DeepL BYOK Translation: take your catalog to 30 languages with no hidden markup

Exporting a 50,000-product catalog into five languages traditionally required a translation agency, a dedicated project manager, and six to twelve months of work. With DeepL integrated into the PIM under a BYOK model, the same operation is measured in hours and at marginal cost. But translating a product catalog is not the same as translating a website: there are precise rules to follow in order to preserve quality, consistency, and marketplace compliance.

Why translating a product catalog is different from translating a website

A corporate website contains a few thousand words of marketing content. A PIM catalog combines industrial volumes, sharp technical terminology, and channel-specific editorial constraints. The three key differences:

• Off-the-charts volumes — an average B2B catalog contains between 10,000 and 500,000 translatable strings (titles, short descriptions, long descriptions, attributes, bullet points, SEO tags). Multiplied by five languages, that means several million segments.
• Strict technical terminology — a "chef's knife" is not translated the same way as a knife in a novel. "18V Li-Ion impact drill" must keep millimeter-level precision in every target language, otherwise the professional buyer walks away.
• Cross-product-sheet consistency — the same attribute "Color: Midnight blue" must always be translated the same way on the 12,000 product sheets that use it. A random variation between "Midnight blue", "Dark blue", and "Navy" breaks search filters and site facets.
• Marketplace constraints — Amazon limits titles to 200 characters, forbids certain words ("guaranteed", "best", "100%"), and requires structured bullet points. Mirakl enforces different rules per operator. Translation must respect those limits locale by locale.

Why DeepL rather than Google Translate or a general-purpose LLM

Several machine translation engines exist. DeepL is not the only one, and it has its limits. But for a product catalog exported to European markets, it is currently the best trade-off in terms of quality, format preservation, and compliance — and it is the engine PixeePIM integrates natively under BYOK.

The arguments in its favor:

• Recognized quality on European technical domains — independent benchmarks regularly place DeepL ahead of Google Translate on FR↔DE, FR↔NL, EN↔IT pairs, particularly on technical B2B content.
• Format preservation — HTML tags, placeholders ({product_name}), markup-specific syntax: DeepL respects structures and does not translate them by mistake. Critical for rich product descriptions.
• Custom glossaries — you can impose your own translations on business-specific terms (more below).
• GDPR compliance — DeepL hosts its servers in Europe and offers a Pro / Enterprise tier with no reuse of data for training.
• 31 languages supported at the time of writing, covering the entire European market (FR, DE, IT, ES, NL, PT, PL, SV, DA, FI, etc.), US/UK English, and high-volume e-commerce languages (JA, ZH, KO).

The honest limits: DeepL is weaker than general-purpose LLMs for creative rewriting, deep cultural adaptation (transcreation), and Asian languages outside of Japanese and Chinese. It is a translation engine, not a copywriter.

The BYOK principle applied to translation

BYOK stands for "Bring Your Own Key". You create your own DeepL account (Free, Pro, or Enterprise depending on your volumes), you generate an API key, and you configure it inside PixeePIM. All translation calls are billed directly to your DeepL account, at public rates, with no intermediary.

Concretely, this means:

• No per-word surcharge charged by PixeePIM. You see your usage exactly in the DeepL dashboard — down to the character.
• Full cost transparency. No opaque "translation credits", no hidden publisher margin.
• Free choice of DeepL plan depending on your volumes. A DTC brand exporting a few thousand product sheets can stay on the Free plan and then migrate to Pro. A reseller with 200,000 SKUs goes directly to Enterprise.

The same principle applies to all other AI use cases in PixeePIM (description generation, attributes, AI Tagging, AI Copilot) with 11 supported providers: OpenAI, Anthropic, Mistral, Gemini, Ollama, and more. That adds up to more than 20 providers and services available under BYOK. The related article BYOK AI: why you should bring your own keys details the cost comparison against managed credits.

Typical workflow: translating 50,000 products into 5 languages

Here is the standard flow of a catalog internationalization project inside PixeePIM, once the DeepL key is configured:

1. Configure target languages — select locales (e.g. en_US, de_DE, it_IT, es_ES, nl_NL) and bind them to distribution channels (Amazon UK, Amazon DE, Mirakl Cdiscount, Shopify ES, etc.).
2. Select fields to translate — by default: title, short description, long description, long text attributes, SEO tags, meta titles. Structured fields (weight, dimensions, references) stay in the source language or get a unit conversion applied.
3. Launch the batch — the job can be scheduled outside peak hours to avoid saturating DeepL rate limits. For 50,000 product sheets across 5 languages, plan a few hours of processing.
4. Manual review of critical cases — Amazon product titles, legal notices, health/safety claims, marketing hooks. A human validates, an in-house editor fixes directly inside the PIM if needed.
5. Publish per locale — each validated locale is pushed to its respective channels. Amazon DE gets the German version, Cdiscount FR gets the original, the Shopify ES storefront gets the Spanish version.

Glossary and terminology consistency

This is the feature that turns a correct machine translation into a production-ready catalog translation. DeepL Pro lets you upload a glossary (1,000 entries per language pair), and PixeePIM uses it on every call.

Typical use case:

• A tools reseller sells drills on Amazon DE. Without a glossary, DeepL translates "drill" sometimes as "Bohrmaschine", sometimes as "Bohrer" (which is actually the drill bit). Wrong product, bad ranking, customer returns.
• With a glossary: you force "drill → Bohrmaschine", "drill bit → Bohrer", "auger bit → Bit", "chuck → Bohrfutter". 100% of product sheets become consistent.

The recommended workflow in PixeePIM: export recurring attribute values (your product taxonomy values), generate a CSV glossary from those values (source version + a translation validated by a domain expert or native speaker), upload the glossary into DeepL, re-run the batch on the affected product sheets. The extra cost is zero (using a glossary is free once it exists), and the consistency gain is huge.

Translation and international SEO

A literal translation, even a good one, does not capture local search intent. A German does not search for "Schraubenzieher mit isoliertem Griff" — they type "VDE Schraubendreher". An Italian does not search for "tappeto da yoga" — they search for "tappetino yoga". The exact word matters for Amazon ranking, Google Shopping, and marketplace search.

The recommended pattern in PixeePIM for serious internationalization projects:

• DeepL for the baseline — mass translation, glossary for terminology consistency, format preservation. Covers 80% of content at marginal cost.
• AI enrichment for market-specific SEO variants — on critical product sheets (top sellers, hero products), a second pass through an LLM via the enrichment pipeline, with configurable web search and configurable Shopping API. The LLM rewrites titles and meta descriptions while integrating market-local keywords and the variants actually typed by buyers.

This two-stage flow — mass translation + targeted enrichment — delivers the best results. The article AI Copilot: query your catalog in natural language details the targeted enrichment mechanic, product sheet by product sheet.

Edge cases: when not to translate automatically

Machine translation is never the right answer for everything. Here are the zones where you should explicitly disable it, or send it through human review:

• Legal notices — warranty terms, mandatory legal notices (CE, RoHS, REACH, energy labeling), safety warnings. Legal text is not DeepL's playground: a law firm or a legal translator is still required.
• Product references, SKUs, EANs, GTINs — these fields are never translated. They are identifiers. They must be excluded from the translation scope right at the configuration step.
• Brands and proper nouns— "Samsung" stays "Samsung". "Black & Decker" stays "Black & Decker". DeepL handles recognized proper nouns well, but for less-known brands it is wise to add a no-translate protection rule.
• Units of measurement — "12 inches" translates literally into German as "12 Zoll", but the German market expects "30.48 cm" or "30 cm". Translation alone is not enough — you need a unit conversion, handled separately in the PIM via localization rules.
• Warranties and after-sales — statutory warranty periods, return policies, after-sales commitments vary by country. A literal translation from French to German can be misleading. Should be rewritten by a human.

What does it really cost

DeepL bills per million characters. Public rates (available on deepl.com) start at a few euros per month for the base Pro plan, plus volume-based pricing above the included quota. The Enterprise plan is negotiated case by case for very large volumes.

Estimate for a typical project:

• 50,000 product sheets
• 500 characters of description on average per sheet
• 5 target languages
• = 50,000 × 500 × 5 = 125 million characters

At this volume, you are typically looking at a few hundred euros of DeepL Pro consumption — a marginal cost for a project that would have cost tens of thousands of euros via a translation agency. To compare with your exact case, check the current DeepL pricing grid — it evolves.

On the PixeePIM side: DeepL BYOK translation is available from the Starter plan. No per-word surcharge, no publisher margin — you pay DeepL directly. For the PixeePIM pricing grid, see the Pricing page.

FAQ

Which languages are supported?
The 31 languages covered by DeepL at the time of writing, which covers the entire European market, US/UK English, and the main Asian e-commerce languages (Japanese, Chinese, Korean). The list evolves — check deepl.com.

Can I translate manually after the automatic batch?
Yes. DeepL translations are stored as editable values inside the PIM. Your in-house editors can fix them product sheet by product sheet, locale by locale, without re-running the batch. Manual corrections are never overwritten unless you explicitly request it.

How do I guarantee terminology consistency across the catalog?
By using the glossary feature of DeepL Pro / Enterprise, with a glossary built from your recurring attributes and validated by a domain expert. PixeePIM makes it easy to export attribute values to a CSV glossary format.

What if DeepL is down?
Batches in progress fail gracefully with automatic retry. The PIM keeps working normally — translation is one function among many. Once the service is back, untreated product sheets are picked up automatically.

Is it GDPR-compliant?
Yes for DeepL Pro / Enterprise, which host data in Europe and contractually guarantee no reuse for training. For highly sensitive data (formulas, recipes, contracts), it is still wise to exclude these fields from the automatic translation scope.

Can I mix DeepL with other translation engines?
Yes. The BYOK system lets you configure several providers and assign them to different contexts via Provider Assignments. DeepL for technical translation, an LLM for marketing transcreation, for example.