Ninety-Eight Percent Match Rate
In Eliyahu Goldratt's The Goal, a plant manager named Alex Rogo brags to a physicist named Jonah about how robots have improved his factory's efficiency by thirty-six percent. Jonah asks a few simple questions—did you ship more product? did you reduce headcount? did inventories go down?—and Rogo realizes, with a sinking feeling, that none of those things happened. The robots just made one station faster. The plant got worse.
Here is what that conversation might sound like at a localization conference.
"You use translation memory?" she asks.
"In every project," I say. "We've been building our TM for almost eight years now."
"Has it really increased your productivity?"
"Absolutely. We're seeing a ninety-eight percent match rate on some projects. Ninety-eight percent. Our translators barely have to touch those segments."
She looks at her coffee.
"So your brand voice is ninety-eight percent more trusted in those markets? You're reaching ninety-eight percent more people effectively?"
I laugh. "Well—no. It doesn't work like that. The match rate is about reuse. Cost savings. But the voice and brand side, that's a different conversation."
"Is it?" she says. "I'd have thought that was the whole conversation."
I feel my smile freeze.
"Let me ask you something," she says. "Are you shipping more completed translations per quarter than you were before you built up the TM?"
I think about it. "I'd have to check the numbers."
"Did you reduce your review team?"
"We can't do that. If anything, the review burden has gone up. We have more volume now, and reviewers still have to check every fuzzy match. Some of the matches are old. Different translators, different years. You have to read carefully."
"So your review costs didn't go down," she says.
"No," I admit.
"What about time-to-publish? Are translations reaching your audience faster?"
I hesitate. "We have a serious problem with turnaround times, actually. It's a pain point with our stakeholders."
She nods, like she expected it.
"But my cost per word went down," I say. "My match rates are above ninety percent across the board. To stay competitive, we have to leverage that TM. If we don't reuse those matches, we lose our cost advantage."
"So you have to keep reusing them," she says.
"Of course."
"Even when the matches are outdated? Even when terminology has shifted, or the audience has changed, or five different translators left their preferences in the database over the last eight years?"
"We do TM cleanup sprints before big projects," I say. "It's not perfect, but—"
"Be honest," she says. She smiles. "Your backlogs are growing, aren't they?"
I look at her. How does she know?
"If you mean the review queue—"
"All of it," she says. "Review backlogs. Terminology inconsistencies. Complaints from in-country teams about tone. Projects that technically shipped but that nobody is confident about."
"We have some of those issues, yes."
"And everything is always late?"
"One thing I'll admit," I say, "is that we have a real problem meeting deadlines. It's becoming an issue with the communities we serve."
"Here is what I think happened," she says. "You optimized the part of the process you could measure. Match rates. Cost per word. Reuse percentages. Those numbers went up, and it felt like progress."
"It was progress."
"It was local efficiency," she says. "But the goal of your organization isn't to match strings. It's to get a clear, trustworthy message to the people who need it—in their language, in their voice, on a timeline that matters. That's the throughput that counts. And on that measure, what happened?"
I don't answer.
"Your TM made one step cheaper," she says. "But it made the whole system more fragile. Review takes longer because reviewers can't trust the matches. Consistency is worse because the database is full of conflicting decisions from different eras. New translators inherit a voice that isn't theirs and isn't anyone's. And you can't stop reusing the matches because your cost model depends on them."
She pauses.
"You're not failing because you're bad at this. You built exactly what the industry told you to build. But the metric you optimized isn't the goal. It never was."
In The Goal, Goldratt's insight is deceptively simple: a system where every station is locally efficient can be globally terrible. The robots in Alex Rogo's plant were producing parts faster than downstream stations could absorb them. Inventory piled up. Bottlenecks shifted. The plant looked efficient on paper and was drowning in practice.
Translation memory does the same thing. It optimizes the production of matched segments—the easiest, most measurable step in the process—while quietly making everything downstream harder. Review, consistency, tone, coherence, trust. The parts of translation that actually determine whether the work lands.
This isn't an indictment of anyone who built or relies on a TM. The tool made real sense in the era it was designed for. The problem is structural, not personal: TM measures what it can count and ignores what it can't. And the things it can't count are the things that matter most.
The question worth asking isn't what's your match rate. It's what's your goal?
Is your translated content actually reaching more people, faster, with greater clarity and trust? Is the voice of your organization—or your translation team, or your faith community—coming through more clearly this year than last? Are the people you serve telling you it's working?
If those numbers are going the wrong direction despite a world-class TM, the TM isn't the solution. It might be the constraint.
At Frontier R&D, we build tools that optimize for the actual goal: clear, trustworthy, human-guided translation at a pace that matches the need. Not faster matching. Better reaching.