Experiences and lessons learned in 22 years of localization

Sean Song is the founder of Translation Boulevard, a company that provides both translation and subtitling services. Previously he worked as a journalist. He holds master’s degrees from Annenberg School of Communication of the University of Southern California, and London School of Economics.


Sean Song


Sean Song

Sean Song is the founder of Translation Boulevard, a company that provides both translation and subtitling services. Previously he worked as a journalist. He holds master’s degrees from Annenberg School of Communication of the University of Southern California, and London School of Economics.


started my career as a journalist, but subsequently became a translator. As such, I considered one of the pioneers in localization.

I speak from my own personal experience, as I have watched how my friends and colleagues in China used the Chinese version of to search for local hotels. So I reached out to to see whether they would be willing to share their experiences with localization.

Their localization head, Emmanuelle Dumas, shared the history of their localization efforts, their achievements and lessons they had learned. She highlighted their “mixed model,” which combines in-house teams, freelance translators and writers, and translation companies. I hope this will be helpful to other global companies that already are working on localization, as well as to others that are just contemplating starting down that path. Small business owners also may benefit from this, as big companies like also started small. I hope you might agree that this effort, even in the oft-ignored translation industry, has turned up a good story.


Emmanuelle Dumas

Song: Could you introduce yourself to our readers?

Dumas: Sure. My name is Emmanuelle Dumas. I’m a French native speaker and by training a linguist and translator. I have a master’s degree in applied linguistics and have worked all my career in the language and translation industry. I started working on game localization and then moved on to work for some language service providers (LSPs).

In September 2017, I joined to manage the language services within Booking’s content agency. The role is really about defining the language strategy and how best to support the business with innovative language solutions.

Song: Tell us a bit about and your localization history. How did it come to excel as a global company?

Dumas: To give you a bit of history, Booking was one of the first online travel platforms. It was founded in the Netherlands in 1996. This certainly helped shape its future, because the Netherlands is a small country, and the Dutch people love to travel! So the need to become international was there from the very early days. It’s in our genes.

Our first office outside of the Netherlands was opened in Barcelona in 2002. In 2005, was acquired by Priceline, an American platform also in the hotel space. Priceline’s acquisition of gave it a foothold in Europe and, for Booking, provided us financial opportunities to expand further internationally.

From its beginning, Booking was very data-driven. As we looked at adding languages, we always did it according to consumer needs, so languages were added based on consumer demand — at the beginning primarily to translate hotel descriptions. And for a long time, our approach to content was really transactional like that. When it worked well, we continued, and we added languages on the basis of demand.

In 2010, we started adding language specialists to work alongside our UX copywriters, and supporting our cycle of A/B testing. From then, it became clear that the language specialists were supporting the user experience in multiple languages, so we have continued to scale up in that way. Today we have 44 languages. I believe that the last languages that we added were Icelandic and Argentinian Spanish. Last year we added Georgian for our partners’ tools, but not yet for our customer content.

Now, of course, we constantly look at what markets we need to develop next. For me, in my role, I consider what key markets the company is focusing on. Are there languages that we need to add; how do we need to upscale our internal teams; or what content will we be sending to our internal language specialists or external teams in the future?

So from the very beginning, translation and localization have been at the core of what we do because Booking has been international from very early on. Translation and localization boost our business because they allow us to be locally relevant. With physical presence in over 200 offices in 70 countries, we are able to evaluate our local relevance every day through those local offices.

This global reach means that our local teams are constantly talking to our partners and also our guests, of course, and we have constant feedback on how locally relevant we are, and what we could do better in-market. Thus I would say that localization and translation are at the core of our business.

Song: Would you have any lessons others could learn from? I mean common pitfalls or experiences to share with other global companies.

Dumas: We are very lucky to have a unique mixed model with internal language specialists being so involved in experimentation with localized copy, so we are able to tell very early on what works in certain markets and what doesn’t. The local teams, working closely with our UX copywriters, are able to give feedback on copy that does not work well and provide an alternative better suited for use in their market, this can happen very fast. Having worked for LSPs in the past, I would say that if we relied exclusively on external resources, the feedback loop would be much slower.

When you can have internal localization experts and specialists constantly working with UX copy, then it enables faster and higher-level localization and better adaptation to market. My experience is that we are on target for each market faster because of our close coordination, and we ensure the right voice of Booking the first time.

Song: The CEO of SmartLink also made that point in a webinar I heard recently. You have direct relationships with the content writers, creators, translators and your consumers. A lot of companies give everything to the LSP, probably because they just want to have a good night’s sleep, but then they wake up to headaches because of miscommunication. If you have only indirect communication with the content creators, they will not know the immediate response from the market.

Dumas: That’s exactly right. I think you can probably achieve this with an LSP, but they likely need longer agreements, more in-depth communication and longer feedback loops to achieve the results that we achieve when our product teams and language specialist teams are basically part of one team within Booking. The other advantage is that our internal language specialists are passionate about the product, about our company, about the brand and about what they are translating. There is a level of involvement and engagement that you get from internal language teams that you cannot get from an external provider.

Song: Is the cost higher to hire an LSP or to maintain an internal team?

Dumas: Well, I think this is very difficult to compare and evaluate because it depends very much on what you include in the costs of localization. In my experience most companies using LSPs need to involve teams in product and marketing with guidelines, terminology validation, review, feedback and final content approval. These are a lot of hidden costs, and for us, it’s a model that would simply not be efficient or scalable enough.

We prefer to look at the value that having our internal teams generate, the edge it gives us to enable multilingual and multimarket copy experimentation so fast, and similarly, to ensure high-quality locally-relevant branded messaging for our marketing campaigns.

Song: I agree. A lot of buyers just want the certainty of saving money. You point out that it is not only hard to compare whether they are saving money, but the quality of the work can differ, too.

Dumas: Nowadays, it seems that time to market is key. We are working with internal stakeholders and also sister companies that are most concerned with both a short time to market and high quality.

In general, our model provides the best basis for fast turnaround and faster change. If we need to take a different approach, we can do it in less time and more flexibility with an internal team.

Song: What feedback do you get from consumers?

Dumas: We collect user feedback in a number of ways to have different touch-points on quality. Of course we have stakeholder surveys for satisfaction and quality. Also, after every major launch or major project with our internal stakeholders, we use feedback loops to find out how our general stakeholders evaluate our quality. Then Booking employees form another layer of feedback, since all of us are also users and we have a culture of feedback! We have a system in place internally whereby everybody can give feedback on the quality.

I’m French, so when I use in my language, if I see something that I find a little bit strange, I can directly provide my comments through our feedback system.

Finally, we also conduct end-user surveys for customers and partners, to find out what the users think of the language on For example, what they think of the forms of address or our communication in general. We try to measure quality from all of the possible angles.

Song: Other companies hoping to globalize may be just starting their localization process now, so they want to hear what’s best for them: to hire an LSP or to manage an internal linguist team. What’s your view as to what’s the best for a global company?

Dumas: There are advantages to every model. Probably the answer is a mixed approach. Our model is mixed. Our internal teams work on a lot of our most critical, branded content, and they are at the core of our localization process. But, like other large companies, we also use freelancers and LSPs for specialized or very scaled content.

For us at Booking, even though I came from a background of having worked with many LSPs, when I joined Booking and its internal content agency team, I was amazed by how quickly and efficiently we can support the business. Essentially, our platform is really our formula for localizing content and constantly optimizing ourselves. We run a lot of A/B experiments to determine best content solution.

Song: Tell me again, what are A/B experiments?

Dumas: A/B experimentation means basically that we are testing different variants of different elements on a page. For example, how a call to action is worded or how certain information is displayed, to see which performs better with our users. With this model, it’s very useful for us to have internal language specialists working closely with our product teams and our copywriters. It enables us to get content ready and localized fast.

Song: What about machine translation (MT)?

Dumas: I knew that you would ask about MT. We now have live MT in 12 languages, for property descriptions, for example. When our MT team was built, roughly two years ago, they started working on neural machine translation. Because at Booking we had so much data on property description that we had been translating for years with freelancers, we were quickly able to build neural machine models to very high quality. Today, after just two years, our MT team has put 12 languages in production.

Song: How many languages are you currently translating into?

Dumas: Currently we are translating in 44 languages.

Song: Wow, 44 languages! I’m very impressed. How do you control the quality of those 44 languages?

Dumas: As I mentioned earlier, ours is a mixed model with internal and external teams and translating very different content types. So we have a created a quality framework that actually supports all of our needs, including content and services. The quality framework looks at quality from a purely linguistic point of view, but we also consider our stakeholders’ feedback and the level of quality they need for the product. In addition, we also look at the end users’ perception of the quality of language on For example, we run surveys to find what our users think about the language, the tone, how we address the users and their experience when they use

We also can do some very targeted quality research. For example, we can pretest on a specific challenge we have with a language or product.

Song: As my last question, do you have a vision for the future or a strategy for the next five years and beyond?

Dumas: Well, our MT team certainly does, and I think they basically are looking at training their model for more languages. That is one of the things they are working on, and also more use cases for machine translation. Now we are using MT only for our property descriptions and user reviews, but not in many languages. Obviously, we are trying to deploy MT on more use cases, and more languages.

On my side, I’m also concerned midterm with what the next languages should be that we need to focus on, the markets our business is focusing on, whether we will need to add new languages and how are we going to go about it.

Now that has grown to be a platform where people come to book much more than just their accommodation, we have many products and a lot of different types of content. Our language team’s strength is that we are integrated. We are looking at basically collaborating with the business on more of their content, what the next leads will be and how we can support them. For instance, last year we localized the Booking Assistant chat product into several priority languages to support our customer service, providing faster solutions to users.

Song: This has been so helpful. Thank you very much.