The KFC problem: what a famous mistranslation still teaches performance marketers

Every localization deck eventually reaches for the same story. KFC's English slogan, an idiom about the sheer enjoyment of the food, entered the Chinese market rendered close to word for word, and came out as an instruction to eat your own fingers off. It's a good story because it's funny, and it's a durable story because the failure it describes never actually went away. It just stopped happening at slogan scale and started happening at headline scale, inside campaigns nobody outside the account will ever see.

The mechanism is exactly the same. "Finger lickin' good" isn't a claim about fingers. It's an idiom whose job is to signal enjoyment, and idioms don't carry their meaning in their individual words. Translate the words and you lose the job the phrase was doing. This is not a rare linguistic edge case. It's the default failure mode of any translation process that scores itself on accuracy to the source text rather than on whether the target reader gets the same message the source reader got.

In a PPC account, this shows up constantly, and almost never gets a headline. A German search ad built around "get more bang for your buck" translated as a literal explosion metaphor instead of the native phrase German buyers actually search for. A Meta hook built on a pun that simply stops being a pun in the target language, so the copy reads flat where it was supposed to land a laugh. None of these trip an obvious error. They read as fluent, grammatically correct sentences. They just don't do the job the source line was written to do, and that gap only shows up later, as a CTR or CPA number that's quietly worse than it should be.

Three things follow from treating the KFC case as a pattern instead of a punchline:

1. Identify the phrase's job before translating it. Is it literal, or is it doing something else, an idiom, a pun, a cultural reference, a claim? That question has to be answered before a translator or an AI model touches the line, not after.
2. Literal accuracy and correctness are different things. A translation can be word-perfect and still fail the only test that matters: does it produce the same reaction in the target reader that the source line produced in the source reader.
3. Automated scoring won't catch this alone. Match rate and glossary compliance are necessary checks, but they were never built to detect a figure of speech. That's a job for a risk signal that specifically flags idiomatic, figurative, or culturally loaded language for human review, before it ships.

That third point is the actual argument for risk-routing over blanket AI translation or blanket human review. A phrase like "finger lickin' good" should never clear as Green on match rate alone. It's exactly the kind of line the routing layer is built to catch and send to a native transcreator, the same way our own engine flags claim density and language risk before a string ever reaches a human queue.