HSBC once spent roughly $10 million fixing a single slogan. “Assume Nothing” was meant to signal careful judgment. In several markets it arrived as “Do Nothing.” Customers read it as indifference. The bank had to overhaul materials and messaging after the fact. The translators had worked from text alone. They never saw the phrase sitting on an actual billboard or in a television spot. Context was missing, and the brand paid for it.
That pattern keeps repeating. Pepsi’s “Come Alive with the Pepsi Generation” landed in Chinese as something closer to “Pepsi brings your ancestors back from the grave.” KFC’s “Finger Lickin’ Good” became “Eat Your Fingers Off.” In each case the words were technically rendered. What was missing was the final check that asks whether those words still work when a real person encounters them in the real place they will appear.
More recent examples hit closer to software and apps. When Amazon opened its Swedish site in 2020, machine translation produced product descriptions that swapped “rapeseed” for the Swedish word for rape, turned roosters into crude slang, and labeled frying pans as items suitable for women. The Argentine flag appeared where Sweden’s should have been. Social media turned the launch into a running joke within hours. The company scrambled to clean it up, but the first impression stuck. Native reviewers looking at the live site would have caught most of it in a single afternoon.
These are not isolated gaffes. Industry data keeps pointing to the same gap. Poor localization routinely costs companies around 20 percent of potential revenue in new markets. Surveys of marketers show that nearly three in ten admit a translation or cultural misstep has already damaged their brand. Almost 40 percent say their worst localization error cost more than $10,000 once lost sales, staff time, and reputation repair were tallied. Manual regression testing after the fact is expensive and slow; the alternative is shipping with errors that customers notice immediately.
Translation and LQA are not the same job. Translation converts meaning from one language to another. LQA puts that converted text back into the actual product—mobile screens, game menus, video subtitles, voice-over timing—and checks whether it still fits, still sounds natural, and still makes sense to a native speaker using the product the way a customer will. A sentence that looks fine in a spreadsheet can overflow a button, contradict a character’s gender on screen, or land as tone-deaf once the surrounding visuals and interaction are visible. German compounds routinely expand 20–35 percent beyond the English source; short UI strings can double. Russian Cyrillic glyphs often demand more horizontal space even when the character count looks similar. Without LQA those expansions truncate labels, overlap icons, or force awkward line breaks that make an otherwise solid app feel unfinished.
Fixing overflow is mostly design discipline applied early. Leave roughly 30–50 percent extra width on short labels. Prefer flexible containers over fixed pixel widths. Test the longest target languages first on the smallest common devices. Pseudo-localization during development surfaces the problem before any real translation begins. Once the localized build exists, a practical mobile LQA pass checks string completeness, expansion on real screens, in-context flow of dialogue and menus, visual fit across orientations, cultural tone, encoding and font rendering, and basic performance under longer text. Screenshots and short recordings attached to each issue make the hand-off to developers unambiguous.
Automated tools can accelerate the mechanical parts—string extraction, consistency checks against glossaries, basic length validation—but they still miss the judgment calls that only a native speaker playing the actual build can make. The most effective setups combine machine assistance for scale with human LQA for the final contextual pass. Skipping that human layer is how “All your base are belong to us” became an internet meme instead of a forgotten 1989 arcade translation, and how modern app launches still generate the same screenshots of clipped buttons and nonsensical error messages.
The disasters share one root: treating localization as a late, isolated translation step rather than an integrated quality process. Brands that treat LQA as optional discover the cost only after the product is live and the screenshots are circulating. Those that build it into the release cycle avoid the expensive rework and the public embarrassment that follows.
Artlangs Translation has spent more than twenty years refining exactly that integrated approach across 230-plus languages. With a network of over 20,000 professional linguists and a track record of successful projects in translation services, video localization, short-drama subtitle localization, game localization, multilingual dubbing for short dramas and audiobooks, plus multilingual data annotation and transcription, the company has helped clients catch the context gaps and overflow problems before they reach users. The difference between a smooth international launch and a viral PR headache often comes down to whether that final native review happened inside the actual product.
