Adding a new locale to a software product takes time, money, and significant coordination. Even when that locale is a dialect variant of a language you already support, many teams treat it as a new project, but it doesn’t have to work that way.
When HiBob, a global HR platform, expanded into Spain, its European Spanish-speaking customers made clear that the Latin American Spanish version of the product wasn’t working for them. Adding Castilian Spanish using a traditional localization process would have meant the same cost and timeline as a new language launch. Instead, HiBob used AI to adapt its Latin American Spanish translation memory (TM) rather than start from scratch.
Pavel Riazanov, HiBob’s Senior Localization Manager, recently joined Stephanie Harris-Yee on Global Ambitions to explain how the adaptation was done. Global Ambitions is a podcast featuring focused conversations with global content program leaders.
Pavel began with the translation memory that HiBob built over years of supporting Latin American Spanish. Before starting translation, he engineered the TM to reflect Castilian Spanish conventions. This pre-populated most of the new locale’s content—content that would have otherwise been translated from scratch at full cost.
Engineering the TM First
Research was Pavel’s first move. Before touching any tooling or workflow, he mapped the differences between Latin American and Castilian Spanish, consulting native speakers to validate his findings on vocabulary, tone, and regional conventions. Only then did he use AI to apply those changes systematically across HiBob’s Latin American Spanish TM, transforming it into a Castilian-ready asset. The new Castilian Spanish TM went through another human review before it was used for translation.
Once the TM was ready, the new locale launched with roughly 70% of its content pre-populated automatically. The remaining 30% was made up of new content and segments with no usable match. These went through AI-assisted translation followed by human review. The result was a Castilian Spanish locale delivered dramatically faster and at a fraction of the cost of a new language. It was also the first-ever AI-translated language in HiBob’s product.
The two phases asked different things of AI. In the first phase, AI applied the dialect changes Pavel defined through research and native speaker consultation, working with an existing asset rather than generating new language. In the second, AI translated the content the adapted TM couldn’t cover, functioning similar to conventional machine translation. Humans reviewed the output of both.
Getting the Order Right
Skipping TM preparation and going straight to translation would have produced a different result entirely. The Latin American Spanish TM would have pre-populated content in the wrong dialect, requiring linguists to review the entire locale to catch mismatches that should never have made it into the translated content.
The time and cost savings Pavel achieved came from eliminating that problem before translation began. AI made the TM transformation fast enough to be practical, but the decision to do it first made all the difference.
When Stephanie asked Pavel what he’d recommend to someone looking to update their own workflows, he had one piece of advice: go back to basics.
“Do not start with tools. Start with clarity. Before you just rush and introduce AI or reinvent a workflow, you have to be very clear on what problem you’re solving, who the audience is, and what kind of language you’re working with,” he said.
Your TM Has More to Offer
The project’s success was largely due to how the AI was directed. By defining the dialect’s differences upfront, Pavel provided the clear instructions needed before asking AI to do anything. This level of preparation transformed the project into a replicable strategic blueprint for launching new languages. For languages with more complex structures, Pavel recommends testing in small batches and validating the quality before moving forward.
Localization teams that carefully evaluate and find ways to adapt their existing assets using AI can find they have more options than they realize. For HiBob, the years of Latin American Spanish translations stored in their TMs pre-populated 70% of a new market’s content before a single new segment was translated.
If you’re expanding into new markets and looking for new ways to get more out of your existing translation assets, contact us to find out more.
Argos Multilingual
5 min. read
Nobody likes being stuck in the middle. When you’re an enterprise localization manager, being caught between a technology provider and a language services provider is a familiar scenario. When the two vendors don’t talk directly to each other, the localization manager becomes the go-between who manages the relationship between their vendors. This isn’t a staffing […]
