79 Billion Words a Year: What Enterprise Localization Actually Looks Like at Booking.com
Booking.com processes 79 billion words of localised content every year. That equates to 200 million words a day across 45 languages and over 100 different content types. At that volume, localisation becomes an infrastructure problem rather than a content problem.
In the latest episode of "The New Fluency", DeepL's podcast about language, technology and business, Mik Szajna, Head of Localisation at Booking.com, sat down with Morana Perić to explain how one of the world's largest travel platforms manages to localise on such a large scale.
The core challenge is knowing where to begin
For companies wondering how enterprises manage multilingual workflows on a large scale, the honest answer is that it's not just about volume. It's about prioritization. “The main challenge is knowing where to begin,” Mik explains. “What is worth your attention and focus, and what maybe not so much?" When localising at Booking.com's scale, it's impossible to apply the same level of attention to everything. The question, then, becomes about localization strategy: Which content types, languages and workflows justify significant investment? This is a fundamental tension for any enterprise localisation program, and it doesn't get any easier as you grow.
“Almost all content goes through automated localization. However, the system is not monolithic. Different content types are handled differently, and the team continuously validates whether each approach delivers the right results,” says Mik.
Defining translation quality
“Ultimately, quality is whatever works for customers. But even that is a bit of a minefield," says Mik.
Booking.com uses a model that combines quality estimation with automated post-editing. When human editors over-edit, the model becomes confused. It applies the wrong threshold. It also feeds back into the machine that produces future outputs. This has important implications for anyone building or evaluating enterprise translation software. In a high-volume, AI-assisted workflow, quality is a system property, not just a property of individual translated items.
The ROI of localization
For localization teams trying to justify investment or prove business value, Mik shares one of the most compelling examples — Booking.com's large-scale blackout experiment.
“Some time ago, Booking.com disabled localized content for a small proportion of its users across all 45 languages for several weeks. We then compared the impact of it with the localised product,“ Mik explains.
“The impact was really significant. It very much proved the premise: can't read, won't buy." This is the clearest possible answer to the question of the value of translation software for a business: If people cannot read content in their language, they will not make a purchase.
Watch the full episode!
Mik and Morana go deeper on experimentation, quality frameworks and what they think the next few years will actually look like for the localization industry. Watch full episode on YouTube: How do you localize 79 billion words a year? With Mik Szajna of Booking.com. Want to hear more? Catch all of Season 1 of The New Fluency on our YouTube.
Designing AI-first localization processes with Claudia Mueller, Trusted Shops Beating the complexity bottleneck with a product mindset with Olga Stokowiec, Maxon Ship fast, translate faster with María José Salmerón Ibáñez, GitLab