Pricing guide
What an AI chatbot really costs — and when a menu bot is enough
The gap between €500 and €15,000 is not intelligence — it is how deep the bot reaches into your processes.
The most expensive mistake chatbot buyers make is paying AI prices for something that could have been a button menu. In 2026 the market splits into three clear tiers. A rule-based chatbot with predefined menus — good for standard questions about opening hours, delivery or pricing — costs roughly €500–2,000. An AI chatbot built on a large language model and grounded in your company data through RAG, answering in natural language from your documents, catalogues and procedures, lands at €3,000–15,000. A bot wired into your operations — taking orders, booking appointments, writing to the CRM — goes past €15,000, because you are paying for integrations, not conversation. Every AI variant also carries a recurring cost that proposals routinely omit: the model API, typically €20–500 a month depending on conversation volume. For international teams commissioning this from a Romanian development partner, EU-based engineering rates are a large part of why these figures stay reasonable by Western European standards. Before fixing any budget, ask the uncomfortable question first: how many of your customers' messages actually need AI to get answered?
Let’s talk about your project
Message us on WhatsApp or send an email — you talk directly to a developer.
office@northdan.com · +40 752 070 247
Key takeaways
Rule-based bot: €500–2,000
Predefined answers, buttons, fixed scenarios. Covers repetitive questions surprisingly well when they are few and predictable.
AI bot trained on your data: €3,000–15,000
An LLM plus RAG over your documents, products and procedures. Handles natural-language phrasings a menu tree would never catch.
Process-integrated bot: over €15,000
Takes orders, books appointments, opens tickets, writes to your CRM. Here the budget goes into integrations and action safety, not conversation.
Recurring API cost: €20–500/month
Every AI conversation consumes paid tokens at the model provider. At high volume, the monthly bill can eventually overtake the original build cost.
What pushes an AI chatbot's price up
The first driver is the source of truth: a bot answering from twenty pages of FAQ is cheap to build; one searching thousands of documents, price catalogues that change weekly and each customer's history needs a serious RAG architecture with indexing, automatic refresh and testing against invented answers. The second driver is actions: the moment the bot can do something — place a booking, take an order — you need validations, confirmations and hard limits, because a mistake is no longer a bad answer but a wrong transaction.
The third driver is quality control. An unsupervised chatbot can promise discounts that do not exist or hand out legally sensitive information. Budget from day one for human escalation and a periodic review of conversations — a few hours a month is enough — otherwise the money saved at implementation gets paid back in confused customers.
When a good form beats an expensive bot
If you receive fewer than 10–15 messages a day and most are quote requests or appointment bookings, a well-designed form plus an automatic email reply solves the problem at a tenth of the budget. A chatbot becomes worthwhile only when repetitive conversations consume real hours of your team's time, or when customers write outside business hours and sales leak away until morning.
A simple test before signing anything: export your last 200 customer messages and group them by theme. If 80% fall into 10 questions, a €1,000 rule-based bot covers them. If the questions are varied, tied to specific products and each customer's situation, only then does the €3,000–15,000 AI option earn its price. The figures are indicative; the proportions of the test hold at any scale.
Frequently asked questions
Why am I paying monthly for the chatbot after buying it?
Because every AI-generated reply calls a provider's language model, which bills token consumption. A few hundred conversations a month means €20–50; thousands can reach €300–500. A rule-based bot without AI has no such cost.
Can the chatbot give my customers wrong answers?
Yes — any language model can produce inaccurate output. The risk drops with strict RAG over your own documents, instructions that force it to say 'I don't know' outside its scope, and automatic escalation to a human. Ask for all three mechanisms explicitly in the proposal.
How long does an AI chatbot take to implement?
A rule-based bot is configured in 1–2 weeks. The AI variant trained on company data takes roughly 4–8 weeks, a good share of which is cleaning and structuring your documents — answer quality tracks the quality of the material it learns from.
Does a chatbot replace my support staff?
Realistically it absorbs the repetitive share — usually between a third and two thirds of volume — and leaves humans the complicated cases and the angry customers. Companies that treated it as a full replacement lost clients; those that treated it as a filter gained time without losing quality.
Similar pages
Related resources
Let’s talk about your project
Message us on WhatsApp or send an email — you talk directly to a developer.
office@northdan.com · +40 752 070 247