Inside Smartling’s Push to Automate Global Translation with Bryan Murphy

Welcome to Revenge of the Nerds. We’re skipping the hype and going straight to the builders. In this edition, we talked about:

• After scaling software and e-commerce platforms at companies like eBay, Bryan Murphy stepped in as CEO of Smartling to modernize the way global content is translated.

• He sees large language models reshaping not just translation quality, but the entire infrastructure behind localization—automating workflows that were once slow and manual.

• Smartling’s focus on AI-driven workflows, linguistic assets, and agent-based systems is moving translation from a human-heavy process to a background service that scales with the business—turning it into a growth driver rather than a bottleneck.

Let’s dive in. No floaties needed…

Revenge of the Nerds

Bryan Murphy, CEO of Smartling

He first became interested in translation when he noticed how slow and manual the process was for getting global content online. That frustration sparked his curiosity.

After being introduced to Smartling through Battery Ventures, Bryan saw a big opportunity in a large but under-automated industry. Instead of joining the board, he chose to take on the role of CEO to help drive that transformation himself.

How will LLMs like GPT or Gemini reshape translation over the next few years?

I see large language models transforming translation in two key ways. First, they improve the translation itself. Second, they change the entire infrastructure behind how translation happens.

In enterprise translation, a big part of the cost and complexity comes from the process, not just the act of translating. Companies have huge ecosystems of websites, products, learning systems, and legal documents. Managing all of that content across languages is the hard part. Automation is critical, and LLMs enable the creation of dynamic workflows that handle translation from start to finish.

When people think about translation, they usually picture the output—the words on the page. But the real cost lies in everything that leads up to that point, such as collecting, preparing, and processing the content. LLMs are helping automate those steps.

They also allow us to build agent-based systems that manage many of the manual tasks that used to slow everything down. For example, we recently launched an MCP server that lets other AI applications connect directly with ours. That means one system, such as a content management platform, can automatically communicate with Smartling to translate text in real time.

So overall, LLMs are improving both the quality of translation and the efficiency of translation. They are helping us move toward a world where translation feels seamless, fast, and deeply integrated into every business workflow.

Where do you draw the line between what AI can do on its own and what still requires human oversight?

AI is great at handling single, well-defined tasks. If you ask it to translate something or summarize a piece of text, it does that very well. Where it still struggles is in performing multiple tasks or managing complex sequences, like ‘book me a flight’ or ‘translate this, then send it to that system, then verify the output.’ Those multi-step workflows are where human oversight is still needed.

In translation, AI can now handle around 80-95% of the work, especially when it is trained on high-quality data and supported by automated quality assurance. But for important content, you still want a human to review and approve the final version.

It is similar to how people use AI tools in their own work. The AI can create a strong first draft or framework, but you still need to apply judgment, context, and a human touch to make sure it is right. AI takes you most of the way there, and humans ensure the end result is accurate, thoughtful, and aligned with intent.

How is Smartling using AI to preserve a brand’s voice while keeping localization scalable?

We use a combination of what we call linguistic assets to help brands maintain their unique voice while scaling localization. These assets include translation memories, glossaries, and style guides, all of which feed into how we train and fine-tune our AI engines.

For example, translation memory ensures that once a phrase has been translated and approved, it stays consistent in every future use. Glossaries define specific terminology or internal language, ensuring that the brand’s preferred words and tone are used correctly in every language.

We also work closely with clients to develop and maintain detailed style guides. This helps capture differences in tone between brands, even within the same industry. For instance, IBM and Apple are both technology companies, but their voices are very different. Our system learns those distinctions so that the same sentence would be phrased in an IBM tone for one client and an Apple tone for another.

All of this is reinforced through computational linguistics, prompt engineering, and continuous model training. That combination allows Smartling to deliver translations that are not only accurate but also sound like the brand itself, no matter the language.

What is the next frontier of multilingual SEO as global search becomes more AI-driven?

SEO is quickly evolving into what I like to call AIO, or AI Optimization. Traditional search engines and AI models crawl, interpret, and recommend content in very different ways. As AI-driven search tools like Perplexity, Claude, and GPT become more common, marketers need to understand how these systems evaluate and surface information.

The rules that defined SEO for the past decade no longer apply. AI models do not rely on the same ranking signals that Google traditionally used. Instead, they focus more on semantic meaning, context, and how well content aligns with a user’s intent.

For global brands, that means multilingual SEO will have to adapt. Translating keywords alone will not be enough. Companies will need to optimize their localized content so that AI systems can understand it in each language and region.

In short, the SEO playbook we have relied on for years is being rewritten. The next frontier is learning how to make content discoverable and relevant to both humans and AI systems across every market.

As AI translation becomes faster and cheaper, does it risk becoming commoditized, or does it make language more strategic?

I actually think faster and cheaper AI translation makes language more strategic, not less. When I was at eBay, I saw firsthand how localized content drove engagement and revenue growth. The more content we translated, the more traffic we got. But because translation used to be slow and expensive, companies often rationed it and focused only on what they could afford.

Now, with automation and AI, we can translate 99% of a company’s content at a fraction of the cost and time. That means businesses can finally localize everything: websites, marketing materials, documentation, training content, all of it.

It is similar to what economists call Jevons Paradox. When the cost of a resource drops, demand for it actually goes up because new use cases emerge. The same thing happened with data storage. As the price of storage fell, people started saving everything because it became so valuable to keep.

The same principle applies to translation. As costs fall, companies translate far more content, which expands their global reach and deepens customer personalization. In other words, cheaper translation does not commoditize language, rather it unlocks its full strategic potential.

What would the ideal translation workflow look like in five years, and how would it differ from today?

In five years, translation will be almost completely automated and integrated into everyday workflows. Instead of translation being a separate, manual process, it will happen continuously in the background as people create content.

Imagine writing a 5k word technical document. As you type, the system is already translating and localizing your work in real time, checking quality, tone, and brand consistency across multiple languages. You will not need to think about translation; it will just happen.

This model is what I call “translation as a service.” It sits alongside your content creation tools and ensures everything you produce is instantly available to a global audience. The human role will shift from doing the translation to managing the systems, training the engines, and maintaining the linguistic assets that keep everything accurate and on brand.

The process will be faster, more seamless, and far more scalable than it is today.

In five years, do you expect less human involvement?

Yes, I expect less direct human involvement in day-to-day translation, but people will still play an important role. Most of the workflow will be automated, with AI handling the translation, localization, and quality assurance in real time.

Humans will focus more on training the engines, maintaining linguistic assets, and monitoring quality rather than performing manual translation work. Think of it like running a production line. The machines handle the heavy lifting, but skilled professionals oversee the process and step in when something unusual happens.

So while the number of people needed for repetitive translation tasks will decrease, the need for highly trained linguists and computational language experts will remain. Their role will shift from doing the work to guiding and improving the systems that do it.

AI translation depends heavily on training data. How do you balance accuracy and speed while preventing data leakage?

Data is everything in AI translation. The quality of the annotated data determines how accurate and reliable the models are. What makes Smartling unique is that we have more than 30B words of professionally translated, verified data. Every sentence in that dataset has been reviewed and approved by a human linguist, which gives us an incredible foundation for training our models.

We use this proprietary data to continuously improve our translation quality, and we protect it very carefully. Preventing data leakage is a top priority. We have strict legal agreements with our AI partners that prohibit them from using our data for their own training purposes.

This approach allows us to maintain control over our linguistic assets while still taking advantage of the latest AI technology. It is how we balance speed and innovation with accuracy, quality, and data security.

Additional Reads

• In conversation at Slator (“On the Real ROI of AI in Localization”),

• Not another CEO (“Early AI adoption and leadership at Smartling”),

• On The Retail Podcast (“Scaling global businesses with language AI”)

Quick Poll

Rate This Edition