For decades, localization has relied on subject-matter experts—also called SMEs—to review translations and make sure they’re right. They also produce content, answer questions, and define terminology so we know how something should be expressed.
Using localization SMEs effectively is harder than it looks, whether those SMEs are specialized linguists with years of knowledge in a specific field or employees with deep market or product knowledge. Ironically, the reason has nothing to do with SMEs themselves.
Working with SMEs gets complicated because nobody has figured out how to pay for know-how. Localization is a business that’s traditionally run on buying words, hours, review cycles, and turnaround times. Those models were designed to manage translation volume as cost-efficiently as possible. Expertise doesn’t fit as neatly on a line item.
What about using technology to solve the problem? It’s true that AI has made translation faster and more scalable. Enterprises are producing more content in more languages, and the output sounds convincing most of the time. The tricky part is that AI makes errors that sound and feel authentic. Without somebody who really knows the subject matter, you have no reliable way to know whether the output is right.
Recently, Argos’ Global Marketing Director Stephanie Harris-Yee sat down with industry consultant Erik Vogt to discuss why SME engagement is so hard to build into an enterprise localization program and what a smarter approach looks like.
Three Kinds of Experts, Three Different Problems
A product manager at a pharmaceutical company and a freelance translator with ten years of medical experience aren’t the same resource and shouldn’t be treated as one. These experts fall into three different SME groups that localization companies rely on.
The most familiar group is made up of product managers, engineers, and regulatory leads at the company that developed the product. They understand not only the product but also the technical and regulatory issues around it, which enables them to judge whether a translation is accurate. Their biggest obstacle is competing priorities—namely, their own day jobs.
“They often have other jobs to do,” Erik says. “Reviewing might not be their primary responsibility. Even translation might not even be their formal responsibility.”
The second group are in-country partners, sales reps, and local market contacts who understand how a product should perform in a specific region, what customers expect, and what local regulations require. Erik describes them as an extension of the sales ecosystem. They carry knowledge that the people who built the product usually don’t have. Localization programs generally don’t have a formal way to include them in the review process.
The third group are specialized linguists who have spent years developing domain expertise in a particular field. The market doesn’t compensate them for that expertise. They move through the same hiring channels as generalists, get evaluated on the same criteria, and get paid using the same per word rates.
“We don’t buy experts,” Erik says. “We buy words and hours, and that transaction tends to bury the value of that subject matter expertise and how to put that experience to work.”
Pricing Out Expertise
Ask a procurement manager to put domain knowledge on a purchase order. There’s no line item for 10 years of medical device experience, no way to price familiarity with a specific regulatory environment. The localization industry built its purchasing model around things it could count, and expertise doesn’t count.
“If expertise isn’t measurable,” Erik says, “it becomes unpurchasable.”
When procurement drives unit costs down, it pushes out the specialist linguists who priced their work on their knowledge. Generalists fill the need, and programs that relied on expert judgment now depend on people who need more oversight from the same internal team members who were already hard to reach. Users lose confidence in the output, and when things go wrong, the enterprise blames the vendor.
“Translators are being commoditized at the exact moment we need them to be acting as domain experts,” Erik says.
AI Enters the Picture
AI has expanded what enterprise localization programs can produce, and teams are translating faster and at lower cost than before. More fluent output also means more content that needs expert review, and that review has gotten harder.
These days, AI output is fluent enough that errors aren’t easy to spot. A translation can pass a surface read and still be wrong about what a product does or how it should be described in a specific regulatory context.
“AI makes fewer obvious and easily identifiable errors,” Erik says. “It’s making more believable mistakes that are harder to detect.”
A specialist who has spent years working on a product knows things no model was trained on. They know the vocabulary a regulatory body uses and how a particular client applies it. When output needs a second look, they know that too.
When specialists are priced out of the workflow, the work of catching errors falls to internal product and regulatory teams. This forces the enterprise to use its most expensive employees to do the proofreading that their language partner would have traditionally handled.
Making the Most of Real Expertise
Expert reviewer time is the scarcest resource in a localization program. In a perfect world, your whole workflow would be designed around it.
A common principle from operations management calls this the Theory of Constraints. It says that when one resource limits the whole system, the answer isn’t to push more work through it. It’s to prioritize it, schedule around it, and make sure it’s only working on what actually requires it. Right now, localization programs generally do the opposite—they attack the volume first and figure out who needs to review it later.
The current model assumes that if the per word rate drops, the program is more efficient. It ignores the fact that as AI volume increases, the cost of verifying those words becomes the dominant expense. In any localization budget, the unit price of a translator is irrelevant compared to the cost of an engineer’s time spent fixing a believable AI mistake. The goal is to stop treating expertise as a line-item cost and start treating it as the primary resource that dictates the pace of the entire system. Without it, every increase in AI volume puts more pressure on the internal team.
Trusting in the Truth
The localization industry spent 20 years building anonymized, commoditized translation. Now AI is making it impossible to pretend that the linguists doing the reviewing are interchangeable. The errors it produces require someone with the expertise to catch them, and that person can’t be a generalist from a vendor pool. The purchasing models just don’t account for the value of smart people in key roles. They were built to optimize what was easiest to measure.
The result is a system where everybody is optimizing for the wrong things at the same time. Buyers are focused on per word prices because that’s what’s in the contract, but the language partner is focused on timelines and volume. Everybody agrees that SMEs are a valuable resource, but no one is paying for the privilege of their expertise.
“The closer we get to AI delivering more and more ‘close enough’ content,” Erik says, “the ability of those subject matter experts to validate the truth becomes more important than ever.”
If you’re looking for a better way to build specialist review into your localization program, contact us.
Argos Multilingual
16 min. read
A confusing message is circulating about the economics of AI. One camp insists that hiring back a junior engineer is now cheaper than using AI to build things; another insists it has never been cheaper to dig in and build apps yourself, no external teams required. Both claims can be true at once. Tokens are […]
