AI assistant for business
An assistant that answers customers and staff from your documents, prices and rules — in your tone, with a source, and hands difficult questions to a person with the whole conversation.
01 05 service 1 of 5 in Automation & AI
What’s included
- Assistant trained on your own documents, prices and policies (RAG)
- Answers in your tone of voice, cites the source, admits when it doesn't know
- Escalation to a human with full context
- Multilingual out of the box
- Voice channel if needed
- Analytics on what people actually ask — and where your content fails them
Who it’s for
- Support teams answering the same 30 questions a hundred times a day
- Companies with a large knowledge base nobody reads
- Sales teams losing leads to slow first replies
In short
An AI assistant answers your customers and staff from your own documents, catalogue, prices and rules — in your tone of voice, with a link to the source. When the answer is not in the knowledge base, it says so honestly and hands the conversation to a manager together with its whole history. It works on the site, in Telegram and WhatsApp, and for the team — inside Claude or ChatGPT with access to company data. We pick the model for the task and budget, your data is not used to train it, and the keys are registered to you. The price is counted in hours and fixed before the start; model usage is paid to the provider directly.
What an AI assistant does for a business
It takes over the questions that are asked again and again — and frees people for the ones that really need them.
-
01
Answers at any hour
Customers get an answer in seconds, at night and on weekends, instead of waiting for a manager.
-
02
Knows your business
Products, prices, delivery terms, schedules and rules — from your own materials, not general words from the internet.
-
03
Helps people choose
Asks clarifying questions and suggests a product or a service that fits the request.
-
04
Takes requests
Collects contacts and details and passes the request to the CRM — nothing is lost in the chat.
-
05
Hands over to a person
When a question needs a manager, the whole conversation goes with it, so the customer does not have to repeat anything.
-
06
Helps the team
Staff ask about orders, stock and internal rules and get answers from the real company data.
-
07
Speaks many languages
Multilingual out of the box: it answers in the language it is asked in.
-
08
Shows what people ask
Analytics of real questions shows where your site and documents fail people — and what to add.
Where the assistant works
One knowledge base and one logic — in every channel where your customers and staff already are.
-
01
On the site
A chat window that answers questions about products and services, helps choose and takes a request.
-
02
In messengers
The same assistant in Telegram and WhatsApp — people write where it is convenient for them.
-
03
For staff
Claude or ChatGPT gets access to orders, stock and internal rules — with rights, a log and the ability to withdraw access at any moment.
-
04
By voice
A voice channel if it is needed: the assistant understands speech and answers out loud.
How the assistant finds an answer
A model does not know your business by itself. The method is called RAG: first find it in your materials, then answer.
-
Knowledge is gathered
Documents, catalogue, price lists, terms, frequent questions — everything the assistant may rely on.
-
Indexed by meaning
The materials are split into pieces and indexed so they can be found by meaning, not by exact words.
-
The right pieces are found
A question in someone’s own words finds the passages that actually answer it.
-
The answer is written
The model answers only from what was found — in your tone and with a link to the source.
-
No answer — a person
If nothing was found, the assistant says so and offers a manager instead of making things up.
-
Everything goes to the log
Questions, answers and handovers are visible: gaps in the knowledge base can be filled straight away.
What can be trusted to the assistant
The boundaries are set before launch: what it answers by itself, what only by the rules, and what always goes to a person.
-
Questions about products and services
By itselfSpecifications, availability, how it works, how to choose.
-
Delivery, payment, schedule
By itselfTerms from your documents, always the current version.
-
Taking a request
By itselfContacts and details go straight to the CRM.
-
Prices
By the rulesOnly from the connected, current price list — never from memory.
-
Order status
By the rulesTaken from the CRM or accounting system after the customer is identified.
-
Bookings
By the rulesOnly into free slots of a connected calendar.
-
Discounts and exceptions
Only a personDecisions that cost money are made by a person.
-
Complaints and conflicts
Only a personAn upset customer needs a person; the assistant only passes on the context.
-
Medical, legal and financial advice
Only a personThe assistant gives general information and passes the question to a specialist.
AI assistant or scripted chatbot
They are often confused. The difference is in how they understand a person — and that decides which one your task needs.
| Criterion | Scripted chatbot | AI assistant |
|---|---|---|
| How it understands | buttons and keywords | the meaning of a question in free form |
| A question outside the script | “I did not understand you” | finds the answer in the knowledge base or hands over to a person |
| How it is updated | the script is redrawn | the documents are updated |
| Cost of a conversation | almost zero | model usage per request — visible and capped |
| Predictability | complete | kept by rules, sources and checks |
| Best for | bookings, statuses, payments | consultations and questions in people’s own words |
How we keep the assistant from making things up
A confident wrong answer is worse than none. That is why quality is designed in, not hoped for.
-
01
Only your sources
The assistant answers from the materials you approved, not from general knowledge of the internet.
-
02
A link to the source
Every answer can be checked: the assistant shows which document it came from.
-
03
An honest “I do not know”
When there is no answer in the base, the assistant says so and calls a person.
-
04
Rules and tone
Topics it does not touch, words it does not use, how it greets and how it says goodbye.
-
05
Test questions
A set of real questions with correct answers is run after every change to the base or the settings.
-
06
A spending limit
Simple questions go to a cheap model, complex ones to a strong one, and a monthly limit protects you from surprises.
What we will need from you
An assistant is only as good as the knowledge it is given. Here is what helps most.
-
01
Knowledge materials
Catalogue, price list, terms, instructions, frequent questions — in any format: documents, tables, site pages.
-
02
Real questions
Examples from chats, emails and calls — they become the test set the quality is checked against.
-
03
Rules
What the assistant may say and what it may not: prices, discounts, promises, competitors.
-
04
Tone of voice
How you speak to customers: formally or casually, briefly or in detail. A couple of good replies from your managers is enough.
-
05
Who receives handovers
Where requests and conversations go: the CRM, Telegram, email — and who answers them.
-
06
Accounts in your name
An account with the model provider and the keys registered to the company — so the assistant belongs to you.
Common mistakes with AI assistants
-
An assistant without your data
A bare model answers with general words and confidently makes up prices and terms.
-
Uploading everything in a heap
Old price lists next to new ones give contradictory answers. Out-of-date materials have to be removed.
-
No way to reach a person
An assistant that cannot hand over a conversation turns a question into irritation.
-
Pretending to be a person
People should know they are talking to an assistant — that keeps trust, and in a number of countries it is already required by law.
-
No spending limit
Without routing between models and a monthly cap, the bill is a surprise.
-
Never reading the log
The questions people really ask are the best guide to what is missing on the site and in the base.
Questions about AI assistants
How much does an AI assistant cost?
Development is counted in hours: it depends on the amount of knowledge, the channels and the connections to the CRM and other systems. The sum is fixed before the start; the estimate is free.
How much does it cost to run?
You pay the model provider for actual use: the more conversations, the higher the spend. Simple questions go to a cheap model, the cost of every request is visible, and a monthly limit can be set.
Where does the assistant get its knowledge?
From your materials: catalogue, price lists, terms, instructions, site pages. It finds the relevant passages and answers from them, with a link to the source.
What if the answer is not in the base?
The assistant says so honestly and hands the conversation to a manager together with its history. The question goes to the log so the gap can be filled.
How is the knowledge updated?
You update the document or the price list — the assistant answers from the new version. Out-of-date materials are removed so they do not contradict the current ones.
Can it quote prices and take orders?
Yes, by the rules: prices only from the connected price list, orders and requests straight into the CRM. Discounts and exceptions stay with a person.
Will it work with our CRM and accounting system?
Yes, if the system has an API or an export: the assistant passes requests on and takes order statuses and stock from there.
Is our data safe?
The work goes through the API, where the provider’s terms forbid training models on your data. The assistant sees only what it is allowed to, and when data must not leave the perimeter, a local model is used.
Can staff use it too?
Yes. Through MCP, Claude or ChatGPT gets access to orders, stock and internal rules — with rights and a log of every request.
Will customers see that it is an assistant?
Yes, and we recommend it: honesty keeps trust, and in a number of countries it is already required by law. The assistant’s name and manner are set to match your brand.
How long does it take?
As a guide, an AI feature takes one to four weeks; it depends on the amount of knowledge and the number of connections. The exact dates are fixed before the start.
Online form
Discuss
an AI assistant
Tell us which questions the assistant should answer and where the knowledge lives — catalogue, prices, documents. We will suggest a solution and estimate the work. The consultation is free.