Authorised context
Validated sources, non-inference rules, terminology and data boundaries.
Clients
Contact
Governed AI · Terminology · Human review
We combine AI, brand context, terminology and professional oversight to translate faster while retaining editorial control, traceability and the quality appropriate to each use case.
An AI system can produce fluent copy while inventing attributes, changing terminology or diluting brand voice. We configure authorised sources, prompts, glossaries and translation memories, then apply automated checks and human review according to content risk and market exposure.
We configure prompts, terminology, reference material and validated source content before generation, then apply automated checks and human review at the level the content requires.
Tell us about your project →Validated sources, non-inference rules, terminology and data boundaries.
Instructions by brand, channel, language, format and communication goal.
Checks for numbers, omissions, terminology, consistency, tags and variables.
Market specialists validate meaning, naturalness and suitability for use.
The workflow defines what AI may transform, what it must preserve and which claims require validation, with traceability from source to final multilingual version.
Classify content by use, visibility, regulation and commercial risk.
Risk classifiedPrepare authorised sources, glossaries, memories, prompts and QA rules.
Sources approvedRun a representative pilot and have it reviewed by destination-market specialists.
Pilot evaluatedMeasure recurring errors and adjust the system before increasing volume.
Review completedScale with the agreed level of review and retain traceability of decisions.
Feedback retainedThe service can include sample analysis, content and risk classification, source preparation, terminology and prompt configuration, pilot translation, automated QA, agreed human review and traceable delivery.
We can work with TMX memories, TBX glossaries, XLIFF files and CMS or PIM workflows.
AI translation · Ontranslation
For Branni Pets, the starting point was machine-translated ecommerce content for Germany. Cultural analysis, international SEO and native-specialist oversight restored brand coherence and commercial usefulness. The case illustrates the principle behind our workflow: fluency is not the acceptance criterion; meaning, terminology, market fit and intended use are.
Discuss the project →Choose this workflow when multilingual content is recurring, scalable and starts from approved source material. Use post-editing when machine-generated translations already exist. Keep human-led translation or full specialist review for legal, medical, regulated, highly creative or high-exposure content where the cost of a subtle error is unacceptable.
Before production, we agree where generated content may be published, which sources are authoritative and who can approve exceptions. The pilot is assessed against explicit error categories rather than general impressions of fluency. We also separate model configuration from editorial policy: changing a tool must not erase terminology, risk decisions or the evidence behind approved output. This makes the workflow portable, measurable and easier for internal teams to govern.
The system receives authorised reference material and explicit non-inference rules. Product attributes, figures, legal wording and named entities are checked against the source rather than accepted because the generated sentence sounds fluent.
Glossaries and translation memories record approved choices, prohibited variants and market-specific language. These resources are maintained as working knowledge, not uploaded once and forgotten when the content or product changes.
Public, regulated and brand-sensitive content receives stronger human oversight than low-risk internal material. The review model is agreed before production so speed never silently replaces the required level of accountability.
Recurring corrections are classified and fed back into prompts, rules and QA. This creates an auditable improvement loop and helps the client understand which problems originate in the source, the model or the publishing workflow.
No. The service adds authorised context, terminology, governance, QA and professional oversight.
We agree the review level according to risk, visibility, regulation and commercial impact.
Yes. They are essential for consistency and for reusing validated knowledge.
We restrict sources, define non-inference rules and validate attributes, numbers and claims with automated and human controls.
Catalogues, web content, support and recurring documentation can be suitable. Sensitive or regulated content requires stronger controls.
We review a representative pilot, classify recurring errors and, where useful, combine specialist assessment with automated checks or metrics such as COMET.
Tools, access, data processing and retention are agreed before production. We do not assume that every public AI system is acceptable for client material.
Yes. We first assess whether post-editing is efficient. If correcting the output would cost more than translating again, we say so before scaling.
Correct existing machine translation against an agreed quality level and purpose.
View service →Complementary serviceCreate and localise product pages from validated data and category rules.
View service →Complementary serviceGovern Product Truth, approvals, generation and publishing across markets.
View service →Tell us about your project
Tell us about the scope, languages and timeline. We’ll respond with a clear proposal.