// CONTENT & TRANSLATIONS
AI Translation and Machine Localization (MTPE)
AI translation alone delivers speed but rarely reliable quality. A pure agency delivers quality but rarely the speed you need. MTPE closes exactly that gap: neural machine translation for the volume, human-in-the-loop for the polish, ISO 18587 as the quality standard. This is what localization looks like for mid-market companies in 2026.
// DEVELOPING PRODUCTION-READY AI SOLUTIONS FOR
Why you cannot wait three months for your launch
Expanding into five new markets means millions of words to translate. Classic agencies take months. Pure AI is done in seconds but makes mistakes. MTPE delivers both: machine speed, human quality. Three features that make the difference.
01
Up to 95 % time savings
Up to 95 % faster than pure human localization
The AI translates thousands of segments in the time a human takes for one chapter. The post-editor checks, corrects, and polishes. Result: time to market in days instead of months, without quality loss on marketing-critical content.
02
Custom engine per brand
Trained on your terminology
Off-the-shelf tools translate "sneaker" sometimes as one term, sometimes as another. We train our NMT engines on your corporate language and store term bases. The AI must use your defined terms, not its own suggestions.
03
ISO 18587 compliant
Certified post-editing process
Our linguists work to ISO 18587, the international standard for machine translation post-editing. That means defined review criteria, documented quality tiers, traceable sign-offs. For regulated industries and marketing-critical content, this is a baseline requirement.
Our three-stage process for proper localization
/ THE PROCESS
01
Step 01
Pre-processing
The AI analyzes the source text and matches it against the translation memory. Passages translated previously are reused as-is, which creates consistency and saves cost. At the same time the terminology database is loaded so brand-specific terms are set correctly from the start.
02
Step 02
NMT translation
A neural machine translation engine trained on your language and industry creates the raw draft. Unlike standard tools (DeepL Free, Google Translate) we use enterprise instances with custom engine training. This noticeably improves the raw output and reduces post-editing effort.
03
Step 03
Post-editing under ISO 18587
A native-speaker linguist, specialized in your industry, reviews nuance, SEO keywords, and tone of voice. Depending on the service level (Light or Full PE), the work ranges from a sense-check to a stylistic polish. The translation memory is updated with the final version so the next round runs even more efficiently.
04
Step 04
Integration into your system
We do not work with email attachments. Our connectors pull text directly from Akeneo, Pimcore, Shopify, or your CMS, translate it, and push it back into the correct fields. That saves a full process step and avoids formatting breaks.
Service level: which depth fits which content?
Not every text needs the same effort. Internal tickets need pure AI. Marketing headlines need stylistic polish. Three service levels cover the full spectrum and are selected per use case.
Service Level
What happens
Typical Use Cases
Raw MT
Pure AI translation, no human review
Internal chats, support tickets, user reviews
Light PE
Linguist checks for meaning, facts, and critical errors
Technical documentation, FAQs, long-tail product copy
Full PE
Stylistic polish, tone of voice, SEO keywords, transcreation
Marketing claims, homepage, newsletters, brand communication
Data protection and governance
When you translate, you upload sensitive data: contracts, unreleased products, internal strategy documents. Xanevo uses enterprise instances exclusively. Your data does not train public models like DeepL Free or ChatGPT. Server location is in Germany and the EU, ISO 27001 compliant. GDPR-compliant data processing agreements are signed before any project starts.
Special topics for localization professionals
Some topics deserve a deeper look. For professionals who want to fully leverage translation memory and market localization, we have dedicated deep-dive pages.
Translation Memory and consistency
How translation memory works, where its limits lie, and how we compensate for the typical TMS weaknesses (segment context loss, reproduced errors) through hybrid architectures.
// AI TRANSLATIONS
Test our quality before you decide.
500 words from your own material, translated for free in your brand voice. You see the real output, linguistically and stylistically, and make an informed decision. No demo account, no contract level required.