AI customer service is software that reads customer messages, looks up the answer in your knowledge base, and writes a reply without a human involved. For European businesses, the good systems also answer in the customer's own language, cite their sources, hand off to a human when they're uncertain, and keep data in the EU. This article explains how it works, what it costs, and how to start. For the product side, see our AI customer service overview.
What AI customer service actually does
An AI customer service agent isn't the same as an old-school chatbot with fixed answers. Traditional chatbots work with decision trees: if the customer types "return," they get answer A. If they type "order," answer B. The moment the question is phrased slightly differently, it breaks.
AI customer service works differently. The system reads the entire question, understands the intent, searches your website, help pages, and documents, and writes a reply in natural language. It can handle multiple questions in one message, remember context within the conversation, and tell the difference between "I want to return this" and "how long do I have to return this."
The technology behind it is called retrieval-augmented generation (RAG). The AI retrieves relevant information from your own sources and uses a language model to formulate the answer. This grounding sharply reduces invented answers, but it does not eliminate them. A model can still misread a policy or fill a gap in your content with something plausible. That's why good systems show citations, so you can check where an answer came from, and hand off to a human when the sources don't clearly cover the question. We go deeper in why support AI agents hallucinate and how to fix it.
Why now, and why the Netherlands and Benelux are ahead
The Netherlands has several characteristics that make AI customer service particularly relevant. The market is multilingual: customers communicate in Dutch, English, German, and French. Traditional customer service requires agents who speak multiple languages, which is expensive and hard to scale. A modern AI agent handles all of these languages in the same model, without a separate translation step.
Dutch businesses are also early adopters of digital technology. The Dutch e-commerce market passed €35 billion in 2024, according to Thuiswinkel.org's market figures, and customers expect fast, digital service. At the same time, labor costs are rising and good customer service agents are hard to find.
Finally, the GDPR and the EU AI Act set requirements for how you handle customer data. European businesses benefit from tools that host in the EU and are compliant by default.
How it works: step by step
- Add knowledge sources. Point the AI to your website URL, help articles, or upload documents. The AI indexes everything within minutes.
- Install the chat widget. Paste a script tag on your website, or install via the Shopify App Store. The widget appears as a chat window in the bottom right.
- Connect email. Forward your support email address to the AI. Same knowledge base, same quality, but over email.
- Review initial conversations. Check the first 50-100 conversations. Where did the AI escalate? Add the missing knowledge. After a week, the system runs at full capacity.
- Set up handoff rules. Decide when the AI hands off: at low confidence, for specific topics (complaints, cancellations), or when the customer asks for a human.
What it costs
The cost structure of AI customer service is fundamentally different from traditional help desks. Traditional tools charge per agent per month, so cost grows with headcount. AI tools often charge per reply or per resolution, so cost grows with volume.
A comparison for a typical European SMB handling 500 conversations per month:
- Traditional (3 agents on a per-seat help desk): three seat licences plus three salaries, every month
- AI-first (Keloa Growth): €49/month for 1,500 AI replies, plus human agents only for complex issues
If the AI resolves 70% (an assumption for this example; your rate depends on your questions and content), it handles 350 conversations. The remaining 150 go to your team. You need fewer agents, or your existing team spends more time on complex, high-value interactions.
Keloa starts with a free Starter plan (50 AI replies per month) and paid plans from €49/month with no limits on the number of agents.
Common mistakes
Neglecting the knowledge base. The AI is only as good as the information you give it. If your return policy isn't on your website, the AI can't find it. Invest the first week in completing your knowledge base.
Setting expectations too high. AI handles a large share of routine questions, not 100%. Complex complaints, emotional situations, and edge cases require humans. Think of AI as your first line, not a replacement.
No human escalation path. Customers should always be able to ask for a human. Without an escalation path, you frustrate the customers the AI can't help.
Choosing the wrong tool. Not every chatbot is the same. Look at: AI quality (test with your own real questions), EU hosting, pricing model (per seat vs per reply), and whether AI is included by default or an expensive add-on.
Which tool fits your business?
For European businesses, a few criteria are especially important:
- EU hosting: your customer data should stay in Europe. Keloa hosts in the EU. Many other tools are US-based, so ask each vendor where data is stored.
- Native language support: the AI should understand and write fluently in Dutch, German, and French, not translate from English.
- No per-seat costs: for small teams, per-seat models are relatively expensive. Per-reply or per-workspace is usually more affordable.
- GDPR compliance: data processing agreement, sub-processor list, data retention policy. Read our GDPR chatbot guide.
Also check our comparison of AI customer service tools for small businesses and the EU AI customer service guide.
Get started in 10 minutes
- Go to keloa.ai and create a free account.
- Add your website URL as a knowledge source.
- Install the chat widget.
- Set up your support email.
- Review the first conversations and refine.
Most teams are live within 10-15 minutes. The AI starts answering immediately based on your website content. After a week of fine-tuning based on real conversations, you'll reach a stable resolution rate.
Frequently asked questions
Is AI customer service reliable enough? Yes, when set up correctly, but not perfect. The AI answers from your own knowledge base and cites its sources, which makes errors rarer and easier to spot. When it's uncertain, it hands off to a human. Reliability depends mostly on the quality of your knowledge base, plus regular review of real conversations.
Does it work in Dutch and other European languages? Yes. Today's frontier language models understand and write fluent Dutch, German, and French, including informal language and regional variations. No translation layer needed.
What if the AI gives a wrong answer? Check the source: is the correct information in your knowledge base? If not, add it. The AI learns from new content immediately. You can also raise the confidence threshold so the AI only answers when it's certain.
Do I still need customer service agents? Yes. AI handles the routine; your team focuses on the complex. The ratio shifts: where most time previously went to repetitive questions, your team now spends that time on complaints, edge cases, and relationship building.
Is it GDPR compliant? With Keloa, yes. Data is hosted in the EU, with a data processing agreement and a public sub-processor list. For US-hosted tools, you'll need to assess compliance separately.