Skip to content
Back to blog
12 min read

Business chatbot with your own data: how it works, what it costs and when not to build one

A chatbot that answers with your manuals, price lists and procedures is not something you buy from a template store. We explain how it works under the hood, what GDPR requires, where to host it and what it costs to build and run.

Share

At ASD Solutions, a custom software development company based in Madrid, we build chatbots for companies across Spain that answer from their own documents (manuals, price lists, procedures) and cite the source, hosted in the EU or on-premise. A chatbot like this costs €600 to €3,000 and is delivered in 1 to 4 weeks: around €600 for an internal assistant over a few documents and around €3,000 for a customer-facing one on your website with handover to a person. Fixed price, excluding VAT. If you get few questions or your documentation is a mess, it will not pay off, and we will tell you before we start.

A chatbot with your own data is an assistant that, before answering, searches your company's documents with a technique called RAG (retrieval-augmented generation) and cites the document each answer comes from. Almost every company that asks us about a chatbot wants the same thing: someone, a new employee or a customer, asks a question and gets the right answer taken from the company's own documents, not from the internet. That has been technically possible for a couple of years and it is no longer expensive, but there are traps: personal data that should not leave the EU, made-up answers when the system is poorly built and monthly costs nobody mentioned. This guide covers how a business chatbot with your own data works, when it makes sense, what it really costs and when you are better off not building one.

What is a chatbot with your own data and how is it different from ChatGPT?

ChatGPT and other public assistants answer with what they learned during training. They do not know your 2026 price list, your returns procedure or the technical manual for your machine. A chatbot with your own data uses the same kind of language model, but before answering it searches your documents for the relevant passages and hands them to the model together with the question, with the instruction to answer only from that information.

This technique is called RAG (retrieval-augmented generation). The practical advantage is huge: you do not need to train your own model, which is slow and expensive, and when you change a document the chatbot knows within minutes. Each answer can also cite the document it comes from, which is essential for your team to trust it and for spotting mistakes.

Two ideas often get mixed up. A chatbot answers questions. An AI agent also takes actions: it checks an order status in your ERP, opens a ticket or drafts a quote. AI agents for business are the next step, and we cover them in depth in our guide to private AI agents for business automation. Here we focus on the first step, which is where almost every company should start.

Internal chatbot for your team or customer-facing?

An internal chatbot carries the least risk and pays back soonest. A typical case: a maintenance company with 30 technicians and 400 pages of manuals, equipment sheets and procedures. Today technicians call the workshop manager to ask for a torque value or an error code. With an internal assistant they ask from their phone and get the answer with a link to the right manual page. If the assistant gets it wrong, a professional who knows the trade notices.

A customer-facing chatbot has more visible impact and more risk. It answers out of hours, filters repetitive questions and hands the important ones to a person. But every mistake is seen by a customer, and that brings reputation and legal liability into play. If the bot promises a delivery time or a price that does not exist, the problem is yours.

Our usual advice is to start inside. After two or three months of internal use you learn which questions come in, which documents are missing or contradict each other and how accurate the system really is. With that data, opening it to customers becomes an informed decision instead of a bet.

What data can a chatbot use without breaching GDPR?

If the chatbot only works with manuals, catalogues or procedures with no personal data, GDPR barely comes into it. The problem starts when documents contain names, emails, customer histories or employee data, or when users type personal data into the conversation. At that point you are processing personal data and you need a legal basis, you must inform the people concerned and you need a data processing agreement with every provider involved, including the AI model provider.

Three decisions cut the risk a lot. First, do not index what you do not need: a chatbot that answers product questions does not need your whole CRM. Second, respect permissions: an internal assistant should not show payroll documents to a salesperson just because they sat in the same shared folder. Third, control conversation logs: how long they are kept, who can read them and whether the model provider uses them for training, which the business tiers of most APIs let you switch off by contract.

There is also a newer obligation that mainly affects customer-facing chatbots. The EU AI Act (Regulation (EU) 2024/1689) requires in Article 50(1) that people know they are talking to an AI system, unless that is obvious, and that duty has applied since 2 August 2026. The Digital Omnibus on AI (Regulation (EU) 2026/1744), which pushed back the high-risk obligations, did not move that date: it only gives generative systems already on the market until 2 December 2026 to mark synthetic content (Article 50(2)). In practice a clear notice when the conversation opens is enough, but it is worth building it in from the start.

Where should it be hosted: external API, EU cloud or on-premise?

There are three options and each has its place. The first is using the API of a large model provider with processing in the EU. It is the cheapest way to start and gives the best answer quality, as long as the contract guarantees your data is not used for training and processing stays in Europe. For technical or sales documentation without sensitive data it is usually enough.

The second is running an open model on your own server with a European cloud provider. Current open models answer very well over a well-defined set of documents, and you control where the data lives and who can access it. It costs more per month because you need a GPU machine, but it removes the dependency on a third party. We explain how to choose provider and region in our guide to GDPR and EU hosting.

The third is on-premise: the model runs on a server inside your company and nothing leaves your network. It makes sense when documents are especially sensitive (health, legal, financial data) or when a contract with your own customers requires it. It needs the largest upfront investment and the most maintenance, so we only recommend it when the need is real.

How much does a business chatbot with your own data cost?

These are our prices for an SME, excluding VAT, with a fixed price agreed before starting. An internal assistant over a defined set of documents (manuals, procedures, FAQs), with web access, source citations and a simple panel for uploading documents, starts at €600 and is delivered in 1 to 2 weeks. A customer-facing chatbot on your website, with handover to a person, permission control and conversation logs, comes to around €3,000 and is delivered in 3 to 4 weeks. If it also has to read live data from your ERP or CRM, that connection is quoted separately as an integration.

Then there is the monthly cost, which is what surprises people most. As a rough guide, with an external API and SME volume (a few hundred questions a day), model usage typically runs between €20 and €200 a month, plus hosting for the application and the search database. With your own model on a cloud GPU machine, rental can range from €300 to over €1,000 a month depending on size. On-premise, you swap that fee for buying the server and maintaining it.

And there is a third cost almost nobody budgets for: keeping documents up to date. A chatbot that quotes a two-year-old price list does more harm than having no chatbot. Someone on your team has to own the content. If you want us to watch quality, costs and updates, maintenance starts at €125 a month (5-hour bank).

When should you NOT build a chatbot?

It does not pay off if you get few questions. If your team answers 10 queries a week, a good FAQ document and a reply template do the same job for a fraction of the cost. Nor does it pay off if your documentation is messy or out of date: the chatbot only amplifies what is there, and tidying documents is a separate project that comes first.

It is also a bad idea when mistakes are expensive and nobody reviews the answer. Legal or tax advice to end customers, diagnoses, binding price calculations: a chatbot can help your team prepare the answer, but it should not answer alone. And if what you really need is for a process to stop being manual (entering orders, matching delivery notes), the solution is probably not a conversation but process automation with custom software.

We say this because it is part of how we work: we do not sell AI when it is not needed. In quite a few conversations the conclusion is that a well-built internal search or a form solves the problem, and it is better to know that before spending money.

How do you get started without taking on too much risk?

The order that works best is this. First, pick 30 to 50 real questions your team receives today, with their correct answers. That list is the yardstick: if the chatbot does not get the vast majority right, it does not go live. Then a first set of documents is prepared, the system is built and tested against those questions.

With an acceptable result, it opens to a small group of internal users for a few weeks, conversations are reviewed and documentation gaps are fixed. We work in two-week sprints, so every two weeks you see the system running and decide whether to continue, adjust or stop. After delivery you get 30 days of included support to fix whatever shows up in real use.

If you are not yet sure whether your case calls for a chatbot, an agent or neither, you can start with a free 30-minute call or with our €190 technical audit, which reviews your systems and documents and tells you what is worth building. More on what we do in artificial intelligence for business.

Prices and facts checked on November 9, 2026.

Frequently asked questions about business chatbots

How much does a business chatbot that answers from my documents cost?

A chatbot that answers from your documents costs €600 to €3,000, excluding VAT: around €600 for an internal assistant over a few documents and around €3,000 for a customer-facing one on your website with handover to a person. On top of that comes the monthly cost of the model and hosting, which with an external API typically runs €20 to €200 a month at SME volume.

Is my data used to train the AI model?

It does not have to be. Business tiers of the main APIs let you exclude your data from training by contract, and if you run an open model on your own server, the data never leaves your infrastructure. The key is to check before choosing a provider and put it in writing in the data processing agreement.

What happens if the chatbot makes up an answer?

A well-built system answers only from your documents, cites the source and says it does not know when it finds nothing. The risk is still not zero, which is why we recommend measuring it against a set of real questions before going live and starting with internal use.

Do I have to tell customers they are talking to an AI?

Yes. Article 50(1) of the EU AI Act requires informing people that they are interacting with an AI system, unless that is obvious, and it has applied since 2 August 2026. The 2026 Digital Omnibus on AI did not change that date. A clear notice at the start of the conversation is enough.

How long does it take to get a chatbot running?

An internal assistant over a defined set of documents can be running in 1 to 2 weeks, and a customer-facing one with handover to a person in 3 to 4 weeks. At ASD Solutions they cost €600 to €3,000, fixed price and excluding VAT. We work in two-week sprints, so you see real progress every two weeks.

What is RAG in artificial intelligence?

RAG (retrieval-augmented generation) is the technique a chatbot uses to answer from your documents: it first searches your manuals, price lists or procedures for the passages that match the question, then the model writes the answer using only that information and cites where it comes from. That keeps it from making things up, and you update what it knows by changing the documents, without retraining any model.

Does a chatbot make sense for your company?

Our €190 technical audit reviews your systems and documents and tells you whether a chatbot, another solution or nothing at all pays off, and what it would cost. Diagnosis in 72 hours, report in 5 business days, deductible from the project.

Request the technical audit
Ignacio José Álvarez-Sierra Diez

CEO & Fundador · ASD Solutions

I am Ignacio Álvarez-Sierra, founder of ASD Solutions. I have 8 years of experience building custom software for companies, focused on Go, Node.js, React and cloud-native architectures. Our own senior team works without outsourcing: you talk directly to the people who write the code.

React · TypeScript Go · Node.js · AWS 8 years of experience LinkedIn GitHub

Technical audit for €190

Diagnosis in 72h and full report in 5 days. Credited to your project if you hire us.

See our full process, pricing and technology stack:

Custom Software Development