Breadcrumbs

AI Workflows

AI can do more in localization than translate a sentence. Wordbee AI workflows put AI to work at three points in your process: producing the first translation, improving translations that already exist, and checking linguistic quality. They run inside your Wordbee Translator projects and read your translation memory and your terminology first, so the output follows your style and your approved terms instead of sounding like a generic engine.

The workflow names used in this documentation, such as MT workflow: 3 Agents, are representative: the name on your platform may look a bit different. The services themselves are the same, and the type of work (translation, post-editing, quality assurance) and the number of agents are always indicated in the name.

Benefits at a Glance

  • Quality that is fit for purpose. Match the amount of AI work and human review to each kind of content, so you can move more volume without treating all of it as high-stakes.

  • Your language data does the heavy lifting. The workflows lean on your translation memory and terminology, so the better curated those are, the more you get back.

  • You decide where people stay in the loop. Automation is set per workflow step, so you switch it on exactly where it makes sense.

What AI Workflows Cover

Area

What it does

Where it helps

AI translation

Produces the first translation of your content. Unlike a machine translation engine, it reads your translation memory and term base for every sentence and follows instructions you write in plain language.

Files reach the Editor already drafted in your own style and terminology, so linguists correct less.

AI post-editing

Rewrites a translation that already exists to bring it up to publishable quality.

Cleaning up machine translation output, and raising partial memory matches without putting a linguist on every one of them.

AI quality assurance

Adds AI checks on top of the built-in quality checks. One agent verifies terminology against your references and checks dates, numbers, codes, tags, and placeholders. A second agent running alongside reads for meaning: content added to or missing from the translation.

Catching what rule-based checking misses, such as a term that is present but wrong for the context, or a dropped qualifier that changes the meaning.

Both AI translation and AI post-editing can run as one agent, or as several in sequence that check and correct the pass before them. The multi-agent setups cost more and take longer, so choose them for content that earns the extra passes rather than as a default.

AI workflows run on text content inside Wordbee Translator projects, in any file format Wordbee Translator can process.

Why Context Changes the Output

Every Wordbee AI workflow reads your own linguistic data before it produces anything. For each sentence it pulls matching entries from your translation memory and matching terms from your term base and passes them to the AI as reference, alongside your written instructions. Soon the metadata you keep in segment custom fields will join that context in the post-editing and quality workflows.

That is the difference between an AI that translates a sentence correctly and an AI that translates it the way your organization translates it. A machine translation engine can be given a glossary or trained on your content, but those are separate assets you build, maintain, and retrain as your terminology moves. Here nothing is precompiled: approve a term this morning and the translation this afternoon uses it.

A machine translation engine may still return the sentence faster, but the output here arrives already carrying your terminology, so the correction work that usually follows machine translation has largely been done before anyone opens the file. Your established team does not get smaller. It gets paired differently, with experts and AI agents each on the work they are better at.

How a Workflow Processes Your Content

An AI workflow does not send your document to the AI in one block. It works through the content in batches. For every batch, the workflow queries your project at that moment: the segments to process, together with the translation memory entries and terms that match them under your configured filters. The batch travels to the AI with that reference material, the results are saved back to your project, and only then is the next batch prepared, with reference material retrieved fresh again. Translations saved earlier in a run become part of your project's own memory and can serve as reference for the batches that follow, which is what keeps a long document consistent from its first segment to its last. In multi-agent workflows, further agents review and correct each batch before it is saved.

All of this preparation is real work, so a run takes longer than a classic machine translation engine, which returns raw output the moment you ask. While a workflow runs you can follow its progress (see Tracking AI Work), and what arrives at the end already carries your terminology and your style: exactly the time you would otherwise spend correcting it afterwards.

Working Alongside Your Team

AI translation runs when your file is prepared, so the document is already drafted before the first job starts. AI post-editing and AI quality assurance are assigned to jobs the same way you assign a supplier, and each starts as soon as its job does, with no one clicking to accept it. AI quality assurance can also run as an automated check between jobs, or on demand from the Editor and from the job's details page.

Your workflow structure does not change. The jobs, the sequence, and the assignments work as they do today, and a step a person handles now can be handled by an AI supplier instead. A project can be fully automated, fully human, or any mix in between, with full visibility of what ran and with what result.

What You Control

You configure AI workflows from Wordbee Translator, per service and per project:

  • Instructions to the AI: your own guidance on top of the built-in instructions, covering tone of voice, brand rules, terms never to translate, register, and domain conventions.

  • Reference material: how similar a memory or terminology match must be to count as reference, and which approval statuses qualify.

  • Scope: which content the AI works on and which it must leave alone, such as anything a person has already edited, or locked segments.

You can also configure the same workflow more than once with different instructions, for example one set for marketing content and another for legal content.

Your AI subscription stays yours. Workflows run on your own provider account and your own key, so you contract with the provider directly, pay them directly, and keep your own usage and quota in view. Any provider can be used, and different steps of the same workflow can run on different models. We configure that with you at setup, and we are working on putting it directly in your administrators' hands.

What Is Coming Next

We are extending the catalogue with further AI workflows, including:

  • Language adaptation: adapt an existing translation to a regional variant, changing only what the target locale requires.

  • A conversational assistant in the Editor: ask for context, alternative phrasings, and terminology clarification while you translate.

Learn More

Talk to Our Team

Which AI workflows fit your content depends on what you publish, how much of it there is, and how much review it needs. We work that out with you before anything is set up, and we run a pilot on your own content so you see the quality on your material rather than on a demo.

Contact your Wordbee account manager or talk to our sales team to arrange a walkthrough.