Decide the scope of localization, not just the language list
Localization is not translation. A literal translation of a course written for one regulatory environment, one set of behaviours, and one workplace culture can be accurate word for word and useless in practice. Decide, market by market, how much adaptation you need: transcription only, linguistic adaptation, cultural adaptation of scenarios and examples, or full re-design for a different learning culture.
The answer changes what you are buying and how you evaluate it. If you only need transcription, you are buying a language service with a technical QA component. If you need cultural adaptation, you are buying editorial judgement, and the provider must have people who understand both the language and the working environment of that market.
Also be explicit about what you do not want changed. Many organizations want legal and policy wording preserved exactly, which is a different instruction from a request for a free translation, and it needs to be stated clearly to avoid expensive rework.
Judge linguistic and editorial quality
Ask for a test translation of a real sample from your content, and have a native speaker in the target market review it. The evaluation should focus on whether the result reads as though it was written in that language, whether terminology is used as the industry uses it locally, and whether the register matches your audience.
Pay particular attention to how the provider handles your glossary, if you have one. A mature provider will ask for it, flag ambiguities in it, and propose improvements rather than working around it. A weak provider will translate inconsistently across modules and never mention the discrepancy.
Ask who does the work, whether it is in-house or freelance, and how the provider ensures consistency across a large set of modules. Consistency is a process outcome, so ask about tooling, translation memory, and a term base. See our guide on how to evaluate AI-assisted content production for how machine-assisted translation draft fits into a human-verified workflow.
Assess the cultural and regulatory adaptation capability
For learning content, the highest-cost failures are usually cultural or regulatory, not grammatical. A scenario set in one country may involve assumptions about hierarchy, gender, payment, or authority that do not transfer. A compliance module may reference a regulatory regime that does not apply in the target market, which is both useless and potentially misleading.
Ask how the provider identifies these issues. A good answer describes a localisation questionnaire or brief, a review by a local subject matter or compliance contact, and a report back to you with recommended changes to the source material. A weak answer focuses only on language quality.
For regulated content, ask whether they can work from your approved source and preserve protected terminology and legal wording, and whether they will flag content that cannot be adapted and should be rewritten instead. It is better to hear that during scoping than to discover it in testing.
Verify the technical and engineering side
Localization is also a build problem. Layouts that worked in the source language will break when text expands, buttons will truncate, captions will overflow, and the screen reader order can break entirely. Ask who performs this testing and whether it happens on real target devices.
Ask about captioning and subtitle handling, since these are frequently the weakest part of a localized build. Check whether subtitles are translated and re-timed, or left in the source language, and whether the timing survives expansion.
Also confirm the technical standards: SCORM 1.2, SCORM 2004, and xAPI packaging, and whether a language variant is a separate package or a parameterized version of one. Ask how the provider versions localized content, so that a fix to the source propagates to every language without creating dozens of divergent variants. Our guide on what affects the cost of custom eLearning development explains how languages and layout rework drive the budget.
Plan the ongoing language workflow
The decision that determines long-run cost is whether your content is designed to be updated in multiple languages. If every policy change requires a new translation cycle across every language, the cost of maintenance grows with the number of languages and the frequency of change, which is unsustainable.
Ask whether the provider can work from a single source with modular components, so that only the changed element needs retranslation and re-layout. Whether they use translation memory and a term base, whether they can accept a partial update rather than a full retranslation, and whether they can report how much content actually changed.
Finally, look at the team structure. A single account contact who understands the whole programme is more useful than a translation vendor, a QA vendor, and a build vendor who do not talk to each other.
Ask how the process is coordinated and who owns the overall result. See our guide on how to select a vendor for compliance training for a case where regulatory accuracy and language access have to be solved together.
Commonly overlooked pitfalls
- Buying translation as a language service when the content actually needs cultural and regulatory adaptation.
- Not protecting policy and legal wording explicitly, and paying for rework when the provider rewrites it.
- Ignoring layout, caption, and screen reader testing for text expansion in the target language.
- Letting each language variant diverge into a separate maintenance branch with no shared source of truth.
- Evaluating providers on native-speaker accuracy alone and missing the engineering and QA requirements.
Next steps
- Decide, per market, whether you need transcription, linguistic adaptation, cultural adaptation, or redesign.
- Commission a paid test localization of a real sample and review it with a native speaker in the target market.
- Require a localization brief, a term base, and a back-translation or in-market review step in the workflow.
- Include layout, caption timing, and screen reader testing in the acceptance criteria, tested on target devices.
- Agree a modular update model with a translation memory so a source change does not trigger a full retranslation.
Related guides
Put this into practice
The vendor evaluation checklist and the RFP template turn a guide into a process you can run this week.