Localisation as transcreation

Translation turns your words into another language. Local re-writing finds out what people in that place actually type, and writes for that instead. The first produces filing words that fit for nothing anybody actually asks.

Why straight translation fails

A translated hidden pocket is a list of correct words that simply aren't the words people use.

• The everyday word is often not the dictionary word. A borrowed foreign word, or a local brand-turned-generic term, is frequently what people actually type. And a good translator will correctly avoid exactly that word.
• How long a question is depends on the place, not the grammar. How people phrase what they want is a local habit, not a rule of the language itself.
• Double meanings don't survive. A single word that usefully covers two meanings in one language usually splits into two separate words somewhere else. Either doubling your cost, or halving your coverage.
• The very category itself can be named differently. What a shop even is can be called something different in a different place.

This matters because every piece of writing is set separately per sign-language, and the desk matches against your sign, the line beneath it, your hidden pocket and your floor, whichever sign-language a given shopper actually gets shown.

Start from branches, not from languages

Every branch has its own default sign-language plus a set of others it also reads, and which ones those are differs sharply from branch to branch. What a shopper actually sees depends on their own location's setting, their own device, which sign-languages you've actually written, and your own primary language. And where nothing matches, it falls back to the branch's own default, or to yours.

Two consequences people miss:

One sign-language can serve several branches at once. A language that's merely "additional" in one huge branch might be the default across a whole group of smaller ones.

The cheapest expansion is usually inside a branch you already have. Adding a sign-language your existing branch already reads is often better value than opening in a whole new country.

How to actually do it

Do your homework in the target language, from scratch. Not "translate the local list and check it after." Start from how the problem is actually described there: a rival's own shopfront in that branch, reviews written in that language, autocomplete with that branch actually set, your own complaints from that market.

Use someone who actually uses shops in that language. Not only a translator. The real question isn't "what's the correct word". It's "what would you personally type."

Check before you commit. A word you're unsure of can be tested with a small paid billboard in that branch before it takes up any real room in your writing.

Write your filed pieces first, your persuasion pieces second. If you can only do one at a time: sign, the line beneath it, and your hidden pocket carry filing weight; your leaflet and flyer carry persuasion. But don't ship a translated sign above a leaflet still in the wrong language and call the market done.

Sort your branches into tiers. A handful get real homework and locally-made photos; the rest get correct, unembarrassing basics. Treating every single branch with the same deep attention produces a pile of mediocre ones instead.

The cheap way in: a machine translation nobody actually sees

Everything above is an argument against machine translation. There's exactly one place that argument doesn't apply, and it's genuinely useful when you're deciding whether a branch is even worth real money yet.

The hidden pocket is never seen by a shopper. So a machine-translated hidden pocket carries no risk to conversion at all, the worst that happens is it fits for words nobody asks for. That makes it a probe: fill it, watch for a season whether anything moves in that branch at all, and use the answer to decide whether to fund the real thing. Experienced shopkeepers call this a bare-minimum opening.

Two conditions, and they aren't optional.

Check every translated word against a real, live search before using it. A translation can be technically correct and still be the wrong word entirely. The clearest example: machine-translating a phrase can return the job title for a role instead of the everyday activity, technically correct, but almost nobody looking for the shop actually types the job title. The translation is right; the word is dead on arrival. A quick popularity check catches this in seconds.

Never extend this trick to anything visible. Your sign, your pitch line, and your window captions are all read by people, and a machine-translated visible page costs you conversion and trust in exactly the branch you're trying to evaluate. If the probe works, the visible writing is the thing you then pay a real person for.

This doesn't apply on the other side of the centre. There, the leaflet is both indexed and visible, so there's no hidden pocket to probe with. And that centre already offers shoppers its own automatic translation of an unwritten branch, so bolting your own machine translation on top mostly just replaces one automatic translation with a worse one you now personally own.

Your window photos are part of this too

Redressing your window for a local audience is the most commonly skipped lever in this whole job, and it gets skipped because it's genuinely expensive, not because it doesn't work. A caption on a window photo is text a shopper reads right there in the desk's own answer; leaving it in the wrong language in a local branch is the single most visible sign that the shop wasn't actually made for that place.

What tends to go wrong

Machine-translating the hidden pocket and stopping there. Produces correct words nobody actually types.

Localising only the visible writing. The hidden pocket is where the actual filing happens, and it's the one thing nobody reviews internally, because nobody ever reads it.

Treating one language as one market. The same language spoken in three different places can mean three completely different sets of words people actually type.

Leaving your primary sign-language wrong. It's the fallback for every branch you haven't written for, the default experience for most of the world.

Opening sign-languages you won't actually maintain. A stale one is worse than none at all, shoppers still see it.