How Topcon scaled localization without a large localization team

Topcon scaled AI translation across engineering, technical publications, marketing, and eLearning by pairing Smartcat AI review agents with a human proofreader and running everything through templates. The result: a 5,000-segment, 31-language review that used to take two weeks now finishes overnight, and translated UI ships in a couple of days instead of twice a year.

In a recent Smartcat webinar session, Brittany Clark, Senior Director of Product Marketing at Smartcat, sat down with Michelle Quirke, Program Engagement Manager at Topcon, to unpack how Topcon got there: the workflow, the metrics that built confidence, and the lessons she wishes she'd known ten years ago.

• Speed without trading off quality: Smartcat AI review agents review each file in about an hour, so a 31-file job triggered at the end of the day is ready for human proofreading the next morning.

• Data, not gut feel, retired internal review: Topcon dropped mandatory in-country review on eLearning once reporting showed reviewers were changing well under 5% of content.

• Every AI change is explained: The review agent leaves a comment on every segment it edits, giving project owners visibility they never had with human-only workflows.

• Templates are the scaling lever: Assignment and project templates turn localization into a few clicks for teams who aren't in Smartcat every day.

• The next frontier is content governance: Topcon wants AI to help find duplicate, conflicting, and outdated content before it's ever translated.

Topcon is a global manufacturer of precision technology, including machinery used heavily in agriculture. Its translation journey started in Latin America, where many machine operators don't speak English — and Topcon had no manuals in Portuguese or Spanish.

A support manager in the region began using Smartcat to translate short quick-reference guides for customers and dealers. Michelle was asked to assess whether the platform was fit for purpose. “I had a look at it, and then I just got sucked into it,” she said.

The early days were painful. The first two manuals Topcon translated were Word files that many people had edited over the years. After translation, Michelle spent 30–40 hours reassembling them, finding hidden images and stray text every time she realigned a page.

Today, she supports engineering, technical publications, marketing, and eLearning teams — without a large dedicated localization team.

Topcon's model rests on two principles: ownership and risk-based involvement.

• Business units own their own translation. Each team takes responsibility for its workflow, so the workload doesn't funnel back to one person.

• High-risk content gets hands-on oversight. Engineering UI strings can run to 20,000 words and 5,000–6,000 segments. “If we get it wrong, we get it wrong in a big way,” Michelle said. So she checks UI projects up front and locks segments that must not change.

That structure, combined with AI, changed the release cadence. Where engineering once shipped two large translation updates a year, some teams now run live projects that push small updates weekly or monthly. Even UI translation turns around in a couple of days.

For eLearning, the time savings are concrete: Michelle measured roughly 2.5 hours saved per course, per language. An agriculture course localized into nine languages saves about 22 hours of hands-on work.

An AI review agent is an AI coworker that reviews translated content in context — checking terminology, meaning, and consistency — and edits segments where needed, before a human proofreader does the final pass.

Topcon's core workflow for eLearning looks like this:

1. AI translation produces the first draft.

2. An AI contextual review agent reviews and corrects it.

3. A Smartcat Marketplace proofreader completes the final human review.

Topcon's company-wide rule is simple: AI agents are welcome in the workflow, as long as a human always follows them.

Marketing is next. Its current workflow uses two Marketplace stages plus an internal review. The planned workflow adds an AI contextual review agent and an SEO agent, followed by a Marketplace proofreader. Internal review becomes an optional add-on for new products, not a default stage.

Michelle ran a head-to-head test. While investigating why long-standing UI translations had changed unexpectedly, she proposed a sanity check across a product UI that had been live for years — in 31 languages.

She ran Smartcat AI review agents over the UI. For comparison, she also built separate agents with Microsoft Copilot, restricted to reviewing only the content she provided.

The other agents couldn't handle the volume, and their output drifted — “they were fine for a little bit, and then they just wandered.” The AI review flags areas of concern, and humans make the final call on what changes go through.

For years, Topcon added a final internal review stage. In-country review is a step where a native-speaking employee who knows the product checks the translation before release. It's an established best practice, and research in regulated fields such as clinical outcome assessment still recommends it. At Topcon, though, those reviewers did it on top of their full-time jobs.

Michelle used Smartcat's reporting to look at every eLearning project and measure how much internal reviewers actually changed. The target was anything below 5%, and most projects sat well under that.

Meanwhile, the stage was a bottleneck. Some projects sat in internal review for a full year, and chasing them took more of Michelle's time than any other part of the process.

So Topcon removed internal review from eLearning, roughly 18 months to two years ago. The proof was in the courses themselves: employees, support staff, and dealers are required to complete them, so they see the translated quality firsthand. Once other teams saw the data and the results, technical publications followed, keeping internal review only for some new-product manuals.

Michelle gave a precise before-and-after for a roughly 5,000-segment job across 31 languages. She triggers all 31 files at once, finishes for the day, and has them ready for human proofreading the next morning.

That speed creates options. When a product team says a language won't ship unless translations arrive by a deadline, Topcon can now deliver an AI-only first pass immediately — so the team can check the UI on screen — while a human proofreader finishes the final version within two days.

The Smartcat review agent leaves a comment on every segment it changes, explaining why. Human translators typically comment only when they need clarification.

Those comments are now Topcon's best tool for bringing other teams on board: show them the edits and the reasoning. One longtime Marketplace translator even validates the agent's comments, sometimes leaving a single word — “great.” As Michelle put it, it's one word, but it says everything.

Standardize the workflow and hide the complexity behind templates. A workflow template is a saved set of stages, assignees, and linguistic resources that every new project inherits automatically, so users pick a template instead of configuring a project from scratch.

When Michelle first trained internal teams, there were no templates. Every user had to learn how to build a project and which linguistic assets to attach. Now, assignment and project templates — plus connectors from Adobe Experience Manager (AEM) and Adobe Workfront — do that work automatically.

For occasional users, that means a couple of clicks and the job is done. For Michelle, it means she no longer monitors every project daily. Notifications tell her when something needs attention; otherwise, “it runs itself.”

Faster translation exposes a bigger problem: knowing which content is current. Topcon's content comes from support, product management, subject matter experts, and technical writers. Over time, that produces near-duplicate quick guides and manuals that say slightly different things.

Regulation is raising the stakes. A new EU requirement means declarations of conformity must be available without a login, and countries such as France require all product content in the local language before a product can be sold.

Topcon is responding by adopting Simplified Technical English for shorter, more reusable manuals, authoring in AEM, and protecting a single source of truth so that a change made once updates everywhere. Michelle is also exploring Smartcat's upcoming Content Update Coworker, which reviews existing content and recommends what to create or update, for an in-house expert to approve.

Her ask for Smartcat: stretch further into the workflow on both sides — before translation and after. The reason it matters is the customer.

That aligns with what we see across the market: 96% of enterprises are expanding their multilingual content, according to Smartcat's Global Growth Report 2026.