How to Migrate Your Translation Workflows to an AI-First Model

Annica:
Hi everyone, and welcome to today’s webinar! We’re going to explore how to migrate your translation workflows to an AI-first model. My name is Annica, and I work with client strategy at Comactiva Language Partner. I'm excited to be joined today by Jourik, a language technology expert with extensive experience in the localization industry.

Jourik:
Thanks, Annica. Great to be here.

Annica:
Before we dive in, could you tell us a bit about yourself?

Jourik:
Of course. My name is Jourik, and I wear a couple of hats in the localization space. I’m the CTO of Yamagata Europe, a mid-sized language service provider based in Belgium. I’m also the co-founder of my own company, CJ International, where I focus on three main areas:

• Translation technology consulting

• Localization engineering, things like automation, scripting, and integrations

• Market research

That’s me in a nutshell.

Annica:
Great, sounds like your background is a perfect fit for today’s discussion.

To kick off: We all agree that implementing AI in the translation process is key. But like any solution, it’s important to know how and where to apply it to create real value. Jourik, in your view, which areas of translation and localization can AI deliver the most value today?

Jourik:
That’s a great question. AI’s potential does seem endless at the moment, and I imagine if we revisit this in a few months, I’ll have even more to say. But right now, I’d highlight three key areas:

• Machine Translation

• Quality Assurance

• More Sophisticated Workflows

Let me elaborate.

With large language models (LLMs), we can actually talk to our models. We can give instructions, what content we’re translating, for whom, what the purpose is, and provide context, examples, glossaries, and more. That’s a big shift from traditional machine translation, where input goes in and output comes out with little room for nuance.

Then there’s Quality Assurance. I honestly believe the QA systems in most TMS tools are outdated. They generate a lot of false positives, lack context awareness, and have limited customization. With generative AI, we can improve this significantly by using contextual prompts, examples, and rules tailored to each project.

Finally, Workflows. Many workflows today are static, document-based, manually triggered, and people-dependent. Generative AI can enable dynamic workflows with features like content classification, quality estimation, content routing (deciding whether a piece goes live or to post-editing or SME review), and content repurposing, for example, turning a blog post into a LinkedIn post.

Annica:
That’s really exciting, so many possibilities. But where should someone start if they want to implement a more AI-driven translation process?

Jourik:
It’s crucial to set the right expectations. You can’t just plug ChatGPT into your process and expect everything to be faster, better, and cheaper overnight. It doesn’t work like that.

Start gradually, and start with a problem scenario, a slow process, a quality issue, or something requiring too much manual effort. Once you've identified a use case, prepare your resources: do you have the data needed? A solid glossary? Common errors from previous MT runs? Information on where time is being spent?

Once you’ve got that, you can start building prompts. AI implementation begins with clear, structured input and guidance. So to summarize:

• Set the right expectations

• Take it step by step, use case by use case

• Ensure you have the resources and data to support the use case

Annica:
Completely agree, data is key. Now, what about human expertise? How should we rethink its role in this new AI-driven landscape?

Jourik:
Human expertise is still essential. AI is powerful, but it’s not perfect, it makes mistakes. If you want human-level quality, a human still needs to sign off.

That said, human linguists need to adapt. Traditional tasks, like working in CAT tools, may decrease, while new responsibilities, like prompt engineering, will emerge. Linguists are actually well positioned to do this, because they understand the importance of clear, precise instructions. The nature of the work will expand, not disappear.

Annica:
Such an important point. Now, there’s often confusion between neural machine translation (NMT) and generative AI like LLMs. Can you explain the difference and when to use each?

Jourik:
Sure. Think of LLM translation as the more flexible, creative sibling of NMT.

NMT is predictable, same input in, same output out. It’s also faithful to the source, which is great for technical or legal content. Generative AI, by contrast, is adaptive and creative. It might give 10 different outputs for the same input, and it’s far more customizable.

So:

• Use NMT for structured content, technical manuals, legal texts, anything where precision matters.

• Use generative AI for creative tasks, marketing copy, social posts, etc. where tone and flexibility are important.

Annica:
That makes sense. Have you seen common mistakes or obstacles when companies try to implement AI-first translation workflows?

Jourik:
Yes. two things stand out:

Mistake: Unrealistic expectations. Thinking you can plug something in and instantly get better results is a trap.

Obstacle: You have to get your hands dirty. Implementing AI requires prompt writing, customization, working with glossaries, integrating into tools and systems. And that means you need engineering expertise, someone who understands APIs, scripting, Python, SDKs. That’s new for many in this space.

Annica:
Agreed. So what kind of expertise should a company have in-house to make this work?

Jourik:
Beyond engineering skills, companies need the right technology stack. Traditional desktop tools won’t cut it anymore. You need systems that connect, tools that talk to each other via APIs.

I’m a big fan of workflow automation tools like Make.com or Zapier. They help you integrate everything, from an email trigger to a finished translation, into automated, seamless workflows.

So my advice: audit your current tech stack. Is it ready for AI? If not, now’s the time to invest and adapt.

Annica:
Absolutely, and with how fast things are moving, flexibility is critical. What works today may change in six months.

Thank you so much, Jourik, for sharing your insights! It’s been a pleasure having you with us.

Jourik:
My pleasure, thanks for having me!

Annica:
Thanks again to everyone for watching. This is our final webinar before the summer, but stay tuned and sign up for our upcoming sessions after the break. Have a great summer!

The Expert Speakers