves. A handoff that makes the customer start over is worse than no AI at all.

Most of the common questions are now answered by the assistant, in the team's own tone. The sales team spends its time on buyers who are ready to talk, not on typing the same price again and again.

As Aviago's founder put it: "with his help we added an AI assistant that can work autonomously and hand tasks over to human admins when needed. … Most importantly, we successfully deployed all of these solutions."

If your buyers write to you on Telegram, WhatsApp or Instagram and your team answers the same questions all day, this pattern fits: an assistant that answers from your own data, in your own voice, and hands off to your team at the moment you choose.

I build these as fixed-price projects. Read the [Aviago project page](https://fhsystems.dev/works/aviago), or [tell me how your sales chat works today](https://fhsystems.dev/contact) and I'll show you where an assistant would help and where it shouldn't.

*Originally published on [fhsystems.dev](https://fhsystems.dev/blog/telegram-ai-sales-assistant-human-handoff).*
