Multilingual drafts. Already written in the reader's language, with your name at the bottom.
A German buyer's enquiry, the spec sheet a distributor wants in English, the terms of sale that need to go to France. Every time, someone stops, opens a translation tool and pieces the text together by hand. This piece hands over the draft already written in the right language, and the person who signs it reads it and sends it.
The draft arrives ready. Emails, spec sheets, price lists and instructions are prepared in the recipient's language, with your usual structure and your data in them.
The terms that matter aren't translated loosely. Product names, legal wording, Incoterms and payment terms live in a glossary agreed once, and every draft sticks to it.
Nothing goes out unread. The person who signs reads it and sends it. Nothing that commits the business to a price or a confirmation ever goes out on its own.
This page covers a single piece of the system. The other pieces, and how we choose which one to start with, are on the services page.
What changes in practice
Drafting disappears from the working day, and what's left is the review, which only someone who knows the customer can do. On technical content the measured difference is big: twelve professional translators who started from a machine draft rather than a blank page produced on average 74% more words per hour.
The figure comes from a test published by Plitt and Masselot in the Prague Bulletin of Mathematical Linguistics, covering 144,648 words of real content and four target languages. Individual gains ranged from 20% to 131%, so the number that counts is yours, not the average.
The time isn't lost in translating. It's lost in starting from scratch every time an enquiry arrives in the wrong language.
The second change shows after a few weeks. Correspondence with the same market starts to hang together: the same name for the same product, the same wording for the same term, the same tone. Today that consistency depends on who happened to be in the office that day.
People buy in their own language, and often only in their own language
76% of consumers prefer to buy products with information in their own language, and 40% never buy from a website in another language: that's the finding of CSA Research's Can’t Read, Won’t Buy survey of 8,709 consumers in 29 countries.
In the same survey, 66% say they use machine translation anyway when information isn't available in their language. The same applies to a buyer who receives a spec sheet only in Italian: they run it through a machine, and your product description reaches them in words you didn't choose.
That's a sample of private consumers in 29 countries, not purchasing departments, and it's worth saying so. But the direction is the same one you see with every reply to a foreign importer: people who get a message in their own language reply, and people who get something to decipher put it off.
For mandatory information, the language is set by the law of the country
In food, it isn't a matter of courtesy. Article 15 of Regulation (EU) 1169/2011 puts it like this: “mandatory food information shall appear in a language easily understood by the consumers of the Member States where a food is marketed”.
The same article adds that each Member State may require one or more official EU languages on its own territory. So the language on the label depends not on the seller but on the market where the product ends up, and it changes from country to country: the full picture is in our guide to labelling by destination country.
That's why legal wording is never reworded. It goes into the glossary in exactly the required form, and the system copies it rather than translating it. A well-translated version of a mandatory statement is still a wrong one.
The glossary is agreed once and applies to every document
During the analysis we write down a short list: product trade names, legal product names, Incoterms, payment terms, units of measure and the sign-offs you use. These are the words that mustn't change between an email and a spec sheet.
That list is applied to every draft, in every language. If your product goes by a particular name abroad, no draft translates it, and if a payment term is written in one specific way, it always comes out that way. The English spec sheet and the covering email stop saying two different things.
The rest of the text, on the other hand, adapts to the reader. The same information is written differently for a German purchasing department and a Spanish distributor, and that's the work the system does well once the list of fixed terms is clear.
What the system does, step by step
Every outgoing document follows the same route, whatever the target language, and every step leaves a log entry recording what happened and when.
| Step | What happens | What you get |
|---|---|---|
| Languagethe reader's | The recipient's language is picked up from the incoming enquiry or the contact record, not asked for each time. |
Nobody has to remember which language to use with that customer, not even someone dealing with them for the first time. |
| Contentfrom your data | The draft is built from your real data, meaning your price list, availability, terms and previously approved text, not from what the model imagines. |
The numbers in the email are yours, and there's nothing to double-check by hand line by line. |
| Glossaryfixed terms applied | The fixed terms are applied after drafting, so no stylistic choice can override them. |
Product names and legal wording always come out exactly the same, in every language and every document. |
| Reviewby a person | The draft reaches the person who signs it ready to go, with anything the system had to decide on its own highlighted. |
You read it and send it in a minute, instead of rewriting it from scratch in half an hour. |
| Sendingand archiving | The text that was sent is saved with the contact, along with the language and version used. |
Next time you pick up from what you've already said, and the customer never gets two different versions. |
The review, which is what makes the rest fit to send
Machine translation is now standard even among professionals: in the European Language Industry Survey 2026, with 1,058 participants in 45 countries, 63% of independent translators work on text that has already been machine-translated.
In the same survey only 23% of those professionals rated the quality of machine output as high or very high, down from 40% the year before. The people who use it every day rate it lower than those who've only tried it, which is why the human review isn't a precaution to drop once the system has matured.
Post-editing machine translation has its own standard, ISO 18587:2017, which sets out the process and the skills of the person doing it. An industry that has written a standard for the reviewer is saying something very clear about how replaceable that person is.
The two mistakes a machine makes without noticing are the same everywhere: a technical term swapped for a plausible synonym, and a tone that sounds more formal or more familiar in that language than you intended. Neither shows up when you check in Italian, and both are obvious in three seconds to someone who knows the customer.
What goes out on its own, and what waits for a person
Outside working hours, a first-contact reply can go out on its own, in the sender's language, using only information you've already approved. During working hours you decide, channel by channel, with a switch.
Nothing that commits the business ever goes out on its own: prices, quotes, discounts, order confirmations and contract terms always wait for a person, whatever time the enquiry arrives.
When a reply goes out automatically, it says so in one line. From 2 August 2026, Article 50 of the AI Act applies, and readers have to know they're dealing with a system before they interact with it, not afterwards. How we say it is on the AI transparency page, and the reasoning behind keeping a person in the loop is on our page about the principles we build by.
Same mechanism, different name in every trade
The mechanism is identical; the document being rewritten by hand today isn't. It's worth looking at your own case, because that's where you see what it costs you every week.
| Sector | The document being rewritten today | Where you see it |
|---|---|---|
| Food and agricultureand export | The reply to a foreign distributor's enquiry, with price list, spec sheets and terms, which today waits for the one person who speaks English. |
|
| Hospitalityaccommodation and events | Quote requests arriving in three different languages, and arrival instructions rewritten from scratch every single time. |
|
| Restaurantsand bars | The menu and allergen information for foreign guests, which change every season and have to be redone in every language. |
What this piece doesn't do
It isn't a sworn or certified translation service. Where a sworn translation, legalisation or the signature of a sworn translator is needed, that step stays with a qualified professional, and the system stops before it.
It doesn't take responsibility for what goes on the label. It prepares the draft and keeps the agreed wording fixed, while compliance of the product on the market stays with the business that places it there.
It doesn't decide who to write to, and it doesn't handle incoming enquiries. The first reply to someone getting in touch is a different problem, handled by Inbox AI; documents that come in and need entering into the business software are handled by automatic document entry.
Questions and answers
How many languages can it handle?
The ones you actually sell in, and it's best to start with two or three. Every language you add brings a glossary to fill in and needs someone who can read the output.
Opening eight on day one means eight languages nobody is checking. Add them when the customers exist, not when the idea of them does.
Who guarantees the translation is right?
A person reading it, which stays mandatory before anything is sent. In the European Language Industry Survey 2026, with 1,058 participants in 45 countries, only 23% of independent professionals rated machine translation quality as high or very high, down from 40% the year before.
Post-editing machine translation has its own standard, ISO 18587:2017. The system takes away the drafting, not the responsibility of the person who signs.
Can it also handle labels and mandatory documents?
It prepares the draft and keeps the glossary consistent, but responsibility stays with the business that places the product on the market. Article 15 of Regulation (EU) 1169/2011 requires mandatory food information to appear in a language easily understood by consumers in the country where the food is sold, and allows Member States to require one or more official languages.
Legal wording is never freely reworded: it goes into the glossary in its exact form and stays that way.
Will the recipient notice a system wrote the draft?
If a reply goes out automatically outside working hours, it says so in one line. From 2 August 2026, Article 50 of the AI Act applies, requiring people to know they're dealing with a system before they interact with it, not afterwards.
Drafts that a person reads and sends are that person's own correspondence, like a letter written with a spellchecker.
How do you measure whether it's working?
By the time it takes to prepare a multilingual message, measured first on a real batch and compared again afterwards. The second number is the share of drafts a person has to rewrite rather than just correct: if it rises, the glossary is incomplete or the language was added too early.
The third, where relevant, is the time between a foreign buyer's enquiry and the reply going out.
Notes on sources
- The language requirement for mandatory food information is in Article 15 of Regulation (EU) 1169/2011, consolidated text on EUR-Lex. It's a legal provision, so there's no sample: it applies to every food placed on the EU market.
- The average gain of 74% in words per hour, with individual results ranging from 20% to 131%, comes from Mirko Plitt and François Masselot, A Productivity Test of Statistical Machine Translation Post-Editing in a Typical Localisation Context, The Prague Bulletin of Mathematical Linguistics 93, 2010, pages 7-16. It's a test with twelve professional translators, on 144,648 words of software content from a single company, using 2010 technology: we cite it for the order of magnitude and the variation between people, not as a promise about your case.
- The 76%, 40% and 66% figures are from CSA Research, Can’t Read, Won’t Buy. The sample is 8,709 private consumers in 29 countries, so it measures retail buying behaviour, not purchasing departments: the direction is useful, but the number doesn't transfer as it stands to a deal between businesses.
- The 63% of independent translators working on machine-translated text and the 23% rating machine quality as high are in the European Language Industry Survey 2026 report, pages 31-33. The respondents are 1,058 language industry professionals and companies in 45 countries, in other words people who sell translation: it's an interested sample, and we say so.
- The ISO 18587:2017 standard defines the process and the skills required for full human post-editing of machine translation.
- This page doesn't report results achieved for a client, because this piece hasn't yet been delivered to a client. The tests mentioned are functional checks run in a test environment.
The other pieces in this group
Enquiries and contacts lost across channels and languagesFifteen minutes, with your case in front of us.
How many foreign-language enquiries come in each month, and how long before the replies go out? If the answer is that nobody knows, that's already the most useful thing to find out, and it can be counted within thirty days. In fifteen minutes on the phone we'll look at it together and tell you where it makes sense to start, even if we never end up working together.
You'll speak to Mattia Esposito, who then builds the system: there's no salesperson in between. If you'd rather measure things yourself before talking, the Diagnostico (in Italian) is twenty questions and five minutes.