In 2026, translation management is no longer a pure language service. It is an interface discipline between writing, IT, legal, and sales. Treat it that way and you save 25 to 40 % of the effort per language. Keep running it as an order placed with an external provider, and you pay more every year.
In my consulting work I see two camps on the subject of translation. One emails sentences individually to a provider and wonders why costs creep up every year. The other has a translation memory, a terminology database, and an integrated workflow between CCMS and TMS. In both, the volume is similar. The cost difference is typically between 30 and 50 %.
The difference does not arise in the translation itself. It arises in the preparation.
What has fundamentally changed since 2018
Three developments have shifted the field in recent years:
Machine translation has grown up. DeepL, Google Translate, and the NMT engines of the major LSPs deliver, for standard technical texts, raw translations that an editor brings to publication quality in 25 to 35 % less time — compared with pure TM pre-population. The prerequisite is well-maintained terminology. Without terminology, the machine produces consistently inconsistent results.
iiRDS has established itself as an industry standard. The intelligent delivery of information for service portals, apps, and knowledge systems only works with structured, modular content. Anyone who does not move along here can still have their content translated, but can no longer integrate it into their customers’ digital delivery processes.
The Machinery Regulation 2023/1230, binding from 14 January 2027, requires the instructions to be provided in the official language of the market. Anyone delivering in 14 EU languages has to keep the translations audit-proof and consistent. From 2027, a faulty safety-relevant translation is no longer just a marketing problem but a conformity gap.
The five building blocks of a sustainable translation management
1. A translation memory with rigorous maintenance. A TM without maintenance dilutes within 12 months. Concrete maintenance tasks: regular removal of duplicates, updates on terminology changes, versioning per product line. Tools: SDL Trados, MemoQ, Phrase, RWS Translation Manager. A well-maintained TM with 80 % reuse halves the translation costs compared with an unmaintained one.
2. A termbase as the single point of truth. Terms are defined centrally (brands, product terms, safety terminology, specialist expressions) and stored bindingly in every language. Tools: SDL MultiTerm, crossTerm, Phrase Term. Maintained by a named terminology owner, not by the translator.
3. Pre-translation with MT, then post-editing. Instead of fully manual translation, the text is first pre-populated by an MT engine. The translator becomes a post-editor with defined quality levels: light post-editing for in-house documents, full post-editing for publication-ready manuals. Effort drops by 30 to 50 %, provided the engine is trained on your own vocabulary.
4. Quality assurance with a rule engine. Automatic checking for terminology adherence, numerical consistency, and formal completeness. Tools: Verifika, Xbench, the built-in QA in MemoQ and Trados. Finds half of the typical errors before the text reaches the editor.
5. Vendor management with a clear data-processing agreement. Which language service provider gets which data? Where is it processed? Is it used for MT training? These questions have to be answered in writing in the data-processing agreement under Art. 28 GDPR. Engineering drawings and service data are trade secrets, and not every LSP treats them accordingly.
Where the effort really sits
A typical breakdown of effort in a production-heavy mid-sized company with 14 languages, measured across our projects:
- The translation work itself: 35 to 45 % of total costs
- Terminology maintenance: 10 to 15 %
- Editing and post-editing: 25 to 35 %
- Vendor management and project coordination: 8 to 12 %
- Tool licences (CAT, TMS, QA): 5 to 10 %
Anyone who optimises only the translation work (the largest block) has not touched two thirds of the costs at all. Real optimisation takes hold in terminology maintenance, in the post-editing workflow, and in the vendor model.
Common mistakes in practice
From recent consulting engagements, four patterns that come up again and again:
Translations are commissioned sentence by sentence. Every single order runs through quote process, PO, delivery, and acceptance. The administrative effort exceeds the translation costs. Solution: batched orders with defined delivery cycles.
The TM is not versioned. Older translations are overwritten with new brand or product terms. At the next reuse, contradictory text comes back. Solution: TM versioning per product release or per quarter.
The LSP contract contains no data-protection clause. Translator platforms send texts to external MT engines (often in third countries) without the client knowing. Solution: a written data-processing clause with a list of all sub-processors and a ban on MT training on customer data.
Language variants are mixed. Swiss German and standard German in the same manual, or British and American English without a strategy. Solution: define the language variant per market bindingly and separate it in the TM.
What the next stage brings
Three developments that will become relevant in the next 18 to 24 months:
Adaptive MT. Engines learn from post-editor feedback and adapt continuously to the house style. With a clean data basis, the post-editing effort drops a further 10 to 20 % after 6 to 12 months.
iiRDS-based delivery. Translations are no longer delivered as complete documents but as modular topics, assembled in the service portal depending on context. Prerequisite: consistent iiRDS maintenance.
LLM-based quality checks. Language models check, beyond pure rule engines, for semantic consistency, terminology adherence, and style compliance. Here there is still a lot of marketing and little robust practical experience. Anyone who wants to keep pace should set up first pilot projects in 2026 without scaling them into production.
What to consider before you change the translation process
Three questions that have to be clarified in every initial consultation:
How many words per year and per language? Below 100,000 words per year in one language, an in-house TMS often does not pay for itself; in that case a well-chosen LSP with its own terminology maintenance is the better choice.
How high is the reuse rate today? Anyone who does not know this does not know their most important lever. An analysis of the existing TM or a sample of the last twelve months gives a robust figure.
Which regulatory requirements are coming? Machinery Regulation 2027, Product Liability Directive 2024/2853, industry-specific standards (MDR, IVDR, FDA). Each of these requirements has consequences for language variants, audit security, and retention obligations.
A translation strategy without marketing slides
Before any tool selection, any LSP change, and any MT introduction, a sober assessment pays off: current volume, current cost structure, regulatory obligations, technical interfaces. Schübeler Consulting carries out this analysis with you and delivers a concrete roadmap, tailored to your language mix and industry.
Initial consultation: info@schuebeler-consulting.de or via the website.
Johann Jörgen Schübeler, Schübeler Consulting