Multilingual customer service without more staff

The first Norwegian customer emails in Norwegian, and someone on the team answers in Swedish. That works, because Norwegian and Swedish are close. Then comes the Danish customer, and the Finnish one, and the German who writes in English about VAT. At some point it stops working, and the question becomes how many languages customer service should handle and what that costs. This article is about building multilingual customer service without staffing a department per language.

Why is answering in English not enough?

Because customers buy and ask in their own language even when they understand English. CSA Research surveyed 8,709 consumers in 29 countries and found that 76 percent prefer to buy products with information in their own language, and that 40 percent never buy from websites in other languages. That concerns the purchase, but the mechanism is the same in customer service: whoever cannot find the answer in their language either gets in touch or leaves.

On top of that, cross-border shopping is growing. In its Nordic report from spring 2026, PostNord describes consumers comparing prices more, shopping more across borders and making clear demands on deliveries and returns. An online store that starts selling in Norway and Denmark therefore gets not only more orders, but more questions in languages nobody on the team writes daily.

Nielsen Norman Group points to a common hole in this picture: a site can be translated while the user-generated content, such as reviews and comments, is not, and then the experience suffers anyway. The same applies to customer service content. The help centre may be translated while the reply templates and status notices exist only in the original language.

Which languages should you actually offer?

The languages where you have customers who buy, not the languages where you would like to have customers. Three questions decide the list:

1. Where do the orders come from? Take the share of orders per country over the past twelve months. A country with two percent of orders rarely justifies its own staffed language, but often a translated help centre.
2. Which language do customers write in? Count incoming tickets per language, not per country. Many Danish and Finnish customers write in English, and some Norwegian customers write in Swedish.
3. Which tickets are they? Order status and returns can be resolved in self-service in any language. Complaints, compensation and contract questions require someone who understands the nuances.

A common and reasonable setup for a Swedish company selling across the Nordics: Swedish and English staffed, Norwegian and Danish in self-service and chat with Swedish agents behind them, Finnish only if the volume carries it. Finnish is not a Scandinavian neighbour language but a language of its own, and anyone assuming Finnish customers will settle for Swedish is usually wrong.

What do the rules say about language in customer service?

There is no general requirement to answer in a particular language, but the information a customer needs to exercise their rights must be understandable. The Swedish Consumer Agency describes what information a company must give before a distance contract is concluded, including name, company registration number, telephone number, email address and street address. And under the Swedish act on contract terms, terms must be drafted in clear and understandable language, so the consumer has a reasonable opportunity to review them.

In practice that means the more important the information, the worse an approximate translation serves. Opening hours and delivery times can stand being machine translated. Right of withdrawal, complaints, guarantees and contract terms should be reviewed by someone who knows both the language and the subject, in every language where they are published. If you sell in several EU countries, national rules are added on top; that is a question for your lawyer, not for customer service.

How do you build the knowledge base in several languages without doubling the work?

By writing the answer once in a source language and linking every translation to the original, so that a change in the original becomes a visible task in all other languages. That is the difference between a multilingual knowledge base and several parallel knowledge bases that drift apart.

Four rules that hold it together:

• One original per question. Decide which language is the source, usually your home language, and that changes are always made there first.
• Linked versions. Every translation knows which article it is a translation of. When the original changes, the translations are flagged as out of date until someone has looked at them.
• The same owner. Whoever owns the subject owns it in every language. The returns policy is owned by logistics, whether the article is in Swedish or Danish.
• Translate the exceptions first. It is in the exceptions that errors cost. A short answer that reads a little stiffly in Danish does no harm; an exception lost in translation does.

We did exactly this with the blog you are reading: every Swedish article has an English version that points back to the original. The same principle applies in a reviewed knowledge base, where help centre, chat and inbox draw their answers from the same articles, in the language the customer writes in. How an individual article should be written to be translatable at all without losing the exceptions is covered in the article on writing help articles.

When is machine translation enough, and when does it do harm?

Machine translation is excellent for understanding what the customer writes, useful for short standard answers, and risky for anything that is a decision. The risks are concrete: an amount that comes out wrong, a deadline that disappears, or a term that means something else in the neighbouring language.

The cost of a bad automatic answer is higher than many expect. Gartner reports that only 27 percent of customers would try a chatbot again after a negative experience. A customer who received an incomprehensible answer is unlikely to come back to that channel, and ends up in the phone queue or at a competitor.

So draw the line by consequence, not by language:

Add a rule that applies whatever the language: the customer must be able to reach a person. In a survey of 3,566 customers in February and March 2026, Gartner found that 87 percent consider it essential to be able to reach an agent when a company uses generative AI. In a language you do not staff, that route needs to be explicit: say in the customer's language that the answer will come in English and when, instead of letting the bot keep guessing.

How do you staff and measure several languages?

You staff the languages where tickets require a person, and let self-service carry the rest. The order of calculation is the same as in the article on staffing, but per language: inflow per hour and channel, handling time, and then subtract what self-service measurably takes. A small language almost never carries a staffed real-time channel of its own, but it easily carries a translated help centre and a chat that answers from it.

Then measure per language, not only in total. Four numbers are enough:

• Share of tickets per language, and how it develops. If one language rises quickly, it is time to review staffing.
• Self-service rate per language. If it sits far below your home language, the help centre in that language is thinner than you think.
• Response time per language. A language that always waits until the next day is a policy, and it should be deliberate and stated.
• Customer satisfaction per language. A lower rating in Danish than in Swedish usually points to translations that are understandable but not precise.

What to do

1. Count tickets per language for four weeks, not per country. That is the list to build from.
2. Decide a level per language: staffed, self-service with handover to English, or nothing. Write the decision down and why.
3. Appoint a source language and require that every change is made there first.
4. Translate the twenty most common questions and have someone who knows the language review the exceptions, not just the short answer.
5. Set a review rule for terms and legal content: they may never be published machine translated.
6. Write the handover text in every language: what happens now, in which language the answer comes and when.
7. Follow four numbers per language every month and read the comments from the languages you do not staff.

Common questions

Is a help centre in English enough?

For customers who read English comfortably it works, but it costs you volume. CSA Research's survey shows that 76 percent prefer to buy in their own language and that 40 percent never buy from websites in other languages, and whoever cannot find the answer in their language contacts you instead. A reasonable middle ground is to translate the most common questions per market and keep the rest in English, with a clear route to a person.

Can we let AI translate answers in real time in the chat?

For understanding the customer and for short standard answers it works well. For decisions about money, deadlines or rights, the answer should come from a reviewed article in that language, not from a translation on the fly. Build the chat so that it answers from the knowledge base when there is a reviewed answer in the language, and otherwise says in the customer's language that a person is taking over.

How do we handle Norwegian and Danish when the team only speaks Swedish?

Write the content in Norwegian and Danish even if you reply in Swedish in staffed channels. It is in self-service that the difference shows, and translating twenty articles is cheap. In email and chat, Swedish often works for Norwegian customers, but Danish customers read Swedish less easily than many Swedes assume. Say clearly which language you reply in, so it becomes a choice and not a surprise.

What does it cost to add a language?

The recurring cost is the maintenance, not the first translation. Count on a one-off effort for the most common questions, and after that time every time the original changes: a changed returns policy becomes one task per language. That is why the link between original and translation is decisive, and why the number of languages should be a deliberate decision rather than something that grows by itself.

Sources

• Survey of 8,709 Consumers in 29 Countries Finds That 76% Prefer Purchasing Products With Information in Their Own Language — CSA Research, 2020
• Ny PostNord-rapport: Så förändras e-handeln i Norden — PostNord, 2026
• Translate User-Generated Content for Global Audiences — Nielsen Norman Group, read 2026
• Informationskrav vid distansavtal: regler för företag — Swedish Consumer Agency, read 2026
• Avtalsvillkorslagen: för företag — Swedish Consumer Agency, read 2026
• Gartner Survey Finds Only 27% of Customers Would Try a Chatbot Again After a Negative Experience — Gartner, 2026
• Gartner Survey Finds 87% of Customers Say Companies Using GenAI for Customer Service Must Provide Access to a Human Agent — Gartner, 2026