Home / Insights / Multilingual Customer Service in Belgium with AI
Guide

Multilingual Customer Service in Belgium with AI

Belgian language law leaves customer service language free and binds invoices, labels, staff documents and public services. One AI assistant can answer each customer in their own language from one knowledge base and hand over to the matching team.

Summarize with AI Prompt copied. Paste it into the chat

A Flemish organisation with customers in Wallonia, Brussels, the German-speaking east and the Netherlands needs customer service in Dutch, French and German, and in practice in English too. Belgian language law leaves the language of a customer conversation free: it binds invoices, product labels, staff documents and public services. So one AI assistant can answer every customer in their own language from one knowledge base, and pass each conversation that needs a person to the team that speaks that language.

This guide is for the IT manager, head of customer service or informatiemanager at a Flemish organisation of 250 to 5,000 people with a ticketing system such as Zendesk, Freshdesk or Dynamics 365 and an ICT partner. Crux Digits builds the AI layer next to that setup; your ICT partner keeps the infrastructure.

Which languages does a Flemish organisation's customer service need?

Four Belgian language regions, plus the Netherlands and English: that is the realistic scope for a Flemish organisation selling across Belgium.

Belgium has a Dutch-language region, a French-language region, a German-language region around Eupen and Sankt Vith, and the bilingual Brussels-Capital Region of 19 municipalities, as the federal employment ministry sets out. Add customers in the Netherlands and English-speaking residents, and you have five variants: Dutch for Flanders, Dutch for the Netherlands, French, German and English.

The useful step is to treat language as a routing attribute rather than a translation job. Dutch for a customer in Antwerp and Dutch for a customer in Utrecht is the same language with different facts: different VAT rules, different consumer law, different delivery terms. So the assistant tracks two fields per conversation, language and country, and the knowledge base marks which country each fact applies to.

What do Belgian language laws require for customer communication?

For a private company, nothing in Belgian law fixes the language of a chat, call or message to a customer. Article 30 of the Constitution makes language use free and allows regulation only for acts of public authority and judicial matters, with Article 129 giving the Communities power over staff relations and legally prescribed company documents.

The Flemish government's Taalwetwijzer states it directly: traders and companies decide which language they use towards customers, with exceptions from business language law and consumer law. Those exceptions are specific:

Legally prescribed company documents, such as the mandatory mentions on an invoice, follow the language of the operating seat (exploitatiezetel), according to the Taalwetwijzer on business. For a seat in the Dutch-language region that is Dutch under the decree of 19 July 1973. Since an amendment of 7 July 2017, Article 5 §2/1 of the consolidated decree allows an additional legally valid invoice version in an official EU or EEA language understood by all parties, but only when the recipient is established in another EU or EEA member state. The Dutch version prevails if the two differ. A seat in Brussels may choose Dutch or French.

Price indications to consumers under Article VI.3 of the Code of Economic Law fall under the same regime. Mandatory labelling, manuals and warranty certificates must be in a language the average consumer understands, taking into account the language area where the product is offered. Food and toys have their own rules: toys sold in Brussels need manuals in at least Dutch and French.

Public services, and private parties entrusted with a public task or a concession, fall under administrative language law: the coordinated laws of 18 July 1966 and the Flemish decree of 30 June 1981, both in the Taalwetwijzer's regulation overview. This guide covers private organisations; for a public service that law governs every answer.

For an AI assistant: the chat reply, the WhatsApp answer and the web-form response are free. The invoice it attaches, a quoted price and a linked manual are not.

What do the language rules require inside the organisation?

In the Dutch-language region, staff contact and staff documents must be in Dutch under the decree of 19 July 1973.

Articles 2, 3 and 5 cover spoken and written contact and every document meant for staff. The operating seat decides, defined as the establishment the employee is attached to. For French-speaking colleagues working from Ghent or Mechelen, work instructions, procedures and scripts exist in Dutch. Article 5 §3 sets a procedure under which the employer must add a translation in one or more languages, when the staff composition justifies it and the employee members of the works council ask for it unanimously. The arrangement runs for one year, must be in writing and is served on the supervising officials within a month. Breaches in the Dutch-language region make the document void and can lead to administrative or criminal fines.

Other regions run other regimes, per the federal overview. In Brussels, staff documents follow the employee's language, and documents for all staff need separate Dutch and French versions. In the French-language region the decree of 30 June 1982 requires French. In the German-language region, written documents must be in German.

Here an AI assistant helps. The procedures your staff work from stay in Dutch, and the assistant still answers a customer in Liège in French from those same documents. Whether an AI-written handover summary counts as a document intended for staff is not addressed in the decree or the Taalwetwijzer. The cautious design writes the summary in the working language of the operating seat and attaches the customer's original messages unchanged.

How does one AI assistant answer each customer in their own language?

Pull quote from Crux Digits: Belgian language law decides the language of the invoice and the staff manual, not of the chat.

It replies in the language of each message, answers only from approved sources, and cites its source so a colleague can check it.

Four design choices make that work. First, detect language per message: customers switch, especially in Brussels. When detection is unsure, a short greeting in Dutch, French and English lets the customer choose. Second, retrieval works across languages, so a French question finds the Dutch master text and the answer is written in French. For opening hours, stock or delivery windows, that is enough. For binding texts such as terms and conditions or warranty terms, the assistant quotes and links the approved version in the customer's language, and hands over if none exists. Third, every fact carries its country, so a Dutch customer never receives a Belgian return rule. Fourth, Article 50 of the AI Act requires telling people they are interacting with an AI system, clearly and at the latest at the first interaction; the European Commission states the transparency rules apply from August 2026. In our reading, a disclosure the customer cannot read is not clear, so write it in every language you serve. Our Article 50 guide covers the wording.

What the team gains is checkable. A customer in Eupen writing on a Sunday evening gets a first answer in German from the same source as a customer in Bruges, and the same question gets the same answer in every language.

How should the assistant hand over to the right language team?

Route on the conversation's language at the moment of handover, written to a ticket field, not on the first message.

The ticketing tools show why. Zendesk's intelligent triage fills a standard Language field with about 150 possible values, but Zendesk notes that the field is not updated on each end-user reply, and using the classifications in workflows requires the Copilot add-on. A customer who opens in English and switches to French can land in the wrong queue. Microsoft's Dynamics 365 Contact Center lets an administrator build a language-based ruleset on a workstream that routes escalated calls to language-specific queues, based on the caller's interaction language with the voice agent. Both work well when the AI layer writes its own conversation-language field at handover and the rule reads it.

A handover that lets a colleague continue without asking again holds seven fields: conversation language, country, the customer's original messages verbatim, a summary, the sources the assistant used, the reason for handover, and anything the customer has already been told. When the French-speaking team is offline, the assistant says so in French, states when the team is back, and creates the ticket in the right queue. The team starts the next morning with summarised French conversations only.

How do you keep source documents consistent across languages?

Give every document one master language, record which master version each translation matches, and let the assistant answer only from translations that are current.

In Freshdesk's multilingual knowledge base, an editor clicks Mark other translations as outdated each time the primary article changes, translators mark their version up to date after editing, and a per-language list shows outdated and untranslated articles. The step is manual, which is exactly where an AI assistant can drift: it reads whatever is published.

Seven fields per source document close that gap: document ID, master language, language of this version, country scope (Belgium, the Netherlands or both), the master version this translation matches, owner, and status (binding or informational). The decision rule follows from the fields. A current translation is used as is. An outdated informational translation is skipped and the assistant answers from the master in the customer's language. An outdated binding text triggers a handover.

Then test it: ask the same thirty real questions in Dutch, French, German and English every month and compare. Any difference between languages points to a source that has drifted, and you fix the document, not the assistant.

What does this look like for one Flemish wholesaler?

The chat runs in four languages, the invoice stays Dutch, and French-speaking staff get French conversations with Dutch procedures.

Assumptions for this example: operating seat in Mechelen, a service team of 20 there including four French-speaking colleagues, customers in Belgium, the Netherlands and Germany, and Zendesk as ticketing system. The assistant answers on website chat and WhatsApp in the customer's language. The invoice it links to carries its mandatory mentions in Dutch. For a customer established in Germany, the invoice can additionally have a legally valid German version under Article 5 §2/1; for a customer in Namur it cannot, because that customer is established in Belgium. A translation of the procedures follows if the employee members of the works council invoke Article 5 §3. Have your legal counsel confirm the invoice and staff points; the Taalwetwijzer is a good starting point.

Off the shelf, or built into your own systems?

Start off the shelf when your sources are public; build when the assistant has to write into your ticketing fields and portal under your own version rules.

SimplyBoost is our own product: an AI support agent that answers from your website, PDFs, FAQ and catalogue on website chat, WhatsApp, Instagram and Facebook Messenger. It replies in Dutch, French, German, Spanish and English, which covers Belgium's three national languages plus English. An account includes three AI agents, so one can serve the public site and another a logged-in portal, where it recognises users by verified email. It hands over to a person or to Zendesk, Freshdesk, Gorgias or Slack with the conversation attached. See SimplyBoost.

A custom build fits when the handover must fill fields in Dynamics 365 or Salesforce Service Cloud, or when translation status lives in your own document system. Crux Digits starts with an AI audit at €2,500, builds a Production-ready MVP at €20,000, and takes it to production from €50,000, all excluding VAT, and hands over the code.

Where does Crux Digits fit?

We design and build the AI layer: language and country detection, the handover payload, the routing field your ticketing system reads, the source-version checks and the cross-language test set.

Your ICT partner keeps identity, network and hosting, and your ticketing vendor stays. We do not give legal advice; we build so that your counsel's answer on invoices and staff documents is configured, not coded. For our work in Belgium, see AI consultant Belgium.

More on the service side: AI customer support agent.

Last updated: 28 October 2026.

Talking to an AI consultant in Flanders?

A consultant tells you where AI pays off; Crux Digits also builds it. A fixed price per step, one named expert, and the Belgian rules and funding built in.

AI consultant in Belgium →

Frequently asked questions

Must our chatbot answer in Dutch because our seat is in Flanders?

No. Article 30 of the Constitution makes language use free, and the Flemish Taalwetwijzer confirms that companies choose the language they use towards customers. The decree of 19 July 1973 binds staff relations and legally prescribed documents such as mandatory invoice mentions, not customer conversations. Your assistant can answer a customer in Liège in French and a customer in Eupen in German. Documents it sends along, such as the invoice, still follow the language rules of your operating seat.

Can we send a French invoice to a customer in Wallonia?

Not as the legally valid version, if your operating seat is in the Dutch-language region. The mandatory invoice mentions must be in Dutch. Article 5 §2/1 of the decree allows an additional legally valid version in another EU or EEA language only when the recipient is established in another member state, so it covers a customer in Lille or Aachen, not one in Namur. Ask your legal counsel how to handle a courtesy translation for Belgian customers.

Do we need a separate knowledge base per language?

No. One knowledge base with a master language per document and versioned translations works better, because every language then answers from the same facts. Retrieval works across languages, so the assistant can answer a French question from a Dutch operational text. Binding texts such as terms and conditions need an approved version per language. Track which master version each translation matches, and let the assistant skip outdated informational translations and hand over on outdated binding ones.

What happens when a customer switches language mid-conversation?

The assistant answers in the language of the latest message and records the conversation language again at handover. That matters for routing: Zendesk documents that the Language field from intelligent triage is not updated on each end-user reply, so a rule on that field can follow the first message. Write a conversation-language field at the moment of handover and route on it, so a customer who switched to French reaches the French-speaking team.
Our AI services AI consultancy AI automation AI agents AI implementation Pricing

Want any of this applied to your business?

We turn these concepts into working tools: grounded, safe and measurable. Start with a free consultation.

Book a free consultation →