Contents
Integrating an AI chatbot into your app is technically well under control in 2026: a large language model (LLM) answers questions, guides users through processes or completes simple tasks on their behalf. Costs range from a modest add-on feature to a project in its own right – depending on whether the bot only answers generally, draws on your company knowledge or triggers actions in your systems. This article shows which variant fits and what to budget for.
Four variants of an in-app chatbot
"Chatbot" is a broad term. For planning and budgeting, it helps to distinguish four levels:
| Variant | What the bot does | Effort | Typical use |
|---|---|---|---|
| Rule-based FAQ bot | Fixed answers to recognised questions | Low | A few stable standard questions |
| LLM chat with system instructions | Free-form answers within defined boundaries | Low to medium | General help, drafting text, ideas |
| LLM with company knowledge (RAG) | Answers based on your documents, with sources | Medium | Product advice, support, internal knowledge base |
| AI assistant with actions (tool calls) | Fetches data and triggers processes, e.g. books an appointment | Medium to high | Self-service, orders, business processes |
In practice, most companies end up with the third variant: a bot that only has general model knowledge cannot reliably answer questions about your products, prices or processes. How connecting your own documents works is explained in RAG explained: making company knowledge usable with AI.
What the architecture looks like
An AI chatbot in an app is more than an input field and an API call. A solid architecture typically includes:
- Chat interface in the app: message history, streaming the answer word by word, loading states, feedback buttons and a visible note that an AI is responding.
- Your own backend as intermediary: the app never talks to the AI provider directly. The API key lives only on the server; otherwise it can be extracted from the app and abused.
- Prompt and context management: system instructions, conversation history, user data and search results are assembled on the server.
- Safeguards: per-user rate limits, filters for inadmissible input, length limits and rules on topics the bot must not discuss.
- Hand-over to humans: for complex or sensitive issues the bot routes to support or sales.
- Logging and analysis: which questions come in, where the bot fails, which answers are rated negatively?
Development costs: guide figures by level
The ranges below are guide figures for integration into an existing or already planned app. They depend heavily on data sources, interfaces and quality requirements:
- LLM chat with system instructions: often a small add-on project of a few weeks, at the lower end of our range for simple apps (EUR 5,000–15,000).
- Chatbot with company knowledge (RAG): including data preparation, search, source citations and an admin interface, frequently in the range of mid-complexity projects (EUR 15,000–50,000).
- Assistant with actions in your systems: depending on the number of interfaces and security requirements, at the upper end of the mid range or as a platform of its own from EUR 50,000.
AI-assisted development can reduce these efforts by up to 40 %, especially for interfaces, integrations and tests. A more detailed breakdown is in our article What does an app with AI features cost?
Running costs: what each message costs
Beyond development there are ongoing costs. Most AI providers bill by tokens – fragments of words; as a rough rule, one English word is around one token, longer or compound words more. You pay for input tokens (system instructions, history, documents) and output tokens (the answer), often at different rates.
For an initial estimate, use this calculation:
- Estimate average input tokens per request (for RAG bots often much more than the user's question, because document excerpts are sent along)
- Estimate average output tokens per answer
- Multiply by the current prices of the chosen model
- Scale to the expected number of requests per month and add a buffer for peaks
Add hosting, a vector database, monitoring and maintenance. Model prices differ by a large factor depending on provider and capability tier – smaller models are sufficient for many support questions. A detailed guide with cost-saving tips is in calculating LLM API costs.
Data protection and legal points
Users often reveal more in a chat than in a form. Clarify early which provider processes the data, whether a data processing agreement is in place, where the servers are located and whether inputs are used for training. Options range from major cloud providers with EU hosting to self-hosted open-source models.
Users should also be able to tell that they are talking to an AI – the EU AI Act sets out transparency obligations for such systems that are being phased in. Details on consent, retention periods and provider choice are in implementing AI features in a GDPR-compliant way (this is not legal advice).
Ensuring quality: testing before launch
A chatbot that quotes wrong prices or invents facts does more harm than good. Proven measures:
- Build a test set of typical and critical questions with expected answers and run it automatically on every change
- Instruct the bot to say honestly "I don't know" when it lacks knowledge, and to hand over
- Have answers backed by sources from your documents
- Deliberately try to manipulate the bot (prompt injection) before users do
- Start with a limited user group and review conversation logs regularly
Conclusion
Integrating an AI chatbot into your app pays off when it has a clear job: answering support questions, explaining products or simplifying processes. Cost and value are driven by the chosen level, a secure architecture with your own backend, a sound data protection concept and systematic testing. If you have a specific project in mind, our offer wizard gets you a fixed-price offer after a free analysis.
Frequently Asked Questions
How much does an AI chatbot in an app cost?
A simple LLM chat can often be built as a small add-on project in the range of EUR 5,000–15,000; a bot with company knowledge is frequently around EUR 15,000–50,000. Running costs per request come on top and depend on the model and usage volume.
Can I store the AI provider's API key in the app?
No. Keys inside an app can be extracted and misused. Communication with the AI provider always goes through your own backend, which also enforces usage limits.
How do I stop the chatbot from giving wrong answers?
You cannot rule it out entirely, but you can reduce it significantly: by grounding the bot in verified company documents, citing sources, using clear system instructions and running an automated test set. For sensitive topics, the bot should hand over to a human.
Can an AI chatbot be GDPR-compliant?
Yes, with a suitable provider, a data processing agreement, transparent user information and data minimisation. Alternatively, open-source models can be run on your own servers in the EU.