Ralf Lemster opened the second Bitesize Session of the 2026 Bitesize series, welcoming Luis Lopes (RWS) and Sara Grizzo and Andreas Merz (from Munich Re’s Language Data and Technology Team). A small poll of attendees showed that the move towards Trados Enterprise is still a future consideration for many ETUG members, and that Trados Studio remains the most popular version of Trados among those attending.
RWS’ Cloud Offering
Luis Lopes, who had spoken at the first Bitesize session of the season, provided an introductory presentation about Trados’ cloud platform. He remarked that many users were probably using GroupShare (ie server-based) still. He provided an overview of the different offerings.
He also addressed RWS’ partnership with Cohere – which has a particular bearing on the AI orchestration front of Trados, which is one particular reason for moving towards a cloud-based setup. Trados was initially predominantly associated with local (desktop and server-based) setup, and its product landscape was dominated by a lot of different translation memory products due to various acquisitions. The overlap meant a lot of overlapping products, and since 2018 they have been moving towards a platform that takes the best features from the competing predecessor products. Now the platform serves all of RWS’ customer segments, and uses just one technology, rather than multiple competing technologies.
By moving to a platform-based setup, it has also made it easier to deliver innovations faster. Historically the problem when presenting about the future, was that many users were one or two versions behind due to the very long upgrade cycles. This was often due to testing new roll-outs with IT. The lag was often one of 2-3 years. The cloud-based platform gets around this, as updates are available to users as soon as a release goes live and the customer gets the notification about this. Customers are now at the cutting edge benefitting from the innovations and AI immediately. At the same time, Trados’ linguist-centric culture, philosophy and heritage has been preserved.
However, moving to an open platform, rather than a closed system, is now a very important aspect. Trados’ App Store still works in the cloud environment. Trados was historically a translation management space or language technology platform, but is now becoming a leader in the AI translation execution space. The platform offers different offerings for different types of customers (from freelancer, language service providers and agencies, and then corporate and enterprise customers), as well as public sector and institutional settings, whose needs are frequently very specialised.
The cloud offering follows a similar model to Office 365, so you can work either in the Desktop or via the browser, depending on your use case. Trados Team, another offering is effectively like a cloud-based version of GroupShare in terms of its features. The cloud approach is also very much focused on teams collaborating, working together, sharing linguistic resources and working with internal teams, but with the possibility to include external terms. Trados Enterprise is very much a custom solution, where typically the customer works with RWS’s sales and pre-sales team, through solutions consultants that look at your organisation and recommend which features are required.
The platform has three key focuses:
- Translation productivity (i.e. in Trados Studio, Trados Online Editor and Terminology) while ensuring quality.
- Collaboration in localisation (i.e. for the project management side and involving your stakeholders in the process) to ensure less manual effort. Here the idea is to stop old workflows with bilingual table formats or PDFs, but to get these steps into the Online Editor, with a layout preview that allows you to edit it as if you were say editing a Word document, and therefore cutting out the manual editing/revision steps (and overcoming segmentation shock). Similarly with vendor management it is about allowing you to work with and involve vendors.
- Translation management has increasingly become about orchestration – the workflows, and now in the AI era about orchestrating AI. Corporations have been quick to embrace AI, but struggle with its orchestration in terms of the guardrails and the control. Here the reporting helps with quality management.
Customers often also have their own ecosystems that they like to connect to – and so there are also connectors to their ecosystems using APIs, apps, security solutions, to ensure that translation is handled internally.
Packages for the cloud offering
Various packages are available for Trados Cloud, starting with the smallest Team (3 concurrent users, 1 premium editor and 4 million words included), followed by Business (5 concurrent users, 5 premium editors, 8 million words) and Enterprise (10 concurrent users, 5 premium editors and 30 million words). All of these have AI features like Smart Review, AI terminology extraction, Smart Insights and Generative Translation (which requires an AI engine). AI features are included by default out-of-the-box as in 2026 there is no sense in offering a language technology platform without AI. Previously this required separate add-ons.
Generative translation, which requires an AI engine, means that you underpin it with an LLM, and in this regard RWS has partnered with Cohere, a Canadian LLM provider, which recently acquired the German AI company Aleph Alpha. Trados’ Machine Translation benefits were previously due to their security, scalability and on-premise nature, although some competitors led in terms of quality for localisation.
Options for translation and AI:
Regarding translation and AI there are three basic options that companies can take, each with different manifestations:
- Connecting to foundation providers (e.g. OpenAI, Gemini, Microsoft) where you have no power over data handling, model changes and very general purpose models (so no control, and risks of cost explosion/pricing where complex prompts are needed to force correct behaviours).
- Building on Open Source – quality lags behind the frontier models, and can be a black box and there is no accountability (e.g. hallucination and harmful outputs). Open Source approaches can end up with needing different models for different language combinations, therefore increasing hardware costs.
- Working with a Partner (e.g. RWS/Trados x Cohere) permits greater control over model development, combination of the frontier AI provider and industry MT/localisation expertise, and models are developed specifically for translation (so no use for image creation, but focused on translation). Language Weaver Pro is the largest dedicated translation model (over 100B parameters).
Quality and safety arguments:
Quality arguments: MetricX benchmarking is used to compare specific language pairs (BLEU scores are no longer a relevant metric). In tests using human evaluation compared against a major provider, tests were done at done sentence and paragraph level. Other considerations for quality deal with assessing overgeneration issued (e.g. oscillatory overgeneration where words and phrases are repeated over and over again; partially detached overgeneration, when the LLM sycophancy comes through as unwanted output, or detached overgeneration, e.g. where an LLM refuses to translate content (particularly relevant in law enforcement use!)) Despite having its own LLM (in Language Weaver Pro), Trados can orchestrate any LLM together you’re your translation memory (e.g. for the caser that your company has a specific policy). Generative Translation is about using your linguistic resources rather than relying on the LLM’s best guess. Finally there was also the Confidence Routing approach, which is designed to strategically decide on how translation is done – e.g. when should a translation be sent to a human translator, when should it be machine translated (first pass by AI) with human post-editing, or when could the content be auto-published. In certain circumstances, there are considerations about where AI generated texts lead the reader to disconnect or not.
Munich Re’s experiences with migration to the cloud.
Munich Re has a team of five staff members in its Language Data and Technology Team, within its Language Services The team is headed by Sara Grizzo and handles everything relating to language data and technology and ensuring that systems are up and running and is also responsible for improvements, processes, training, documentation and internal and external language consultancy. Andreas Merz joined Munich Re recently, having also had experience in past roles of cloud-based translation management implementation.
Migration Timeline
The project was a lengthy one, involving many different migration steps addressing a wide range of activities. Previously they had been using TMS although that was only part of their complex landscape as they also had a GroupShare server to allow them to share projects and resources with their vendors. Terminology management was handled in MultiTerm together with QuickTerm, and then there were Trados Studio installations for the actual translation. By moving to the cloud they were able to move away from all of these different solutions, including Trados Studio and to exclusively use the Online Editor. There are only a couple of Trados Studio licences kept for troubleshooting and testing purposes, but now both internal and external translators use the Online Editor.
The major milestones along the way were that the migration project started in November 2024, Trados Enterprise went live at the beginning of October 2025 and TMS was finally switched off in December 2025. Each stage was however far more complicated: the project began by conducting a full inventory about the existing configuration and resources. It took into account the workflows, configurations, TMS fields and the groups, roles and permissions that exist in the system. And then there was the decision-making process about what they wanted to have in the target solution – and in this regard defining templates was new territory after the various configurations and workflows and resources configured in the background on TMS. Under the previous setup, the configuration was very static – with little chance for changing things on the fly.
Use of consultants and dedicated interns
From February/March 2025 they then brought in a consultant from RWS, Ziad Chama, who was involved throughout the configuration stage and also continues to provide consultancy to them. External expert consultants can really help as an addition to the internal expertise for processes and resources, and is a very useful sparring partner e.g. for developing new templates, discussion of broader features.
An intern also did a lot of the manual work to migrate vendors from TMS to Trados Enterprise as it was not possible automatically migrate items like costing models and rates from vendors. During the configuration stage it was also necessary to ensure that marketing, documentation and staff training was conducted as well as setting up single sign on (SSO) access.
Terminology and TM data migration
The terminology migration took place in September 2025, and went from MultiTerm together with QuickTerm to an out-and-out QuickTerm repository. QuickTerm terminology was then mirrored into Trados Enterprise. The migration of data from TMS took more or less all of the summer – after all TMS was also running while the migration was under way. To handle this issue, they used professional services to help with the migration of the Delta TMs (i.e. those that had changed in a live environment since the migration). Another big step with the migration was to conduct a big clean-up and anonymisation of TMs. Once the new Trados Enterprise setup went fully live there were still ongoing delta TM migrations, training and assorted fine tuning exercises. The fine tuning is still ongoing ahead of rolling out new features soon. Regarding training, with each new release in Trados Enterprise there are new features, which may need adjusting, staff members trained and also closer examination done of new features to ensure their use is optimised.
Challenges:
- Personal Identifiable Information (PII) anonymisation: with moving to the cloud, it allowed MT integration into workflows, which was a game changer. However, to allow this to happen, the Compliance department required source text anonymisation. This meant hiding personal data, e.g. names, e-mail addresses, bank account details, driving licence numbers etc. In addition, databases also had to be cleaned up regarding PII. Trados Enterprise does not have a built-in function to handle this yet. Instead they chose to have a student intern develop a PII Anonymiser tool, coded in Python that leverages internal LLM capacity. Now the tool is integrated into workflows, so that project managers can anonymise the source text units before they are sent to DeepL, with the xliff file containing the text that contains the PII. Then while rebuilding the target file Trados Enterprise reinserts the PII data. Then all that is needed, during the DTP and layout verification is to ensure that the data is in the right place, which saves having to manually insert some XLIFFs and looking for matching data. The tool also works with XML files, which can be used for TM cleaning. The tool also works for anonymisation before reimporting data into new TMs.
- Translation Memories: during the inventory stage, it was established that there were 115 translation memories. Naturally there were a lot of duplications between the TMs, for example due to synchronisations between TMSS TMs and the GroupShare TM. Regular in-depth maintenance of TMs is very time consuming. So the migration was also used as the occasion for the necessary consolidation of TMs, but given that everything is no on a single platform, there are no longer any needs for so many duplicate TMs. Also in Trados Enterprise, use is being made of the fields in translation memories (e.g. a single field covering whether it is internal or external), rather than having to have duplicate internal and external translation memories. When evaluating the volumes of translation units that were going to be sent to Trados Enterprise, it was clear that there were enormous volumes of data that might be sent unnecessarily. The approach was therefore taken to archive everything prior to 2020 and have that stored elsewhere. In this regard the archival approach was done using the Creation and LastUpdated dates of TUs. i.e. not just TUs created in the last five years, but also last updated in the last five years. Another advantage for having cleaner, newer data is not only for translation purposes, but also for training LLMs or chat bots, or a RAG approach.
- Training/Change Management: it is very important not to overlook the people who will be working with the new setup, and the user of the new platform need to get used to a new platform, especially with so many different stakeholders, from job requesters, internal translators, external translators, and project managers. As good as a perfect configuration may be, ultimately users still have to be able to cope with it, especially the translators and project managers.
- Translators had to get used to working exclusively in the Online Editor and letting go of Trados Studio. UI differences are part of this process, as users miss familiar features initially, as they may have been renamed or are now in a different menu. The best way to obtain a good feeling about the new environment is naturally using actual translation assignment.
- Project Managers had to get used from moving from very static TMS configuration to a template-based setup – it had already been decided during the configuration process not to have a template for every conceivable scenario, but to have less templates and more flexibility. Now project managers have to go through the tempates and decided whether they need all the workflow steps, and which resources are needed. This needs them to take more action. This made having a lot of meetings, and requesting feedback and understanding what they needed, what their favourite features are. Even now there are regular feedback session to “check in” that everything is working or whether adjustments are needed. This approach has helped to ensure that internal resistance has been broken down.
Lessons learned:
Enjoying a bird’s eye perspective over the entire project provides a lot of insights into how people work. This is what makes it essential to also consider the communication and project management aspects of large projects. In this regard it can be essential to talk to others who have gone through the same process. In this regard use communities – like ETUG or the RWS community – to share your experiences with other users, sometimes they can contribute their own experiences that mean you don’t have to start from scratch yourself.
- Communications: It is important to also consider how you approach the management of the company, especially regarding how the outcome of the project will improve the situation. The tried and tested “saving time and money” approach can prove dangerous as you might have to pay in terms of a drop in quality. The aspect of centralising translation assets can be an approach to take, especially where there are compliance and regulatory considerations.
- Project planning timelines: need to be conservative – unless regulatory reasons impose a very fixed deadline, in which case you have no choice. Allowing yourself extra time helps regarding addressing any issues that crop up after the initial planning phase. Similarly, also do not be afraid to revisit and adjust your initial assumptions: circumstances and people can change. Faster technological development cycles mean rapid changes in technology (as shown by the way new AI models and how they can change the market). If you do change, make sure that you communicate openly with your key stakeholders and management
- Stakeholder engagement: in a big company there are likely to be different stakeholder groups with different interests – typically for example the IT department will get involved in the project, and getting them on board is essential. What for you is a localisation and translation project, might be viewed by IT as an IT tool, which they want to take over the ownership of. You need to talk early to the people you work with frequently, e.g. translation requesters, management, and explain from a translation angle how your translation memory helps the organisation in terms of translation efficiency. In turn also make sure that you involve the users of the solution – the translators and project managers, to ensure that they like using the tool, and that it does help them in their day-to-day work. In this regard, make sure you provide sufficient training and the possibility to test customizations.
- Customisations: it can be useful to try to do as much as possible using the out-of-the-box setup – as every customisation may necessitate further workarounds. After a while, if everything needs a workaround, it can erode trust in the target solution. While customisations are often developed especially for your needs, they can also come with their own special problems.
- External consultants: Using an external consultant can be particularly useful in terms of having an outsider’s perspective regarding training. An external consultant can prove useful if internal relationships between departments can be strained, especially if they have the technology expertise, and can take some of the burden of your shoulders. And this is also really helpful if you have difficult technology questions.
Features Wishlist:
Despite how pleased Munich Re are with the outcome of the migration to the cloud, there are naturally a number of wishes that are unfulfilled, which they chose to highlight, which could be useful for others considering a migration:
- Improvements to the assignment feature: currently it is not possible to reassign a translation task that has say been assigned to an internal translation team to an external team (e.g. through an LSP) at any stage in the workflow – currently this is only possible when the task is active. This is challenging for the project management team, as they need to be aware that something they might have to change along the way, that they might otherwise not be aware of.
- In-built anonymization features: as mentioned earlier, from an insurance or reinsurance perspective, you do not wish to have health information available publicly on the Internet. And this is particularly important in translation. Anonymisation of PII tokens to permit MT use becomes essential. And while a workaround is possible, it would be great to have this feature built-in.
- Online TM maintenance: If you are working with one or more big translation memories, currently there are no ways in Trados Enterprise to break the translation memory down by pages, as there is in Trados Studio. This can be quite cumbersome. It would be nice to have an option for the number of translation units per page and the number of pages of you have.
- Full advanced display filter functionality in Trados Enterprise: Studio has the Advanced Display Filter (ADF), which provides more options when it comes to looking for source and target. This is really useful for quality assurance steps. Enterprise allows searches in the source and target, but doesn’t have the same flexibility.
Concluding remarks: Munich Re’s migration to Trados Enterprise was real success and an achievement that we can be proud of. The platform is powerful and there are plenty of useful features and the regular addition of new features is a further positive step. The feedback we have received and the feedback from the community have shown that we can adapt it to our specific needs fairly independently. This has been a gamechanger compared with TMS, where every little amendment became time-consuming and where we had to read out to professional services. Now we just open the settings and that is it. Our job requesters appreciate the clean and intuitive interface of the customer portal – some have even managed to get up and running without training, without reading the instructions (and setup projects and collect the translations at the end).
Q&A:
Why was 2020 chosen as a cut-off point for “archiving” TUs? Do you have any requirements from compliance about retention periods for TMs? There are no compliance requirements, so it was our choice, and a practical one based simply on it being approximately the last five years’ TUs. It is perhaps worth mentioning that the cut-off was based on last created and last used dates – so e.g. segments created in say 2015 or 2019 but last used in February 2021 would be considered as newer than five years’ old.
Why did you make the decision to stop using the Studio Desktop application as it could be used to open Enterprise projects? It wasn’t so much a licence cost issue, than more of a decision from an IT perspective: and the fact that retaining it means involving IT every time we need something installed on our machines. This gets very time-consuming if you need to involve them every time there is a new version of a single plug-in. Also in Trados Studio, every user could independently change project settings from Studio, which led to inconsistencies in data. Now we configure them on the platform and then everybody has to stick to them. In addition in TMS we used to have problems with XLIFF file corruptions, and we struggled to upload the XLIFFs back to the platform. Now the data stays in the platform so this is not an issue. Also if you work in the Online Editor, it is saved in real time, which is a vast improvement over an XLIFF file saved on the C drive, which if some gets ill, means you have to start from scratch again.
Does Trados Enterprise not have the same level of support for regex-based search compared with Studio? This is something we have support looking at currently, as the RegEx search functionalities don’t seem to be working properly. Sometimes when searching TM searches, we don’t always get the results that we would expect.
Other than the Advanced Display Filter, what are the main features missing in the Online Editor from translator’s point of view? Due to the higher frequency of updates, there are often features added that we don’t necessarily notice have been fixed – although many have been implemented in the end. Since we started using Trados Enterprise, we have also noticed that plenty of features have been implemented. However, one that has been particularly annoying for translators have been when needing multiple review cycles, and we are reviewing someone else’s work – e.g. using an LSP and then doing an internal double review. Where the external vendor changed fuzzy matches or maybe 100% matches, you could only see the new version of the segment and not the fuzzy match. This made it difficult for our internal translators to spot what the external translator actually changed in the segment. In Studio it is still there, but is missing in the Online Editor. This has now been corrected, which is a relief. However, different colours for tracked changes for different reviewers are still missing. At times we are required to do two reviews because of confidentiality or different reasons. If the first reviewer makes changes, and the second reviewer makes their own changes, the colour is the same for both. However, the Online Editor is vastly improved over the one in TMS, which was far more basic. The preview function has also improved. One nice function is also the possibility to have the active source and target segment showing underneath each other rather than left and right.
Was the transition from TMS to Trados Enterprise step-by-step e.g. by departments, document types or order types? A big bang / leap-of-faith approach was taken, i.e. from today you use the new solution. We made announcements on the Intranet in advanced and contacted our regular customers. TMS was kept online for projects that were already underway, but all new job requests or projects were sent through Trados Enterprise and the job request function for TMS deactivated. Once all the projects that were in progress have been concluded TMS was switched off completely. Admittedly, we deliberately chose a relatively quiet period of the year (September/October) for the changeover.
Were there any issues with connecting any other tools or software you’re using in your translation processes? We had massive challenges with Virtual Desktop Infrastructures (VDIs). The VDIs are the virtual machines our vendors have to use when they work with us. We don’t send anything via e-mail or through SharePoint. Vendors login into these virtual machines and access Trados Enterprise. They are able to only download the files there for verification mainly. Otherwise it’s a very, very secure environment and that had been reconfigured from scratch and the connection of the access was very tricky for cloud operations, due to IP restrictions regarding access. This was quite an issue even shortly before go live. The VDI environment means that nothing can be copied to the outside world (only from outside into the VDI). You can’t print anything, and any downloads are only visible within the VDI. And you can’t browse the Internet from within the VDI.
In contrast, the switch from MultiTerm to QuickTerm went very smoothly. Only one tiny issue is still unresolved due to our historic MultiTerm setup. Our database was originally a MultiTerm structure, so we have had issues with links not getting mirrored correctly to Enterprise. A consultant has written a script to implement this, but the implementation has not yet been completed. There have been some issues with images attached to terminology entries, but this is a comparatively small issue, but would probably be a bigger issue for manufacturing or engineering use. Munich Re employees have terminology access through QuickTerm, so terminology work is not done in Trados Enterprise.
Did you miss this session?
By registering for the rest of the ETUG Bitesize Sessions 2026, you’ll receive access to the video of this session and the presentation (password-protected access).
The next ETUG Bitesize will take place on Thursday, 24 September, 2026 from 10:00 – 12:00 CEST and will be on Unleashing the Full Potential of Terminology for LLM-based Translation: Terminology-Augmented Generation Using the Quickterm Studio Plugin – with a presentation by Klaus Fleischmann (Kaleidoscope).