Marketing localisation: one campaign, every market, without rebuilding it each time

1. 1

   Tier the markets before you brief anything

   Not every market earns its own creative treatment, and pretending otherwise is how localisation budgets get spread thin enough to be useless everywhere. Sort markets into three tiers: the ones that get an original concept because the audience, the category or the competitive set is genuinely different; the ones that get the shared concept with local casting, local offer and locally written copy; and the ones that get the shared concept in their own language with the commercial details corrected. Size the tiers by revenue and by how different the buying behaviour actually is, not by how loud the local team is, and write the tiering down so the exceptions are decisions rather than accidents.

2. 2

   Write the campaign as a definition with slots, not as a finished master

   A brief that produces one beautiful film and then asks eleven countries to adapt it has already decided that ten of them get a compromise. Write the campaign as the parts that must hold everywhere and the parts that are meant to change: the promise and the proof are fixed, while the hook, the casting, the setting, the offer, the season and the register are slots each market fills. This is the single change that turns localisation from a queue of adaptation requests into a generation run, because a slot can be filled in parallel and a finished file can only be reworked in sequence.

3. 3

   Hold the local differences as overrides on one brand definition

   One definition carries the colour values, the logo rules and the product truth. On top of it sit named overrides per market: the type stack for each script with a real commissioned weight rather than a default substitute, the dialect and level of formality, the colours that carry a different meaning locally, the claims that are legal in one country and not in the next, and the calendar that market runs on. Kept as overrides they stay visible and auditable; kept as separate files they quietly become a parallel brand nobody at the centre can see. There is a longer treatment of the governance side at /use-cases/brand-consistency.

4. 4

   Generate each market version natively instead of resizing one

   One concept becomes a matrix: hooks against scenes, against the placement sizes each channel demands, against the markets and languages you sell in. A vertical video, a feed square, a marketplace listing image and an email header are four compositions rather than four crops, and an Arabic version is a mirrored composition rather than a flipped export with the logo landing where the eye exits. Generating from the shared definition inside one system, which is what /platform/creation-studio is for, is what keeps the hundredth variant recognisably the same brand as the first. The mechanics of building and sizing the grid are covered at /use-cases/ai-ad-generator.

5. 5

   Route review to native speakers by risk, not by volume

   Reviewing every localised asset end to end is arithmetic nobody wins: five minutes each across three hundred assets is most of a full time role spent confirming that logos are the right size. Tier it instead. Anything carrying a price, a regulated claim, a legal mark or a recognisable person goes to a named reviewer in that market. Anything rendered in a script the generation model draws unreliably gets read at full size by someone who reads it, because text that looks plausible to a non reader can be nonsense to a reader. Routine variants of an already approved concept pass on the rules alone and get audited by sample.

6. 6

   Tag by market and dialect at creation, then read per market

   Every asset carries the variables behind it, market and language and register included, in a naming scheme that survives export from the ad platform. Give each market its own budget and a fixed window, and compare variants within one platform rather than across platforms, since each delivery system optimises differently and will hand your cells unequal spend within days. Then look at what the localisation actually bought: hook rate and completion for what the creative did, conversion and cost per acquisition and return on ad spend for what it was worth, and how long each version held before it decayed, which is usually different per market and is the number that sets the local refresh schedule.

7. 7

   Write the market finding back into the next brief

   This is the step that makes the whole exercise compound instead of repeat. A winning asset expires within weeks; a finding about a market is good for years. When the local dialect read beats the neutral one in one market and loses in the market beside it, when the in context scene wins on paid social in one country and the plain packshot wins in another, when a market fatigues in eleven days and its neighbour in four weeks, those belong in the overrides that constrain the next few hundred assets rather than in a slide someone presents once. A team that does this arrives at each new market already knowing more than the last time, and that accumulated read is the part a competitor cannot buy.

Why adapting a finished campaign costs more and lands worse

The standard model is sequential. A campaign is built for the largest market, approved by everyone who has an opinion, and only then handed out for adaptation. By the time it reaches the third market the concept is fixed, the budget is mostly spent, the season it was written for is closer, and the only lever a local team has left is the copy. So layouts get flipped rather than recomposed and the logo lands where the eye exits, headlines written to a syllable count in one language overflow in another, a model who reads as aspirational in one country reads as foreign in the next, and a seasonal frame built around a holiday that market does not observe is quietly dropped. Every one of those is a small compromise, and the sum of them is why the second and third markets underperform the first and get read as evidence that the market is difficult rather than that the process was.

Composing each version from the same definition changes the sequence rather than the effort. The promise and the proof are decided once, the slots are filled in parallel, and every market gets a first version rather than a derivative one. It also makes the exceptions legible: when the local difference is written as a named override on a shared definition, someone at the centre can see that a market is running its own dialect, its own offer and its own calendar, and can tell the difference between a deliberate divergence and a brand that has quietly forked into eleven brands. The teams who do this ship into every market in the same week, which is the difference between arriving for a retail moment and arriving after it.

Reading which market version actually earned the return

Most localisation programmes are judged on delivery: how many markets were covered, how fast, at what cost per asset. Those are production metrics and they say nothing about whether the localisation worked. The question worth answering is narrower and much more useful: did the locally written version beat the neutral one in this market, by how much, and for how long. Answering it requires each asset to carry its market, language and register from the moment it is created, in a scheme the ad platform reporting preserves all the way to the export, because that association cannot be reconstructed afterwards from campaign names.

The findings are usually more specific than teams expect, and they rarely generalise across a region. A dialect that lifts conversion in one Gulf market can be neutral in the one next to it. A market can prefer the same hook as the home market but fatigue on it three times faster, which means the creative was right and the refresh schedule was wrong. Two markets that look identical in a media plan can want opposite scene treatments. Read within one delivery platform rather than across platforms, hold the test window even when day two looks decisive, and be careful with small markets, where low spend produces numbers that look like findings and are noise. Then the part that matters: a finding belongs in the overrides that constrain the next generation run, not in a quarterly review. That is what makes a localisation programme compound, and it is the only part of it a competitor with the same budget cannot copy this quarter.

Where marketing localisation falls short

It does not decide whether the market is worth entering, and it does not rescue a proposition that does not travel. A product priced for one economy, a delivery promise your logistics cannot keep locally, a payment method the market does not use, and a competitor who already owns the category there are not creative problems, and localising against them buys you a well made argument for something the audience is going to decline anyway. The most useful thing a per market read sometimes does is make that unmistakable early, while the media budget is still mostly unspent.

It also does not supply cultural knowledge you do not have. A system can hold the rules you give it and apply them with more discipline than any human, which is exactly why a wrong rule is worse than no rule: humour, religious sensitivity, gesture, colour meaning, what counts as aspirational versus showing off, and the difference between a dialect that sounds warm and one that sounds like an outsider imitating it are all judgements that need a person from that market. Anything regulated is stricter still, since advertising rules, health and financial claims, price display and comparative advertising differ country by country and stay with a named human reviewer rather than waiting to be automated. Generated text in scripts that image models draw unreliably needs a native speaker reading it at full size before any spend goes behind it.

And the measurement has real limits. Several market variants touch the same buyer in an overlapping region, view through effects are invisible in most setups, small markets produce results that are indistinguishable from noise, and a market with a short test window will hand you a confident number that does not survive the next month. Treat per market asset results as a strong directional signal, rerun a surprising winner before rebuilding a country plan around it, and keep enough human review capacity to match whatever generation volume you unlock, because a review queue that cannot keep pace becomes a worse bottleneck than the studio it replaced.