AI employee · Support
AI Support that knows your product
It answers customers on your website and in messengers around the clock — from your documentation, guides, and past tickets. The hard cases it hands off to a human specialist along with the full context of the conversation.
- Replies in seconds
- Knows your documentation
- Escalates to a human
Why support can't keep up
Most tickets are the same questions, and the answers are already in the documentation.
| Характеристика | Значение |
|---|---|
| Customers wait hours for a reply | Overnight and on weekends the tickets pile up. By the time someone replies, some customers have already gone to a competitor or left a bad review. |
| The same questions over and over | Up to 70% of tickets are routine: how to pay, where the documents are, why something isn't working. Your specialists answer them on a loop. |
| Knowledge lives in people's heads | A new hire takes months to get up to speed, and some answers are known to just one person — who's on vacation. |
Tasks
What AI Support does
It learns from your knowledge base, guides, and past conversations — and answers the way your best specialist would.
- Answers on your website, in Telegram, and by email — instantly, around the clock
- Draws on your documentation, guides, and history of resolved tickets
- Asks for details when a question is unclear
- Hands the hard cases to a human specialist along with the conversation history
- Guides the customer toward the goal: a request, a payment, a booking
- Shows what people ask about most — a signal for improving the product
Human vs AI employee
| Response time | 15 minutes to a day | seconds |
|---|---|---|
| Hours | on a schedule | 24/7 |
| Routine questions | take up to 70% of the time | handled automatically |
| Cost | from 45,000 ₽/mo | from 15,000 ₽/mo |
How we launch it
| Характеристика | Значение |
|---|---|
| 1Gather the knowledge | Documentation, guides, FAQs, ticket history |
| 2Train it on them | The model answers strictly from your materials, without making things up |
| 3Put it on a channel | A widget on your site, a Telegram bot, or an email connection |
| 4Review and refine | We go through real conversations and expand the knowledge base |
Guided launch
The first months — with a dedicated manager
The AI is not launched “as is” and left unattended. For the first 1–3 months a dedicated manager runs your process: tunes the system to you, checks every result and fixes mistakes — while the AI learns on your real data in parallel. Nothing goes live without a human in control.
| Характеристика | Значение |
|---|---|
| 1–3 monthsDedicated manager | A dedicated specialist runs your process by hand together with the AI: tunes it to your documents and rules, checks every result, corrects mistakes. The system learns on your data. |
| Decision pointAutopilot or ongoing support | Once quality is stable, you decide: move the process to autopilot (cheaper, the AI works on its own, the manager on standby) or keep permanent support. |
| BeyondOngoing oversight, if you want it | For critical processes — primary documents, paid ads, contracts — many keep the manager for good: they backstop it, develop scenarios and own the result. |
This removes the main fear of automation — “what if it makes a mistake”. Until the AI has proven its accuracy on your tasks, a human is accountable for the result. We agree the format and cost of support around your process.
Where your data is stored
Conversations with customers contain personal data, so processing runs on our infrastructure in Russia or on your own server. The model answers only from your materials — it doesn't send the content of conversations to foreign services or use it to train anyone else's models.
More about a local neural network on your own server →FAQ
Won't it start making up answers?
The model answers from your knowledge base, not from general assumptions. If the answer isn't in your materials, it says so honestly and passes the question to a human — that behavior is enforced strictly.
How fast can it go live?
If your documentation already exists — 3 to 7 days. Most of the time goes not into the tech, but into gathering and tidying up the materials it will answer from.
What about complex tickets?
It doesn't try to solve them itself — it escalates: handing them to a specialist along with the conversation history and its own read of the problem. The human steps in with context already in hand, not from scratch.
Can it run in Telegram and on the website at the same time?
Yes — the channels connect in parallel to a shared knowledge base. A customer gets the same answer in the website chat and in the messenger alike.
Can it lead toward a sale?
Yes, if it's built into the script: suggest the right plan, take a request, book a consultation. You draw the line between support and sales yourself.
Related reading
We cover this topic in more depth on the blog
- Building a Support Chatbot That Doesn’t HallucinateA support bot stops inventing answers when it replies from your documentation through RAG and escalates honestly, unlike scripted bots or bare ChatGPT.AI adoption · 8 min read · Read →
- AI Chatbot vs Scripted Bot: Which One You NeedA script is predictable for clear-cut processes; AI understands free-form questions but costs more. The comparison, and why a hybrid usually beats either alone.Development · 7 min read · Read →
- Adding an AI Assistant to Your WebsiteAn AI assistant answers visitors instantly from your own content, guides them to an enquiry and escalates the hard cases. Embedded on top of the existing site.Development · 7 min read · Read →
We'll launch it on your documentation
Show us your knowledge base or website — we'll build a demo of AI Support on your own materials.
Contact
Let us try it on your data
Describe the task and send a couple of typical documents or examples — we will show how the system handles them, before any contract.
- info@stitex.com
- Telegram
- @StitexBot
- Response time
- Within 1–2 business hours
B2B & Commercial:
- All prices include 22% VAT
- Electronic tax invoices & formal agreement
- 3-month warranty on completed code