AI chatbots for SMEs: what they actually cost (and when they're not worth it)

At a ten-person real estate agency outside Namur, the same question comes up ten times a day: is this property still available, at what price, can we visit this week. A garage facing the same question about a repair timeline, a shop rewriting the same quote: the problem is identical, only the industry changes.

The fix rarely costs EUR 30,000 to 150,000, the range most benchmarks quote when they're calibrated for large companies. At the scale of an SME with 2 to 49 employees, a chatbot connected to existing documentation runs in the thousands of euros, with a three-figure monthly subscription.

Rule-based chatbot vs AI chatbot

A rule-based chatbot works through a decision tree: buttons, a predefined path, a scripted answer. It routes people to the right page or collects an email, but the moment a question falls outside the script, it gets stuck.

An AI chatbot (LLM + RAG) understands a freely worded question, retrieves the answer from the company's actual documents, then rephrases it in context. It combines several sources, keeps track of a conversation, and improves with use.

📌 A rule-based chatbot handles questions anticipated in advance. An AI chatbot handles the questions customers actually ask, in their own words.

🔍 The test before starting: how many times a week do you answer a variation of the same question, and does that answer already rely on information that exists somewhere (CRM, spreadsheet, documentation)? If the answer is no to both, it's not the right moment yet.

Three use cases, three trades

Availability and pricing. At a real estate agency, a chatbot connected to the CRM answers availability questions without an agent interrupting a viewing to check a spreadsheet. At a garage, the same logic applies to repair timelines and slot availability. Observed result at SME scale: a good share of first-level questions handled without human intervention, response time down to seconds.

Finding scattered information. Contract terms, internal procedures, price scales: at a service firm without a structured intranet, a chatbot connected to these sources acts as an internal search engine, useful to the team as much as to the client. 👉 It also makes the knowledge transferable, rather than dependent on one person answering from memory.

Qualifying a request. A shop or a trade business can have basic questions (need, budget, timeline) asked upfront before routing to the right person. ⚠️ This assumes enough volume: below a handful of contacts a week, a plain form does the same job without the chatbot's cost.

What it actually costs

A quote for "a chatbot on our documentation" can range from EUR 12,000 to 45,000 depending on the exact scope (INSEIL, 2026), with document volume and integration depth mattering more than industry.

At SME scale, the real ranges observed on the 2026 market:

Two market players confirm this order of magnitude: between EUR 1,800 and 6,000 for a question-answering chatbot with qualification (Hutch Agency, 2026), and an SME sweet spot between EUR 5,000 and 12,000 for a documentation-based RAG chatbot (Radiank, 2026).

💡 What actually moves the price is rarely the AI model itself. It's the integration with existing tools and the work on the document corpus.

The ROI math, in one line

Monthly savings = automated requests x average cost of a human interaction.

Example for a 10-person agency: 50 requests a week, 10 minutes each at a EUR 28/hour fully loaded cost. A chatbot handling 60% of them saves roughly EUR 560/month, or EUR 6,700/year. On a EUR 6,000 setup, the payback period lands around 9 to 11 months.

What makes a project fail

• Starting from the tool rather than the volume. "We want an AI chatbot" isn't a brief until someone has counted the repeated questions.
• A corpus that doesn't exist yet. If the information lives in one person's head, it has to be written down first.
• No human fallback planned. A chatbot that can't hand off to a human on an out-of-scope question frustrates more than it helps.

The Nightborn method

A free 30-minute discovery call to say honestly whether a chatbot is the right answer, with no hard sell. Then a scoping workshop that prices out what it's worth and what it delivers. Building only starts once the scoping confirms it's worth it, connected to the tools already in place rather than replacing them. That's the same principle that let Medicheck get documents and emails moving without manual rework.

What to remember

• An AI chatbot for an SME runs in the thousands of euros, not tens of thousands.
• It needs a real volume of repeated questions and a usable corpus. Without both, it's not the right time.
• The payback period typically sits between 6 and 12 months at this scale.

Judging whether a chatbot is worth it doesn't take a EUR 10,000 audit. A 30-minute conversation is enough.