At SlatorCon London 2026, Phrase CEO Georg Ell and Tripadvisor Director of Localization Riccardo Cocco discussed one of the biggest questions facing localization teams right now: should companies build their own AI localization systems or buy existing solutions?
Moderated by Slator Managing Director Florian Faes, the panel looked at the growing excitement around internal AI projects and the risks companies often overlook when moving from pilots to production.
Today, it is possible to build working AI translation prototypes in a matter of days using frontier models and APIs. The output often looks fluent, creating the impression that localization can now be done cheaply and at scale.
But according to Cocco, this simplicity can be misleading. Internal teams may produce text that appears correct in another language, he said, but still miss the harder parts of localization: brand consistency, cultural nuance, compliance, and making content resonate with users across markets. As these systems scale, the problems become more visible. “The cracks are going to show up sooner or later,” Cocco warned, pointing to risks such as inconsistent multilingual experiences, legal or compliance issues, and potential brand damage.
He also shared an example from Tripadvisor involving an internally built AI system that generated multilingual travel tags. While the output initially looked successful, localization reviewers quickly found culturally problematic stereotypes in languages such as Spanish and Catalan because the prompts had been designed mainly with American travelers in mind.
The Hidden Cost of Internal AI Localization
Ell described himself as firmly “on the side of builders,” but argued that companies should think carefully about what is worth building internally and what is already available in the market. Companies, he said, should focus on what truly differentiates their business rather than rebuilding technology that already exists and works well.
Cocco agreed, arguing that companies should focus on what’s important for them and their brand and leave the infrastructure layer to specialized providers. He used a travel analogy to make the point: “You focus on selling the best travel package to your customers. You don’t build the planes.”
Both speakers argued that companies often underestimate the long-term operational cost of maintaining AI localization systems. Ell said many companies fall into a “sunk cost fallacy,” assuming that internal engineering resources are effectively free while overlooking maintenance, scalability, governance, privacy, model updates, and future feature requests.
Localization Value Is Moving Beyond Translation
The panel also highlighted how localization value is increasingly moving beyond translation itself. Localization teams, the speakers said, should focus on intent, adaptation, and business outcomes rather than translation alone. The challenge is no longer simply producing linguistically correct content, but ensuring multilingual content achieves the intended commercial or user outcome in every market.
That shift also changes how localization teams need to measure success internally. Cocco argued that traditional localization key performance indicators (KPIs), such as word counts and turnaround times, no longer matter much to executive teams. Instead, localization leaders need to connect their work to business metrics such as conversion, customer engagement, or support outcomes. Ell said Phrase is increasingly focused on measuring what he called “intent proximity” — whether content actually achieves its intended purpose.
“We’re really interested in this idea of intent. No one’s actually interested in a translation; they’re interested in content that sells something, motivates the reader or viewer to do a thing. So how do we capture that intention at the beginning of a process and measure success at the end? […] The translation step on its own is not where the value is created. It’s about ‘intent proximity’ — getting closer to the intention of the business process.” — Georg Ell, CEO, Phrase
As the discussion wrapped up, both speakers encouraged localization teams to position themselves as strategic partners in AI adoption rather than defenders of legacy workflows.
At Tripadvisor, Cocco said the company focuses on building systems where internal applications connect to language technology platforms (LTPs), which then connect to AI models. This allows the company to focus on user experience and business logic, while relying on LTPs to keep up with the fast-moving model layer.
That flexibility became an important theme during the audience Q&A. Responding to a question about how quickly internally built systems risk becoming obsolete, Ell argued that enterprises should assume anything they build today will eventually become redundant as models evolve. The solution, he suggested, is not to avoid building entirely, but to avoid building rigid systems. Companies should think in terms of composable architectures that can be continuously upgraded rather than monolithic systems that become difficult to maintain as the AI landscape changes.
The panel ended with a final question from Faes: What is the one uncomfortable question companies should ask before scaling an internal AI localization prototype?
Ell said business leaders first need to define what success actually means and think carefully about how they plan to maintain these systems over time. Cocco focused on accountability, asking what happens if an AI localization system “fails silently,” how quickly teams would notice, who would be accountable, and how the issue would ultimately be fixed.
A recording of the panel will soon be available to Slator subscribers on the Video Library.
