Ecommerce Localization: How to Translate Product Catalogues, Online Shops and Product Data at Scale

Ecommerce localization works at scale when you translate product data, not pages: export it from your PIM or shop system, separate specifications from descriptions, lock units and codes, build a product glossary, and let translation memory reuse every repeated attribute. Machine translation with post-editing suits large ranges; claims and safety texts need people. DEMA Solutions translates catalogues, shops and product data this way.

Parcels delivered on a porch

Who this guide is for

This guide is for the people who own product content in companies that sell across Europe: e-commerce managers, product information and master data managers, and marketing managers at manufacturers and brands that run an online shop, feed marketplaces and distributors, and still print a catalogue every season.

Their questions are practical. What needs translating, and in which format? Can machine translation handle ten thousand descriptions? How do we stop the shop, the catalogue and the packaging from using three different words for the same thing? What must we legally provide in each language? And how do we keep it current when products change every week?

What ecommerce localization covers

Ecommerce localization adapts everything a customer in another market reads before, during and after a purchase: product names and descriptions, attributes and filters, size guides, delivery, payment and returns information, transactional emails, checkout microcopy, legal pages and the safety information that must accompany the product. It also adapts formats: decimal separators, dates, currencies, address forms and the size systems customers expect.

For a manufacturer or brand, the shop is rarely the only channel: the same product data feeds marketplaces, distributor portals, data sheets and the printed catalogue. Translate the product content once and publish it everywhere, and ecommerce website translation, catalogue translation and marketplace listings become outputs of one controlled data set. Budgeting a multilingual website in general is covered in our guide to website translation cost; this article focuses on what makes product content different.

Start with the data: PIM, CSV and XML exports

Large product ranges should never be translated by copying text from a live shop into a spreadsheet. The content already lives in a structured system, and the translation should travel in and out of it in the same structure.

Most PIM systems, ERPs and shop platforms export product content as CSV or Excel files, as XML, or as XLIFF through a translation connector. The principle is always the same: one record per product or variant, one field per attribute, a stable product identifier on every row, and new columns or elements for the target languages. The identifier is what lets your team import the translations back into exactly the right products.

A good export contains only the fields that need translating and marks clearly those that must never change: article numbers, EAN or GTIN codes, internal keys, image URLs, prices and numeric values. Fields with HTML or placeholders are flagged so tags stay protected, and discontinued products and duplicates are removed first, because otherwise they are counted and paid for. Before the full project, run a trial export and import of twenty or thirty products, check the result in the shop, and agree the file encoding, which avoids accented characters arriving broken in a French or German shop.

Cardboard boxes and paper shopping bags

Attributes, units and sizes

Attributes look easy and cause the most errors. Colour, material, finish, capacity, “compatible with”: each is a short value, often one word, that appears on thousands of products and drives the filters customers use.

Short values have no context. “Natural” can be a colour, a material or a finish, and “black” may need a different form depending on the gender of the noun it describes in Italian, Spanish or German. The safest approach is to translate attribute values as a controlled list, once, with the attribute name next to each value, and then apply the approved translation to every product automatically. Thousands of strings become a few hundred decisions.

Units and numbers need a firm rule. Dimensions, weights, capacities, power ratings and temperatures must never be converted during translation unless you have decided to do so for a market. The decimal separator does change by locale, and consistently: 1.5 kg in an English shop becomes 1,5 kg in Italian, German and Spanish, and a thousands separator must never be mistaken for a decimal one. A dimension that silently becomes ten times larger is a returned order, or a product that does not fit.

Sizes are their own problem. Clothing, footwear and some technical products use size systems that differ between markets, so decide with your product team whether you show the original system, a converted one or both, and treat the size guide as controlled content.

Product descriptions versus specifications

A product page combines two very different kinds of text, and treating them alike wastes money in one place and risks quality in another.

Specifications are factual, repetitive and structured: technical data, materials, dimensions, compatibility, care instructions, contents of the box. They must be exactly right and identical wherever they appear, and they reuse the same phrases across a whole range. That makes them ideal for translation memory, a strict glossary and, on large volumes, machine translation with post-editing. The goal is accuracy and consistency, not style.

Descriptions sell. Short and long descriptions, benefit lines, category introductions and collection copy are written to persuade a particular buyer, and they depend on tone and on what matters to customers in that market. Product description translation needs native linguists who write well, and the most visible lines, such as headlines and claims, may need transcreation. Translating marketing materials such as brochures, launch emails and campaign pages belongs in the same creative layer. What that creative layer involves, and how it differs from an accurate translation, is explained in our article on transcreation vs translation.

Separating the two in the export and the quote puts the budget where it changes results.

Repetition and translation memory leverage

Product content is among the most repetitive text a company produces. The same attribute values, safety sentences and care instructions appear on every product in a family, and a new season’s catalogue reuses most of the previous one.

A translation memory stores every approved source and target segment and recognises it the next time it appears. Repetitions within an export and matches with earlier translations are reused, and similar sentences are offered to the translator as a starting point, so a variant only needs its differences translated, and a quarterly update may contain little more than the new products and the changed fields.

That leverage benefits you only if the memory belongs to you, in a portable format, and if your provider reports how much of each batch was reused. The reasons are set out in our article on who owns your translation memory, which matters most when you change supplier or add a second one.

A product glossary before the first batch

The most expensive inconsistency in product content is not a wrong word. It is three right words for the same thing, one in the shop, one in the catalogue and one on the packaging, translated at different times by different people. Filters break and customers hesitate.

A product glossary prevents this. It lists the terms that matter in every language: product line names and whether they stay in the source language, components, materials, technical terms, attribute values and words you never want used. It is approved once by your product and marketing teams and the market owners, then applied in every channel. Product names and category labels should also use the words local buyers actually type into a search box, which is not always the literal translation. Building a glossary and keeping it alive as the range grows is explained in our guide to keeping terminology consistent across translators and suppliers.

Machine translation with post-editing for ecommerce localization at scale

For tens of thousands of records, machine translation with human post-editing is often the right balance of speed and cost, if it is used on the right content and set up properly.

It suits specifications, attribute-heavy text and long technical descriptions, where language is predictable and the glossary carries most of the risk. Engines such as DeepL produce fluent output for many language pairs, but fluency is not accuracy: an engine that does not know your terminology will translate a product line name, pick a word your buyers never use or alter a unit. The output is only as good as the glossary and memory that steer it and the linguist who corrects it.

It is less suitable for short persuasive copy, and it is not the route for safety warnings or legal texts, which need full human translation and revision. A sound plan maps each field in the export to a treatment: post-editing for specifications, human translation for descriptions and safety texts, transcreation for headline copy. How such a setup works is described on our machine translation and post-editing service page, while the question of when post-editing is good enough has its own article.

Images with text and printed catalogues

Feature callouts, dimension drawings, infographics and banners with slogans are invisible to a text export and often missing from the first quote. List them early, ask your design team for editable source files, and decide which images need a language version and which can become text-free.

Printed catalogues, price lists, brochures and flyers are a different production. They are built in a layout application, often automatically from the PIM, and translated text is frequently longer than the source: German compounds and Italian or Spanish phrasing can push a product name or table cell beyond its space. Expansion has to be solved in the original layout files, with the same styles and page structure, not by shrinking type until it is unreadable, and cross-references, indexes and article numbers must stay correct.

That is the work of multilingual desktop publishing: our in-house DTP team works in your original design files and returns print-ready documents and PDFs. Anyone comparing catalog translation services should ask whether layout is included or whether translated text will simply be handed back for someone else to place. The steps of a combined translation and layout project are described in how multilingual DTP supports document translation.

Legal product information and safety texts

Some product content is information the law requires, in the right language for every market where the product is sold.

The General Product Safety Regulation (EU) 2023/988, which has applied since 13 December 2024, requires consumer products to be accompanied by instructions and safety information in a language consumers can easily understand, as determined by the Member State where the product is made available. Sellers offering products online must also show clearly in the listing the manufacturer’s name and contact details, information identifying the product including a picture, and any warning or safety information that must accompany the product, in a language consumers can easily understand. For many shops, safety texts now have to appear on the product page, not only in the box.

Sector rules add to this. Textile products need their fibre composition in the official language or languages of the Member State where they are sold. Products under harmonised EU legislation, such as machinery, electrical equipment or toys, carry instructions with their own language rules, and consumer information on contracts, returns and warranties is subject to national language requirements in several countries. Have your compliance or legal team confirm which texts each market requires, then treat them as controlled content: translated by specialists, revised by a second linguist, approved once and reused unchanged.

Keeping the shop, the catalogue and the updates in sync

The first translation of a range is a project; everything after it is a process. Products launch in the source language first, specifications change, and the language versions fall behind until the German shop lists a material the product no longer uses.

Three habits prevent it. Make the PIM or shop system the single source of truth, and let a change to a source field flag the related language fields as outdated. Send updates in regular batches, weekly or monthly, rather than product by product by email. And export only new and changed fields, with unchanged text already matched against the memory, so you never pay again for a product because one attribute changed. Printed catalogues follow the same logic on a longer cycle: if the catalogue is built from the PIM, translated data flows straight into the new edition.

Quality checks before products go live

Product content has a specific error profile, so the checks before publication should target it. Beyond revision by a second qualified linguist, these catch the problems customers notice:

  • Numbers and units: every value, unit and decimal separator compared with the source, with no unagreed conversions.
  • Codes and identifiers: article numbers, EAN or GTIN codes, model names and protected fields unchanged.
  • Terminology: glossary terms used consistently across shop, catalogue, packaging and attribute lists.
  • Length and display: product names, filter labels and buttons that fit in the shop, the app and the catalogue.
  • Tags and placeholders: HTML, variables and line breaks intact after import.
  • Safety and legal texts: the approved wording, complete, in the required language for each market.
  • In-context review: a sample of products checked in the staging shop and the catalogue proof by a native reviewer.

What drives the cost of catalogue and e-commerce translation

Product content is priced by the word and by the type of treatment, but a few factors decide most of the total. The general mechanics are explained in our article on what drives technical documentation translation cost; for product data, five matter most.

The real word count. The export, not the number of pages, defines the volume. Removing discontinued products, duplicates and fields that never change is the cheapest saving available.

Repetition and memory leverage. Repeated attributes, safety notes and matches with earlier translations should be counted separately from new words. Two quotes for the same export can differ widely simply because one provider counted repetitions and the other did not, so ask for an analysis that shows the categories.

The treatment per field. Post-edited specifications, human-translated descriptions and transcreated campaign lines are different services at different price levels, and mapping each field correctly is the biggest lever on the budget.

Layout and images. Catalogues and images with text add DTP work that depends on how cleanly the source files are built, not on word count.

Update rhythm and number of languages. Cost grows almost in line with the number of languages, so prioritise the markets that bring revenue. A steady flow of small batches with high leverage costs far less over a year than occasional re-translations of a range that has drifted.

A practical sequence for your first market

Companies that do this well follow a similar order. They start with an inventory of every place product text lives, from PIM fields and shop templates to emails, images, packaging and the catalogue, and clean the data before exporting it. Next, the glossary and attribute values are approved before the first large batch, because every later change touches thousands of records. A pilot follows: a representative set of products exported, translated, imported and reviewed in the shop and a catalogue proof, while problems are cheap to fix. Then the full range goes through, each field in its agreed treatment, and the update process is agreed, so the second market starts from an asset rather than from zero.

When you compare providers, ask them to walk through that sequence for your data: how they handle your export and protected fields, how they report memory matches, whether you own the memory and glossary, who revises and who checks in context, and whether layout is included. Our website translation services apply the same approach to online shops and product pages.

How DEMA Solutions handles catalogue and e-commerce translation

We work from your exports, in the format your PIM, ERP or shop system produces, such as CSV, Excel, XML or XLIFF, and deliver files ready to import. Native linguists for each market translate descriptions and adapt tone, while specifications follow your glossary and memory. Headlines, collection copy and claims can be transcreated as part of our marketing translation and transcreation work, and catalogues, brochures and price lists are laid out by our in-house DTP team in your original design files.

The process is ISO 17100 certified, with a second qualified linguist revising every translation, as described on our quality page. Your translation memory and glossary are built from the first batch, belong to you and are delivered in portable formats (TMX and TBX), and transparent leverage reporting shows how much of each update is reused. Our teams in Italy, Serbia and Spain work with linguists in every market you sell to.

Where TUBE fits

For brands whose product range changes every week, TUBE keeps each update under your control. Your glossary and translation memory control machine translation, every approved translation goes into a memory you own and can export, and leverage reporting shows how much of each batch comes from text you have already approved.

Frequently asked questions

The adaptation of everything a customer reads in an online shop for another market: product content, filters, size guides, checkout texts, legal pages and safety information, plus formats such as decimals, dates and currencies. For manufacturers it usually includes the product data that also feeds marketplaces and catalogues.

Export them with a stable product identifier, protect fields that must not change, prepare a glossary and use translation memory so repeated text is translated once. Specifications can often be post-edited; persuasive descriptions and safety texts need human translation and revision.

For specifications and attribute-heavy text, machine translation with full post-editing by a qualified linguist is often suitable, especially when the engine follows your glossary. For headlines and claims it rarely is, and for safety and legal texts it is not recommended.

The format your system exports and imports cleanly: CSV or Excel for most PIM and shop exports, XML for structured feeds, XLIFF where a translation connector exists. Keep a unique product identifier on every record and agree on encoding before the first batch.

Yes. Under the General Product Safety Regulation (EU) 2023/988, instructions and safety information must be in a language consumers can easily understand, as determined by each Member State, and online listings must show warnings and safety information in that language. Sector rules, such as textile labelling, add their own requirements.

Not automatically. Keep specification values exactly as in the source and apply only the local number format, such as the decimal comma. Converting sizes or units is a product decision for each market, based on tables approved by your product team.

Make the PIM or shop system the single source, flag changed source fields automatically, send regular batches with only new and changed content, and reuse the translation memory so unchanged text is not paid for again.

Get your product range ready for new markets

If you are preparing an online shop, a marketplace feed or a printed catalogue for new markets, send us a sample export and tell us which languages you need. We will analyse the real word count and the repetitions in your data, and propose the right treatment for specifications, descriptions, safety texts and layout.

How can we help?

Tell us what you need and we’ll get back to you with the solutions.

Table of Contents