Contents8
In short
Translation sits between the customer and the team, both ways: the customer writes in their language, the team reads and replies in its own.
Canned replies, the help centre and automatic emails are translated once, reviewed, then reused.
A dispute, a complaint or a message that commits you goes through someone who reads the language, and personal data decides the translation tool.
A new market does not stop at a translated home page. From the first sales, questions arrive in the customer’s language: a late delivery, an invoice to correct, a product to exchange. Hiring one person per language only makes sense once volume is established, and that is often what holds an opening back.
Machine translation changes the calculation: the same team can answer in several languages, with review placed where it counts. This article says what it brings to support, what is prepared once and for all, and what stays with a person who reads the language. It extends our articles on website translation and opening new markets.
Why support in the customer’s language matters
After-sales service is the moment when a customer decides whether to come back. CSA Research’s survey measures it:
of respondents are more likely to buy the same brand again if customer care is in their language
of those most confident in reading English still prefer customer care in their own language
The second figure matters for a company that thinks English-only support will do: even customers at ease in English prefer to be answered in their own language.
What translation changes for the team
The principle fits in one loop: everyone writes in their own language, and translation bridges both ways.
A customer message translated both ways
- Arrival
The message arrives in the customer’s language, which is detected.
- Reading
The team reads the translation in its language, with the original alongside.
- Reply
It is written in the team’s language, often from a canned reply.
- Return
The reply is translated into the customer’s language, with the glossary, and reviewed when it commits you.
Several support tools build in this loop. Zendesk’s documentation, for example, describes automatic translation of inbound and outbound messages on all its channels except voice, email included, and keeps the original message next to its translation in the event log. Other tools offer comparable features, and a shared email inbox can get the same loop through an automation: we describe this kind of set-up in automate customer support: AI sorts, humans review.
The benefit is concrete: the person who knows the product answers directly, instead of passing the message to a colleague who speaks the language but knows the case less well.
What is translated once and for all
Much of what a customer reads is not written on the fly. Those texts deserve careful translation and review by someone who writes the language, because they will be read hundreds of times:
- Canned replies: order tracking, lead times, return procedure, forgotten password. Translated and reviewed once, they are reused as they are.
- Automatic emails: confirmation, shipping, invoice, reminders. Date, price and address formats count as much as the words, as localize, don’t just translate reminds us.
- The help centre: a question well documented in the customer’s language is a question that never reaches support.
- Names and terms: product, offer and option names, kept in a brand glossary that the translation tool applies to every message.
A translation memory keeps these validated texts: when a canned reply changes, only the modified sentence goes back to translation.
What stays with a person who reads the language
Machine translation can be wrong without saying so: a lost negation, a misplaced amount, a tone that is too curt. On an order-tracking question, the stakes are low. On a dispute, they are not.
Machine translation is enough
A question about an order or a delivery
A request for product information
An acknowledgement or a canned reply already reviewed
A person reviews or writes
A complaint, a dispute, a contested refund
A message that commits you: warranty, terms, promised lead time
An unhappy customer, where tone matters as much as substance
This split makes no profession redundant. It moves the work: the translator or native reviewer translates canned replies and the help centre, validates the glossary and reviews sensitive messages, instead of translating every parcel-tracking email. A team member who reads the language, even without writing it perfectly, can also act as a second pair of eyes on delicate messages. The tool is chosen on your own messages, following the method in DeepL, Google or an LLM, and the place of review follows the logic of translating your website with AI: automatic or reviewed?.
Personal data decides the tool
A customer email contains a name, an address, an order number, sometimes more. Translating it means sending it to a translation service: you need to know where it goes, how long it stays there and whether it trains a model. The answers change with the plan, not just the provider. The obligations that apply are summarised in AI Act and GDPR: what an SME using AI must do.
From the field
Our data governance grid puts named customer data in the most sensitive category: it only goes to a self-hosted model, on our server or the client’s, or to a European provider bound by contract, never to a consumer plan. The channel matters too: on Palmora Property, a six-language real-estate listings site we operate, WhatsApp being blocked in China, contact goes through WeChat, LINE or Telegram depending on the visitor’s language.
One last point on transparency: telling customers that the reply was machine-translated is honest, and lets them flag a sentence they misunderstood. Zendesk provides a “Translated by AI” notice that the administrator can turn on.
Hire later, and only when volume justifies it
Translating support does not replace hiring forever: it postpones it to the moment it pays off. While a market is being tested, the existing team answers with a reviewed translation. When volume in one language grows, a person who speaks it becomes a profitable investment, and they find canned replies, a glossary and a memory already in place.
The cost of translating one message is small; what costs is the preparation: glossary, canned replies, native review, tool set-up. We detail per-character prices in website translation cost: per word, per page, per market. To decide which market to test first, see German, Spanish or Italian: which market to open first and opening an English-speaking market: beyond language.
Customer messages are also a source: the words customers use to describe a problem are the words they type into a search engine, raw material for keywords abroad: redo the research, don’t translate it. For a catalogue, recurring questions point to the product pages to rework, following product pages for export. And if you publish software, the application’s own messages are translated separately, as explained in translating software or an app: strings, context, updates.
Where to start
List the ten replies your team sends most often, have them translated and reviewed in the language of your first market, and store them in your support tool. Then turn on translation of incoming messages, and decide in writing which types of message go through review before sending.
To set up this loop with your tool and your data rules, tell us about your support and your languages: we start from your real messages.
Frequently asked questions
Can I answer a customer in a language I do not speak?
Yes, with two-way translation: you read their message translated, you reply in your own language, and the reply goes out translated. For routine questions, this is reliable if the canned replies have been reviewed. For a dispute or a message that commits you, have the reply reviewed by someone who reads the language before sending it.
Should I tell customers the reply is translated?
It is recommended. A short notice, such as “machine-translated reply”, is honest and lets the customer flag a sentence they misunderstood instead of staying with a misunderstanding. Some support tools display this notice automatically when the administrator turns it on. It does not prevent a careful reply: it explains an awkward turn of phrase.
Which tool should I use to translate support emails?
First, the translation feature of your support tool, if it has one. Otherwise, a translation service or an AI model connected to the inbox through an automation. The choice depends on the quality measured on your own messages, on glossary support, and above all on what happens to the text you send, since it contains personal data.
Is a translator still useful if support is machine-translated?
Yes. They translate and review what will be read hundreds of times, such as canned replies, the help centre and automatic emails, validate the glossary and review sensitive messages. Machine translation handles the volume of routine exchanges; the translator keeps the quality of what commits the company.
Sources
- Zendesk, Translating conversations in the Zendesk Agent Workspace (new tab)support.zendesk.com
Zendesk, Translating conversations in the Zendesk Agent Workspace
Read on support.zendesk.com(new tab)Jim · founder of Brain Plus AI
Also in translation and new markets
Practical guide8 min
Brand glossary: keeping translations consistent
The same product called three different things on three pages, and readers wonder whether there are three products. A brand glossary fixes the terms that are not up for debate, before the first translation, whatever the tool or the translator.
Understand7 min
Translation memory: what a company gains from it
A sentence translated once should not be paid for twice. The translation memory makes sure of that at every site update, provided it belongs to you.
Choose and price7 min
DeepL, Google or an LLM: which translates your trade best
Every provider publishes a test it wins, and independent evaluations crown a different winner from one year to the next. For a specialised text, the only reliable answer comes from a trial on your own texts, language by language, judged by someone who reads the language.