AI is changing not only translation production, but also the business model of language service providers (LSPs). Customers expect shorter delivery times and competitive prices, while the requirements for quality, terminology and reliability remain unchanged.
This creates a central challenge for LSPs: How can AI translation be offered profitably without increasing quality risks or jeopardising their own margin through uncontrolled review and post-editing efforts?
The answer lies not solely in powerful AI models. What matters is a professional production process that combines translation, quality assurance, human expertise and economic calculation.
This article shows how LSPs can structure AI translation as a service, define quality levels and calculate the actual production effort.
The customer journey begins with the right service
AI translation is not a standardised standalone service. Depending on the content, target audience, risk and desired level of market readiness, the process can range from a controlled raw translation to a fully reviewed publication.
LSPs are therefore advised to define the customer outcome first. Only then should the technology, processing steps and responsibilities be determined. This creates an offering that describes not only a volume of translations, but also a clear quality and review process.
Overview of service components
- Preparation: Review of the source material, target language, terminology, style guidelines and intended use.
- AI translation: Creation of an initial target text taking into account available company and language guidelines.
- Post-editing: Linguistic and content revision in line with an agreed quality target.
- Subject-matter review: Review by a person with appropriate subject-matter or technical expertise.
- Customer approval: Submission for review by the customer and structured processing of feedback.
- Completion and documentation: Finalisation, handover and, where applicable, updating of terminology or style guidelines.
These components do not need to be fully included in every project. What matters is that the LSP and customer define before the start which steps have been commissioned and who remains responsible for each decision.
A clear role model prevents gaps in the workflow
When AI translation is incorporated into existing services, tasks and handovers change. Human work does not automatically disappear; instead, it is distributed differently. A role model makes it clear who safeguards quality, consolidates queries and makes the final decision.
Project management
Project management translates customer requirements into a reliable process. It clarifies the target audience, content type, languages, deadline, quality level and approval process. It also coordinates the handovers between AI translation, post-editing, subject-matter review and the customer.
For recurring assignments, project management should also check whether requirements can be standardised. This includes, for example, consistent briefs, defined escalation paths and transparent documentation of customer feedback.
Post-editing
Post-editors revise the AI-generated text in line with the agreed objective. This involves more than just grammar; it also covers meaning, terminology, readability, tone of voice and compliance with customer-specific requirements.
The scope of editing should not be assumed as a flat rate. Source quality, subject area, language combination, text type and consistency of the requirements all influence the actual effort involved.
Subject-matter review
A subject-matter review is particularly relevant when translations have legal, technical, medical, safety-related or brand-strategic implications. The reviewer assesses the text from a subject-matter perspective and clarifies passages that may be linguistically plausible but critical in terms of content.
The subject-matter review may be the responsibility of the LSP, the customer or a jointly defined party. This responsibility should be explicitly described in the quotation.
Customer approval
The customer often bears responsibility for publication in their own market environment. Their approval should therefore be treated as a defined process step rather than as unplannable rework.
Clear rules for feedback, deadlines and revision rounds are helpful. This makes it possible to distinguish between agreed corrections, new requirements and additional content.
Three quality levels for different types of content
A clear quality framework helps customers select the appropriate scope of services and provides a basis for internal planning and costing.
The following three quality levels are example service models for LSPs and do not constitute an independent standard or certification. The agreed scope of editing, review criteria and responsibilities should be clearly defined and adapted to the language pair, subject area and customer requirements. Relevant normative requirements should be considered separately where applicable.
Level 1: AI translation with defined checks
This level is suitable for content with limited risk where rapid understanding of the content or internal use is the priority. Examples include internal working documents, initial drafts or large volumes with low publication sensitivity.
The process includes a preliminary review, AI translation and a defined plausibility check. A complete linguistic and subject-matter review is not automatically included at this level.
Before commissioning the service, it should be clarified whether the target text will remain internal, which errors are tolerable and whether certain terms must be used consistently. For published or legally relevant content, a higher review level is usually more appropriate.
Level 2: AI translation with post-editing
This level combines AI translation with systematic linguistic revision. It is suitable when the target text needs to be professionally usable while a clearly defined review framework can be agreed at the same time.
Depending on the assignment, post-editing considers meaning, terminology, style, grammar and readability. The scope should be described using specific criteria so that the LSP and the customer share the same quality objective.
For recurring content, it is advisable to make terminology and style guidelines centrally available. ownvia connects AI translation with terminology, corporate knowledge, individual requirements and professional translation workflows.
Level 3: AI translation with post-editing and subject-matter approval
This level is intended for content with a high external impact, increased subject-matter risk or binding publication requirements. Following AI translation and post-editing, an additional subject-matter review is carried out by a suitably qualified person.
Subject-matter approval can focus on the entire text or on defined areas of risk. The key requirements are a documented review assignment, clear responsibility and an agreed approval process.
Even at this level, AI does not replace subject-matter judgement. It can support the process, while content responsibility remains with the people designated for it.
Assessing quality level and content type together
The choice of level should not depend solely on the volume of content. A short contract may carry greater risk than several pages of internal work instructions. Systematic review questions can help with the assessment.
- Is the content used internally or published externally?
- What consequences would a factual, terminological or stylistic deviation have?
- Does the text contain safety-relevant, legal, medical or technical statements?
- How important are brand voice, tone and consistent terminology?
- Who can answer subject-matter questions and grant final approval?
- How frequently do the content, terminology and customer requirements change?
The answers can be used to develop a binding service matrix. This matrix can be used in the quotation, project briefing and quality assurance.
Costing: assess effort transparently instead of applying flat rates
Efficient organisation begins with separating the individual effort drivers. AI translation is only one part of the overall process. For realistic costing, preparation, reviews, communication and any potential correction rounds must also be taken into account.
As a guide, LSPs can consider the following items separately:
- Project effort for analysis, briefing and workflow setup.
- Volume and language combination of the AI translation.
- Effort for post-editing and the agreed quality level.
- Subject-matter review, client coordination and approval management.
- Terminology maintenance, style guidelines and recurring process adjustments.
- Additional iterations resulting from new content or changed requirements.
A quotation can therefore be structured as a combination of base effort, volume-dependent items and quality-dependent review services. The appropriate billing model in each case depends on the project model and the agreement with the client.
Blanket savings promises should be avoided. The economic benefit depends, among other things, on the source material, language pair, degree of repetition, review intensity and process maturity. A reliable assessment can only be made once these factors have been evaluated in a pilot project or using representative content.
Categorising enterprise packages by volume and rollout
For licence planning, individual projects are not the only factor that matters. The expected annual volume, the number of users involved, the number of languages and whether an initially limited pilot or an organisation-wide deployment is planned are all decisive.
Enterprise 10K: Regular production use
Enterprise 10K is designed for companies and LSPs with a regular professional translation requirement. Up to 10,000 activations per year provide a suitable framework when AI translation is already being used in recurring customer or business processes.
The package is particularly suitable when terminology, style guidelines, company knowledge and professional translation workflows are to be integrated into the process as binding requirements. It therefore provides guidance for productive processes with a predictable, medium volume.
Enterprise 100K: Centrally scaling larger volumes
Enterprise 100K is aimed at companies, language departments and LSPs with a high, continuous translation requirement. Up to 100,000 activations per year provide a framework for larger translation volumes and broader use across teams or business units.
This package is suitable when AI translation is to be organised not only in individual projects, but as a central part of professional translation processes. Terminology, style guidelines and company knowledge remain important control elements.
Enterprise Custom: Taking individual requirements into account
Enterprise Custom is intended for particularly high volumes, organisation-wide rollouts or use cases that go beyond standardised packages. The scope, activation volume and framework conditions are agreed to suit the respective use case.
This option may be suitable when multiple organisational units, extensive language portfolios or specific process requirements come together. The appropriate configuration should be based on a concrete analysis of the planned rollout.
From an idea to a robust pilot project
A pilot project should test more than just the technical translation quality. It should map the complete customer process and demonstrate how roles, reviews, approvals and effort work in practice.
- Select representative content: Use material that reflects typical text types, specialist terminology and quality requirements.
- Define the quality level: Document which checks are included and what result is expected.
- Define roles: Determine who is responsible for project management, post-editing, specialist review and client approval.
- Gather requirements: Record terminology, style, company knowledge and sensitive content.
- Monitor effort: Measure not only the volume of translation, but also coordination, review and correction cycles.
- Make a decision: Translate the findings into a standardised offering and a suitable licensing framework.
Such a pilot provides a better basis for decision-making than a general assessment of AI. It shows which content is suitable, where human review remains necessary and which process requirements are still missing for the client's operations.
The next step for your LSP
AI translation becomes economically manageable for language service providers when performance, quality, roles and volume are considered together. A clear workflow protects against unclear expectations and makes collaboration with clients transparent.
ownvia supports companies and LSPs in combining AI translation with terminology, company knowledge, individual requirements and professional translation processes. Book a personal ownvia demo or arrange an initial, no-obligation consultation, to assess your translation process and a suitable starting point.