Source
Approved content and assets. Structured product files, website copy, campaign content or everyday business text enter the same system, and marketplace content keeps its required structure from input to export.
Agentic Localisation System
Product and marketplace content moved across languages, brands and local teams through agencies, spreadsheets and email. We put in place one localisation system that brings translation, terminology, market review and accepted corrections together.
The source content was rarely the problem. The difficulty was keeping approved terminology, changing brand guidance, market requirements and local decisions consistent across every request.
Agency loops added time and cost.
Terminology and market corrections lived across spreadsheets, email and individual reviewers.
Market teams often saw content only after significant work had already happened.
Every request starts with the source content and the knowledge relevant to that brand and market.
Approved content and assets. Structured product files, website copy, campaign content or everyday business text enter the same system, and marketplace content keeps its required structure from input to export.
Terminology, guidance and previous decisions. Each brand has its own knowledge base of approved terms, language guidelines, policy documents and accepted translations. Switch brands and the system adopts that brand’s terminology, tone and rules before it drafts anything.
The draft is created with the relevant context already attached: the brand’s terminology, its guidance and what each market has accepted before. The model translates inside that context, not the source in isolation.
Local teams review the translation where it will actually appear, correct it, approve it, and feed useful decisions back into the system. A direct link, no account needed, whether the reviewer is from sales, content or ecommerce. Approved content returns to the marketplace file or export it came from.
The model works with approved terminology, brand guidance and market context rather than translating the source in isolation.
Approved terminology and reviewer decisions become reusable context for future requests.
Public content can use external models where appropriate. Sensitive internal material can use private or local model options when required.
We ran 48 client product pages across seven languages using the same model with the glossary switched off and on.
The results also show where local market judgement still matters. Accepted corrections become useful context, while evaluation continues as models, terminology and market requirements change.
One path from draft to market review.
Accepted terminology, reviewer corrections and market decisions become part of the knowledge available to the next request. Evaluation continues as models, terminology and brand requirements change.
The tool translates. Markets decide. Every approved correction makes the next draft better.
If translation itself is not the bottleneck, we can look at where terminology, context, review or market knowledge is getting lost and whether an intelligent system would help.