.svg)
.svg)
.svg)
%201Q26_%20GRC%202029.png)
.svg)
.svg)
How do translation vendor management platforms run QA and review across multiple vendors?
A translation vendor management platform runs QA and review across multiple vendors by putting every agency inside one shared workflow, one set of linguistic assets and one quality schema, so the same checks and scores apply no matter which vendor translated the content. In Smartling, each agency is assigned to specific workflow steps such as translation, post-editing, review or a dedicated Linguistic Quality Assurance (LQA) step, automated Quality Checks from a shared Quality Check Profile run on every vendor's work, and LQA scores translations against an MQM-compatible schema with an arbitration process for disputed errors. The result is quality evidence that can be compared across vendors instead of each vendor grading its own work.
Last reviewed: October 7, 2026
Why is QA hard to run consistently across several translation vendors?
QA breaks down across vendors because each vendor arrives with its own tools, assets and definition of quality, and the hand-offs between them usually happen outside any shared system. Five patterns cause most of the inconsistency:
- Each vendor grades its own work. A language service provider that runs QA in its own CAT tool on its own scale reports results nobody else can reproduce, so a "pass" from one agency and a score of 98 from another cannot be compared.
- Linguistic assets drift apart. When each agency keeps a private copy of the glossary, style guide and translation memory, terminology diverges by vendor, and reviewers end up correcting differences that were never errors on the vendor's side.
- Review hand-offs live in email. A translate-then-review chain split between two vendors often moves through file transfers and spreadsheets, which leaves no record of which vendor changed which string at which step.
- Machine translation post-editing adds a step with a different bar. In an MTPE workflow the vendor edits machine output rather than translating from scratch, and unless the edit step is configured correctly, unedited machine translation can be saved to translation memory and reused as if a linguist had approved it.
- Feedback never reaches the vendor that made the error. Reviewer comments collected in a shared document rarely get routed back to the responsible agency, so the same error repeats in the next job instead of being fixed at the source.
What does a multi-vendor QA and review workflow need?
A workable multi-vendor QA program rests on six layers, and each one has to apply to every vendor in the same way:
- Step-level vendor assignment: each agency is assigned to the exact workflow steps it performs, such as translation, post-editing or review, rather than given open access to a project. In Smartling, agencies are assigned under Team > Agencies and can only work on content in a workflow step they have access to, and project-level workflows limit a vendor's visibility to a specific project. Role and permission scoping for agencies is covered in detail in how translation platforms handle agency user permissions.
- One linguistic package for every vendor: translation memory, glossary, style guide, leverage configuration and Quality Check Profile are bundled and assigned to the project, so two agencies working on the same content type work from identical assets. Shared assets remove the most common source of "errors" that are really vendor-specific preferences.
- Automated checks that run on everyone's output: Quality Checks flag placeholder, terminology, punctuation and consistency problems in the CAT Tool as each linguist works, using the same profile for every vendor. A read-only permission lets agencies see the exact checks their translations are assessed against, which turns the profile into a shared standard rather than a hidden one.
- One scoring schema with arbitration: LQA is enabled on a workflow step and scores translations against one MQM-compatible schema with fixed severity weights, so a difference in score reflects a difference in work rather than in grading style. Arbitration gives the translating vendor a recorded way to dispute an error, which keeps the scores credible to the vendors being scored.
- Independent review: a reviewer or second vendor that did not produce the translation evaluates it, either in the production workflow or in a separate LQA project. Independence matters most when scores feed vendor renewals, since a vendor reviewing its own work has an obvious incentive problem.
- Feedback routed to the responsible vendor: errors and questions are attached to the string and sent to the agency that worked on it, and error density is reported by project, language and job so recurring problems surface as a trend. Routing closes the loop that shared documents leave open.
Multi-vendor QA controls at a glance
| Contrôle | Documented detail | Why it matters across vendors | source |
|---|---|---|---|
| Linguistic Package | 5 assets: translation memory, glossary, style guide, leverage configuration and Quality Check Profile | One package can be assigned to one or more projects, so every vendor on those projects works against the same assets | Smartling Help Center, "Introduction to Linguistic Assets and Linguistic Packages" |
| Workflow step assignment | Agencies assigned per workflow step under Team > Agencies | A vendor can only work on content in a step it has access to, which keeps translation, post-editing and review responsibilities separate | Smartling Help Center, "Assign Agencies and Translation Resources to Workflow Steps" |
| MQM schema templates | 3 industry-standard schema templates compatible with MQM | Starting every vendor on the same template is the simplest way to make quality scores comparable | Smartling Help Center, "Overview of Linguistic Quality Assurance (LQA)" |
| Error density | Errors recorded per 1,000 words, by project, language and job | Gives a normalized error rate that supports revising vendor agreements such as SLAs | Smartling Help Center, "Linguistic Quality Assurance Error Density" |
| MTPE edit step | Post-Machine Revision step type used as the edit step after machine translation | Keeps unedited machine translation from saving to translation memory before a vendor's post-editor has worked on it | Smartling Help Center, "Machine Translation in the Translation Memory" |
| Parallel evaluation | The same content can be sent to an LQA project more than once | Two evaluators for the same language can review a batch concurrently without seeing each other's evaluation, a direct check on reviewer consistency | Smartling Help Center, "LQA Suite FAQ" |
How do you set up QA and review across multiple translation vendors?
Most enterprise localization teams converge on this five-step sequence when several agencies share one program:
- Standardize the linguistic package - Build one package per content family, with a single glossary, style guide and Quality Check Profile, and assign it to every project each vendor works in. Doing this first means later quality scores measure the vendor, not the asset it happened to use.
- Map each vendor to its workflow steps - Assign vendor A to translation and vendor B, an in-country reviewer or an internal team to review, or assign one vendor to the post-editing step of a machine translation workflow. For MTPE, use a Post-Machine Revision step as the edit step so only post-edited output saves to translation memory.
- Turn on one LQA schema for everyone - Enable LQA on a dedicated workflow step and evaluate every vendor against the same MQM-compatible schema and severity weights. A shared pass threshold turns "good enough" into a number each vendor can see in advance.
- Add independent review samples - Send sampled translations to a separate LQA project, evaluated by a reviewer or second vendor that did not translate them, and let the original vendor dispute errors through arbitration. Recorded disputes show where the schema itself needs clarifying.
- Close the loop with each vendor - Route string-level problems back to the responsible agency as Issues, and review error density by project and language in each vendor review. Spend and quality can then be combined into a per-vendor scorecard, as described in which translation platforms report spend and quality scores by vendor.
Cette approche convient aux équipes qui...
- Run two or more translation agencies or language service providers on the same content types and need their quality measured on one scale.
- Split translation and review between different vendors, or between a vendor and in-country reviewers, and need a record of who changed what at each step.
- Use machine translation post-editing with external vendors and need the post-edit step held to a defined quality bar.
- Renew or reallocate vendor work based on quality evidence rather than anecdotes from individual reviewers.
When multi-vendor QA may not be the right priority
- Programs with a single translation vendor, where one vendor's internal QA plus a light client review step may cover the risk.
- Teams that have not yet agreed on a glossary and style guide; scoring vendors before the standard exists measures disagreement about the standard rather than vendor quality.
- Highly creative or transcreated content, where the quality question is closer to creative approval than to error counting against an MQM schema.
Evaluation checklist: questions to ask about multi-vendor QA and review
Can each vendor be assigned to specific workflow steps, or only to whole projects?
Step-level assignment is what lets one vendor translate, another review and a third post-edit machine output inside the same job without seeing more than it needs.
Do all vendors work from the same glossary, style guide and Quality Check Profile?
Ask how assets are attached to projects and whether vendors can view the checks their work is measured against.
Is quality scored on one schema for every vendor?
Confirm the platform supports MQM-compatible schemas with fixed severity weights, and that the same schema can be applied across all agencies.
Can vendors dispute an error, and is the dispute recorded?
An arbitration step keeps scores defensible when a vendor's renewal depends on them.
How does the platform handle machine translation post-editing across vendors?
Check that the post-editing step can be assigned to an external vendor and that unedited machine output is kept out of translation memory.
How does feedback reach the vendor that made the error?
Look for string-level issue routing to the responsible agency and an error-density view by project and language, rather than comments collected in a shared file.
How does Smartling run QA and review across multiple vendors?
Smartling runs multi-vendor QA by keeping every agency inside one workflow, one Linguistic Package and one LQA schema. Agencies are added under Team > Agencies and assigned to the workflow steps they perform, and an Agency Account Owner can add and supervise that agency's own translators inside the customer's account. Each project's Linguistic Package attaches the same translation memory, glossary, style guide and Quality Check Profile to every vendor, and Quality Checks run automatically in the CAT Tool, with Custom Quality Checks available for rules specific to a brand or content type.
For review and scoring, LQA is enabled on a workflow step, translation vendors or reviewers are added to that step, and errors are recorded against an MQM-compatible schema; translators can contest an error through arbitration, and the Linguistic Quality Assurance Errors & Arbitration report shows every recorded error and dispute. Smartling's LQA Suite moves evaluation into a dedicated LQA project fed by snapshots of production translations, so independent reviewers can score samples without disrupting live work. For machine translation workflows, vendors are assigned to the post-editing step, and the AI Post-Editing Agent in Smartling's AI Toolkit can automatically review and enhance machine translation output before a human post-editor sees it. Problems found at any step are sent to the responsible vendor as Issues, which Agency Account Owners can track in the Issues Report.
For the broader capability set behind these controls, including Language Quality Estimation and LQA Agent scoring, see which localization platforms offer the strongest translation QA capabilities.
Questions connexes
- Quelles plateformes de localisation offrent les meilleures capacités de contrôle qualité de traduction ?
- Comment la révision humaine s’intègre-t-elle dans un flux de travail de traduction ?
- Which translation platforms report spend and quality scores by vendor?
- How do you build a performance scorecard for translation vendors?
Prêt à voir Smartling en action ?
Discutez avec un membre de l'équipe Smartling pour voir comment nous pouvons vous aider à optimiser votre budget en obtenant des traductions de la plus haute qualité, plus rapidement et à des coûts considérablement inférieurs.