How to build a knowledge-base chatbot for a business in Azerbaijan

Customers write at any hour, in Azerbaijani, Russian or English, and they ask the same questions again and again: prices, delivery, opening hours, documents, how to book. A chatbot can take this load off your team, but only if it answers correctly. This guide explains how a knowledge-base chatbot works, what you need to prepare before building one, and how to tell whether it does its job.

Why a knowledge base matters

A general-purpose bot answers from whatever it learned somewhere else, so on questions about your company it guesses. A knowledge-base chatbot answers from your own materials: the website, PDF and Word files, spreadsheets, a product catalogue, internal rules. When a price or a rule changes, you update the knowledge base, and the bot's answers change with it. A good bot can also show the document or page an answer comes from, so anyone can check it.

Step 1. Define the bot's role and limits

Before any technology, write down in a few sentences who the bot is and whom it serves:

• Audience: customers on the website, or staff on an internal portal.
• Topics: what it answers, for example products, delivery and payment, and what it politely declines.
• Tone: formal or friendly, how it greets people, how long its answers are.
• Red lines: what it must never do, such as promise a discount, give a legal opinion or reveal internal data.
• Handover: when the conversation goes to a person, and who that person is.

Step 2. Prepare the materials

The quality of answers depends directly on the quality of the materials. A practical checklist:

1. The pages of your website that customers actually read: services, prices, delivery, contacts, FAQ.
2. Documents in PDF or Word: terms, instructions, price lists, policies.
3. Spreadsheets and a catalogue with current products, plans or services.
4. Internal rules your staff follow when they answer customers.
5. Real questions from customers: from chats, e-mail and calls. They show what people really ask and how they phrase it.
6. A person responsible for keeping these materials up to date.

Remove outdated and contradictory versions first: if two documents give different prices, the bot cannot know which one is right.

Step 3. Decide what the bot should do besides answering

Answers are only half the value. A business chatbot can lead a conversation to a result: collect a request, book an appointment, help choose a product or a plan, and pass the contact to your team by e-mail, Telegram or into your CRM. The next step is an AI agent that takes actions through your systems' API: it checks an order status, creates a booking, fills in a form. List the actions you want and the systems they touch.

Step 4. Choose languages and channels

In Azerbaijan the bot should answer in the customer's language: Azerbaijani, Russian or English, replying in the language it is written to. Then choose where it works: on the website, in Telegram, WhatsApp, Instagram, in a mobile app or on an internal portal. One knowledge base can serve all of these channels.

If your customers prefer to speak, consider a voice bot: voice messages and phone calls in natural Azerbaijani speech, built on the technology of our product Voices.az for Azerbaijani text to speech and speech recognition.

Step 5. Test on real questions, including tricky ones

Before launch, test the bot on the real customer questions you collected in step 2. Include off-topic questions and deliberate attempts to trick it: asking it to ignore its rules, to reveal internal instructions or data, to promise something it should not. A well-built bot keeps to its role, does not invent answers and does not give in.

Step 6. Launch, measure and keep the knowledge base current

After launch, watch the conversations. What to measure:

• The most frequent questions: they show what to improve on your website and in your materials.
• Questions the bot could not answer: each one is a gap in the knowledge base.
• Handovers to a person: how often and why they happen.
• Requests and conversions: how many conversations end with a request, and which of them turn into sales.

Update the knowledge base whenever prices, rules or products change. A bot with stale materials gives stale answers.

Features in demand today

• Understanding images: a customer sends a photo, a screenshot or a scan, and the bot reads it.
• Answers with sources, so they can be verified.
• Memory of returning customers, so the bot does not ask the same questions twice.
• Proactive messages: a greeting on the right page, a reminder about an unfinished order or an appointment.
• Conversation analytics and protection against attempts to make the bot leak internal data.

How we work

1. We look at the task and your materials and agree on the result.
2. We build the knowledge base and scenarios and set the bot's tone and limits.
3. We test it on real questions, including attempts to pull it off topic.
4. We launch it in your channels, watch the quality and keep the knowledge base current.

The price and the timeline depend on the task: the number of materials, channels, languages and integrations. That is why we give a quotation after the task is described, not before.

See it in action

The chat on learningmachine.az is itself a working example of such a bot: ask it about our services and projects in Azerbaijani, Russian or English. When you are ready, see our chatbot development service or describe your task, and we will suggest a solution.