In short: in 2026 an AI chatbot answers 60 to 80% of recurring customer questions using your own documents, and books appointments or looks up orders. Costs: €50 to €300 per month for SaaS, €2,950 to €20,000 for a custom build. It pays off once your team spends more than 10 hours a week on the same questions.
What an AI chatbot can and cannot do in 2026
The chatbot of 2019 was a decision tree: click 'opening hours' and you got a canned reply. The chatbot of 2026 is a language model (Claude, GPT or an open model) that is handed your own knowledge. The technique is called RAG, retrieval-augmented generation: for every question the bot first pulls the relevant passages from your own sources (website, FAQ, manuals, terms, product data) and writes an answer from them. No training, no script per question, and a change to your price list shows up immediately.
What such a bot does well:
- Answering questions from your own sources. Delivery times, return policy, what is in package X, whether product Y suits situation Z. With citations, in your brand's tone.
- Looking something up or acting through an integration. Where is my order, which table is free on Saturday, is this course still available. For that the bot talks to your online store, calendar or CRM.
- Scheduling appointments and intakes. A quote request, a trial class, a reservation. The bot collects the details, checks availability and writes it into your system.
- Handing over to a human. With the full conversation history, so your customer does not start over.
What not to expect from it:
- Judgement calls on exceptions. Goodwill gestures, emotional complaints, a discount outside the rules: that stays human work, and a good bot hands over.
- Knowledge that is written down nowhere. If your return policy only lives in a colleague's head, the bot cannot find it. Documenting is 30% of the work.
- Being flawless without boundaries. A language model without guardrails sometimes invents an answer. So you restrict the bot to your own sources, let it say 'I do not know' and test it with real customer questions before launch.
Use cases by sector
Retail and e-commerce. 'Where is my order' makes up 30 to 40% of all tickets. A bot that pulls order status from Shopify or WooCommerce and starts the return process takes the pressure off your inbox. Product advice ('which size', 'does this fit my model') is the second win.
Hospitality. Reservations, group requests, allergens, parking, 'are you open on Easter Monday'. A bot on your site and on WhatsApp handles that while your staff is on the floor. Connected to your reservation system it books the table directly. See what we build for hospitality.
Fitness and sports. Scheduling trial classes, membership questions, timetables, cancelling or pausing. For a gym the bot is mainly a salesperson outside opening hours: a large share of trial-class requests comes in during the evening.
Installation and construction. Reporting a fault, requesting a quote with photos and an address. The bot completes the intake so your planner only has to slot it in.
Costs: SaaS, no-code or custom
There are 3 routes; which one fits depends on how much you want to connect and how much control you want.
| Route | Examples | Cost 2026 | Strong at | Limit |
|---|---|---|---|---|
| SaaS chatbot | Intercom Fin, Zendesk AI, Tidio, Chatbase | €50 – 300 per month, often plus €0.50 – 1 per resolved conversation | Fast to launch, ready-made widget, inbox included | Limited integrations, data sits with the vendor, price scales with volume |
| No-code or low-code | Voiceflow, Botpress, n8n with an LLM API | €30 – 150 per month plus 20 – 60 hours of your own work | Flexible, affordable, quick to adjust | Someone on your team has to maintain it, integrations stay fiddly |
| Custom (RAG plus integrations) | Your own bot on Claude or GPT via the API, connected to your systems | €2,950 – 20,000 one-off plus €20 – 150 per month in API costs | Full control, your own data, deep integrations, EU hosting | Higher entry price, you need a build partner |
The API costs of the language model itself are modest: an average conversation of 6 messages costs between €0.01 and €0.05 in 2026, depending on the model. The real cost is in structuring the knowledge, building the integrations and testing.
ROI calculation: support hours × hourly cost
Work it out in 4 steps, with your own numbers:
- Count the questions. Look back 2 weeks in your mailbox, WhatsApp and phone log. Say 40 questions per working day.
- Measure the time per question. Including looking things up and replying, 5 minutes on average. That is 200 minutes, over 3 hours a day.
- Estimate what the bot takes over. 60 to 70% of volume in the first 3 months is realistic. Take 60%: 2 hours a day, 44 hours a month.
- Multiply by the cost of an hour. A customer service employee costs €30 to €40 per hour including employer charges. 44 hours × €35 = €1,540 per month.
Against that €1,540 stands a SaaS subscription of €150 per month, or a custom bot of €7,500 that pays for itself in 5 months. Add the revenue side: evening questions answered instantly convert better than those answered the next morning. Below 10 hours a week of recurring questions the sums get tight, and a good FAQ page is usually the better first step.
Privacy, GDPR and the AI Act
A chatbot processes personal data: names, email addresses, order numbers. That falls under the GDPR and, since August 2, 2026, under the transparency rules of the European AI Act as well. What to watch:
- Say that it is AI. The AI Act requires that a visitor knows they are talking to an AI system. A bot that poses as an employee is not allowed.
- Data processing agreement with the AI vendor. Anthropic, OpenAI, Microsoft and Google offer business terms under which your conversations are not used for training. Never use a consumer ChatGPT account for customer data.
- Data minimization. Let the bot retrieve only what it needs. An order status requires an order number plus postcode, not a full customer profile.
- Retention and logging. Decide how long you keep conversations (30 to 90 days is common) and state it in your privacy policy.
- EU hosting where possible. The major models are available in EU regions; keep your knowledge base and logs on a European server.
- Healthcare and finance: extra care. Health and financial data are special categories. Do not let the bot disclose file information there; have it refer to a secure portal instead.
The Dutch Data Protection Authority warned repeatedly in 2024 and 2025 about chatbots that misinform customers or pose as humans. A customer service bot does not fall into the AI Act's high-risk category, but the transparency duty and the GDPR always apply.
Integrations: CRM, online store, calendar and WhatsApp
A bot without integrations is a talking FAQ. The value is in what it is allowed to do:
- Online store (Shopify, WooCommerce, Lightspeed). Order status, starting a return, checking stock. Through the standard APIs, 1 to 3 days of work per integration.
- CRM (HubSpot, Teamleader, Pipedrive). Creating leads with every intake question filled in, attaching the conversation history to the contact.
- Calendar and bookings (Google Calendar, Calendly, Zenchef, Virtuagym). Showing availability and booking directly.
- WhatsApp Business. Since July 2025 Meta charges per template message sent rather than per conversation; replies to customer questions within the 24-hour window are free.
- Accounting and ERP (Exact Online, AFAS). Invoice status and outstanding items for business customers.
Each integration is a set of 'tools' the model may call, with strict boundaries: reading is allowed, writing only what you explicitly permit. That keeps the bot predictable.
Step-by-step plan: live in 4 weeks
- Week 1: inventory. Collect 200 real customer questions from email, phone and WhatsApp. Cluster them. Decide which 3 integrations cover the most volume. Write down the knowledge that is documented nowhere.
- Week 2: build. Set up the knowledge base, choose the model and tone, build the integrations, set up the handover to a human (email, Slack or your helpdesk).
- Week 3: test. Have 3 colleagues ask the bot 100 questions, including hard ones. Measure the share of correct answers; below 90% you do not go live. Finalize the privacy policy and the AI disclosure.
- Week 4: launch and measure. Start on 1 channel, review the conversations daily, add missing knowledge. After 30 days you know the real takeover rate and the savings.
RABZI estimate: an AI chatbot from RABZI starts with an Automation Sprint at €2,950 excl. VAT (1 process, 2 weeks), from €7,500 for an assistant with 3 integrations (online store, CRM, calendar or WhatsApp) and from €19,500 for a RAG assistant spanning several departments. Ongoing management and tuning from €950 per month. Everything runs on your accounts and your data. See AI integration and automation or the full pricing. If you want to look beyond a chatbot, read AI automation for SMEs.
Frequently asked questions
How much does an AI chatbot cost for a small business? A SaaS chatbot costs €50 to €300 per month, depending on the number of conversations. A custom bot that uses your own documents and connects to your online store or calendar costs €2,950 to €20,000 one-off plus €20 to €150 per month in API costs.
Can an AI chatbot use my own documents and website? Yes, in 2026 that is the standard. Through RAG the bot searches your website, FAQ, manuals, terms and product data and answers questions with citations. Change a document and the bot knows immediately.
Is an AI chatbot allowed under the GDPR and the AI Act? Yes, provided you are transparent (the visitor must know it is AI, mandatory since August 2, 2026), have a data processing agreement with the AI vendor, retrieve only the data you need and set a retention period. Health and financial data carry extra requirements.
Does a chatbot replace my customer service team? No. A good bot takes over 60 to 80% of recurring questions and hands the rest to a colleague with full context. Your team gains time for exceptions, complaints and sales.
Conclusion
In 2026 an AI chatbot is no longer a gimmick but a colleague who uses your own knowledge 24 hours a day and talks to your systems. It pays off once you spend more than 10 hours a week on the same questions, and the return sits in the integrations. Start small, with 1 channel and 3 integrations, measure for 30 days and then expand. Want to know what a chatbot would deliver for your business? Book an intro call and you will have an estimate with a fixed price and date within 2 business days.