Conversations without a person
How many conversations end with the bot's answer, without a handover to staff.
A chatbot that answers customers from your own documents instead of guessing. When it does not know the answer, it says so and hands the conversation to a person, and you see every such question in the panel.
A chatbot earns its keep on repeated questions. If every conversation is different, a good form and a quick call work better.
Not every chatbot needs a language model. We match the solution to your customers' questions, not to the hype.
| FAQ bot (no language model) | Knowledge-base chatbot | |
|---|---|---|
| How it answers | Recognises the intent of the question and gives an approved answer | Searches passages of your documents and composes an answer from them |
| Best for | A few dozen repeated questions about delivery, returns, payments | Questions in customers' own words about products, specifications, terms, manuals |
| Cost per question | None | Model provider fee on your own account, estimated in the quote |
| Languages | Any language we prepare answers in | Answers in the customer's language from a single knowledge base |
| Out-of-scope question | Offers contact with a person | Says it does not know and hands over; the question lands in the panel |
| Deployment price | from €780 net | from €1,560 net + care from €105/month |
The technique is called retrieval-augmented generation (RAG). The language model does not answer from memory, only from passages the system found in your materials.
Most offers skip this part. For us it is a condition of handover.
Source: Regulation (EU) 2024/1689 (AI Act), Article 50, EUR-Lex.
A chatbot should take load off people and sell. That is why we track what can be verified from day one.
How many conversations end with the bot's answer, without a handover to staff.
A list of topics missing from the base. A ready plan for additions and website content.
Customers rate answers with a thumb, and we fix poor answers in the base.
GA4 events: conversation, product visit, order. You see whether the chatbot pays off.
An assistant that helps choose a wheel and answers in five languages, with no fee per question.
The assistant recognises the intent of a question, copes with typos and answers in five languages. It runs without a language model, so there are no per-query fees.
See the case studyNet prices. We give the exact amount after reviewing your customers' questions, before you decide.
For a few dozen repeated questions, no language model and no per-query fees. Plugin for WordPress, WooCommerce or PrestaShop.
Answers from your documents with the source, human handover, a panel of conversations and unanswered questions, GA4 events.
Monthly review of unanswered questions, knowledge-base updates, answer fixes and a results report.
Advanced assistant with product catalogue search and several languages: €2,080-5,190 net. You pay for the language model directly to the provider on your own account; we estimate it in the quote. Wider business automation is described on the automation and AI page.
A bot that answers common questions (delivery, returns, payments, opening hours) without a language model starts at €780 net. A knowledge-base chatbot that answers from your documents and comes with a conversation panel starts at €1,560 net, and knowledge-base care starts at €105 net per month. An advanced assistant that searches the product catalogue in several languages costs €2,080-5,190 net.
No. If customer questions repeat and have fixed answers, a bot that recognises the intent of the question is enough. That is how the ALU-GUM shopping assistant works: it copes with typos, answers in five languages and costs nothing per query. We add a language model when customers ask in their own words about things written in longer documents: terms, technical data sheets, manuals.
A knowledge-base chatbot answers only from materials you approve: FAQ, terms, price list, product pages, manuals. For every answer it records which passage it used. If the answer is not in the knowledge base, it says so and offers contact with a person instead of guessing. Unanswered questions land in the panel, and we fill the gaps as part of ongoing care.
Yes. Since 2 August 2026, Article 50 of the EU AI Act requires that people interacting with an AI system are informed of it, unless it is obvious from the context. In our deployments the notice is visible with the first message, and the bot never pretends to be a human agent.
Conversations are stored in the EU, by default for 90 days, and only you and the people you choose can access them. If we use an external language model, its provider is listed in the data processing agreement and receives only the question and the relevant knowledge-base passages. The chatbot does not ask for data it does not need.
It should not, and ours does not. It answers only from approved documents, so asked about 60 days for returns it gives the period from your terms, not the one the customer would like to hear. Discounts, deadlines and exceptions are handed over to a person.
A FAQ bot usually goes live within 1-2 weeks of receiving the answers to common questions. A knowledge-base chatbot takes 2-4 weeks, most of it spent organising materials and testing on real questions from your inbox.
Any website through a short script, and in WordPress, WooCommerce and PrestaShop as a plugin or module with settings in the admin panel. The widget waits for cookie consent where required and does not slow down the first page load.
See also: business process automation and AI, system integrations, online stores.
Based on them we will tell you whether a FAQ bot is enough or a knowledge base is needed, and what it will cost. We reply within 24 hours.