All videos
0:00 / 0:00
ai

Episode 4: How do you localize 79 billion words a year? With Mik Szajna of Booking.com

DeepL16 June 2026Watch on YouTube

Description

As Head of Localization for one of the world's largest travel platforms, Mik Szajna is responsible for localizing roughly 200 million words every single day. Automation and AI are an essential element in tackling the challenge - but so too are strategic thinking, an appetite for experimentation and innovation, and an intelligent approach to localization quality and proving business impact. This is a great episode of "The New Fluency" - here are some of the highlights: Scale demands intentionality "There are obviously lots of challenges when you deal with a scale like this. At Booking, we localize 79 billion words annually. That's 200 million words a day across 45 languages, and over 100 different content types. So the main challenge is: where do you even begin? What is worth your attention and focus? Obviously, when you localize at such scale, you have to automate things — that's a given. But even with automation, there are so many different solutions, and some are going to work for some content types and languages, and others maybe not so much. It's about being intentional about where you apply the technology and how you do it." Don't assume, and don't do too much at once "We are all still trying to figure out how the technology works. There are no clear answers. It's a bit of a black box. We can measure the outputs, we can look at how consistent the results are, but there's always a bit of risk. One thing that companies sometimes do wrong is to try to do too much at the same time. You try to experiment with AI for translation, and then quality assurance, and then quality control, and then let's see how it does with terminology. It creates a bit of noise, and the combined effect of all those efforts can actually backfire." Rethinking quality for an AI-first world "We are rethinking a little bit what quality is. At the end of the day, quality is whatever works for customers. At Booking, we apply the concept of good enough quality specifically for localization — we have four quality levels that we apply to different content types and products, and we try not to over-localize. That's relevant in the context of AI, because if we're working with a model that does quality estimation and automated post-editing, and human post-editors over-edit, then that affects the model. The model gets confused. It's about striking a balance between what works for customers, what the technology can do, and looking at quality through the business lens." Proving the value of localization "About a year and a half ago, we did a large-scale blockout experiment, where for a small percentage of our traffic, we turned off localized content for a good few weeks. Then we compared the impact against the localized product and extrapolated across all 45 languages. The impact was really significant. It very much proved the premise: country won't buy. That allowed us to position the value of localized content within the business in a very clear way. And because almost all of the content at Booking goes through automated localization, we actually equate the value of localization with the value of automated localization." Soft skills are the new superpower "We recently went through a big hiring exercise — we hired 40 people across different languages. For those people who didn't make the cut, it wasn't necessarily a lack of technical skills. The technical skills were often there. What was sometimes missing was the soft skills: communication, being able to work cross-functionally. With all of the technology, soft skills are becoming extra important, because you have to explain complex technical concepts to stakeholders and influence them, and navigate a complex business landscape. AI has become a common denominator for many of us. In localization, we've dealt with AI for a while — machine translation is AI — but now it's something we share with different parts of the business. We're part of conversations we weren't part of before, and you need to speak the language of the business." #TheNewFluency #LocalizationAtScale #LocalizationAI

What you'll learn

  • Localization at scale of 79 billion words annually requires intentional application of AI and automation, not everywhere at once
  • Defining quality as what works for customers, not perfection, prevents over-editing that confuses AI models
  • Soft skills like communication and cross-functional collaboration become crucial when implementing AI technology
  • Proving localization value through large-scale experiments helps convince stakeholders of business impact

Frequently asked questions

How do companies like Booking.com localize such massive volumes of content efficiently?
Through intentional application of automation and machine translation at scale, with different quality levels for different content types. It's about strategically choosing where technology has the most impact.
Why is it risky to run too many AI experiments simultaneously?
Running simultaneous experiments with translation, quality assurance, and terminology creates noise that can backfire on the combined effect of efforts. This makes it difficult to measure what actually works.
What does Booking.com mean by 'good enough' quality in localization?
Quality that delivers the desired outcome for customers, not perfection. This is flexible per content type and product, and prevents over-editing that confuses AI models.
What skills were often missing among candidates who weren't hired?
Not so much technical skills, but soft skills like communication and ability to work cross-functionally. These become crucial when explaining complex technical concepts to stakeholders.

Topics