Localisation Is a Costume

Most localisation is translation with a colour palette.

That is not cultural adaptation. That is a costume.

The distinction matters. A costume changes how something looks. Cultural adaptation changes how something works. Most companies — including most companies that believe they are culturally competent — stop at the costume. They translate the strings, swap the flag icons, adjust the date format, and declare the product “localised.” The product is not localised. It is dressed up.

The difference between a product that works across borders and a product that merely appears in multiple languages is the difference between surface and structure. Surface is what you can change with a stylesheet. Structure is what requires you to rethink the architecture.

What Surface Looks Like

Surface localisation is the visible layer. The part the project manager can put on a spreadsheet. The part that produces a satisfying checklist.

Language. Translate the interface strings. Replace the content. Run it through machine translation or, if the budget permits, a human translator. The output is text in another language. The output is not communication in another culture. Translation converts words. It does not convert meaning. A sentence that is grammatically correct in German but uses American casual register is a sentence that has been translated and not adapted.

Visual elements. Swap the imagery. Replace the American stock photo with a European stock photo. Change the colour palette to something that “feels local.” The colour palette does not make a product local. It makes a product colourful.

Formats. Date formats, number formats, currency symbols, measurement units. DD/MM/YYYY instead of MM/DD/YYYY. Kilometres instead of miles. Euros instead of dollars. These are mechanical substitutions. They are necessary. They are the minimum. They are also the maximum, for most companies.

Legal compliance. GDPR notice, cookie consent, terms and conditions in the local language. This is legal obligation, not cultural adaptation. A product that is legally compliant in the Netherlands and culturally foreign in the Netherlands is a product that will not be used in the Netherlands.

Every one of these items is real work. Every one is necessary. And every one, individually or together, constitutes the costume — the surface layer that creates the appearance of cultural adaptation without the substance of it.

The substance is structural.

What Structure Looks Like

Structure is what operates beneath the surface. It is the set of assumptions the product makes about the user — assumptions about hierarchy, trust, communication, time, relationships, and uncertainty. These assumptions are invisible to the development team because the development team shares them. They become visible only when the product crosses a cultural border.

Reading direction and information hierarchy. Arabic reads right-to-left. This is not a styling preference. It is a perceptual architecture. The eye enters the page from the right. The primary information should sit on the right. Navigation should flow from right to left. The hierarchy of importance maps to a spatial logic that is the mirror image of the Latin convention.

Mirroring a left-to-right layout is not sufficient. A right-to-left layout that was designed from the right is structurally different from a left-to-right layout that has been flipped. The spatial relationships between elements, the weight distribution across the page, the location of the call to action — these carry different meaning when the reading direction changes. CSS logical properties make the flip technically trivial. The design work that follows is not trivial. It requires understanding how Arabic-reading users scan a page, where their attention falls, and what spatial arrangement communicates hierarchy in their perceptual system.

This is structural. A stylesheet cannot do it.

Hierarchy perception. Hofstede’s power distance index measures a culture’s acceptance of unequal power distribution. In high power-distance cultures — Malaysia, the Philippines, Mexico, Saudi Arabia — information from a hierarchical authority carries more weight than information presented without attribution. An AI tool that presents all information with equal authority, no source hierarchy, no institutional endorsement, is structurally misaligned with how these users evaluate trust.

In the Netherlands, Denmark, and Sweden — low power-distance cultures — the same flat presentation is expected. The tool that works in Amsterdam fails in Riyadh. Not because of a missing translation. Because the information architecture embeds a cultural assumption about authority that the development team never examined.

Trust signals. What makes a user trust a product differs structurally across cultures. In Germany, trust is built through technical specification, certification, and institutional endorsement. In Japan, trust is built through relational history and group consensus — the product must be endorsed by someone the user already trusts. In Brazil, trust is built through personal warmth and perceived relationship with the brand.

eBay learned this in Japan. In 2000, eBay entered the Japanese market with its American platform, functionally identical to the US version. They translated the interface. They adjusted the currency. They wore the costume. eBay demanded credit card payments in a market that preferred cash-based transactions. They offered a peer-to-peer trust model in a culture where trust flows through established relationships and institutional endorsement. Yahoo Japan Auctions, which had partnered with Softbank — a trusted local institution — captured ninety-five per cent of the market. eBay withdrew in 2002.

The translation was correct. The structure was wrong.

Formality registers. The distinction between formal and informal address is not a cosmetic preference. In German, the difference between Sie and du is a social contract. An AI tool that addresses a German executive as du in a professional context has violated a register expectation that has no equivalent in English. The violation is not “annoying.” It is disqualifying. The user does not think “the formality is wrong.” The user thinks “this tool does not understand my professional context.”

In French, the tu/vous distinction carries similar weight. In Japanese, there are multiple levels of formality — keigo, the honorific system, involves three distinct registers (teineigo, sonkeigo, kenjougo) that must be calibrated to the relationship between speaker and listener. An AI tool that defaults to casual Japanese in a business context is not making a style choice. It is making a social error.

The formality register is structural because it determines the entire tone of the interaction. It cannot be patched after the fact. It must be designed from the beginning — and designed differently for each culture.

Uncertainty tolerance. Hofstede’s uncertainty avoidance index measures a culture’s comfort with ambiguity. In Greece, Portugal, and Japan — high uncertainty-avoidance cultures — users want definitive answers. An AI tool that hedges with “This might be the case” or “There are several interpretations” triggers not nuance but distrust. The hedging reads as incompetence.

In Denmark and Singapore — low uncertainty-avoidance cultures — the same hedging reads as intellectual honesty. The same words, the same interface, the same product. Different cultural system. Different trust response.

This is not something a translator can fix. It is an interaction pattern that must be designed per culture.

The Costume Economy

The localisation industry has built a business model on costumes.

The Localization Industry Standards Association, before its dissolution in 2011, defined localisation as making a product “linguistically and culturally appropriate to the target locale where it will be used and sold.” The definition includes culture. The practice, overwhelmingly, does not.

CSA Research — the research arm of the localisation industry — found that 72.4 per cent of consumers are more likely to buy a product with information in their own language, and that 56.2 per cent consider the availability of information in their own language more important than price. These numbers are cited by every localisation vendor on earth. They are used to sell language services. They are not used to sell cultural adaptation — because cultural adaptation is harder to deliver, harder to price, and harder to measure on a spreadsheet.

The industry has optimised for what it can sell: translation, formatting, and surface compliance. The result is a global market of products that speak dozens of languages and understand approximately zero cultures.

The costume economy is efficient. It produces trackable deliverables. It generates progress reports (“47 of 52 localisation items complete — 90 per cent done”). It satisfies management. It fails users.

The ninety per cent completion rate represents ninety per cent of the surface. The structural adaptation — the part that determines whether users adopt the product — is not on the list. It cannot be on the list. Cultural adaptation is a system property, not a checklist item.

Why Structure Is Invisible

The development team does not see the structural assumptions because the assumptions match their own culture.

This is not negligence. It is the nature of cultural systems. Culture is invisible to those who share it. A fish does not see water. A development team in San Francisco does not see the cultural assumptions embedded in its product because those assumptions feel universal. Casual formality feels natural. Left-to-right layout feels natural. Egalitarian information presentation feels natural. Hedged confidence language feels natural.

They are not natural. They are cultural. They feel natural because the development team lives inside the culture that produced them.

Trompenaars described this as the fish-in-water problem in Riding the Waves of Culture: the most important aspects of culture are the ones that are most difficult to observe from inside, precisely because they constitute the medium in which everything else operates. You cannot examine the medium while swimming in it.

The consequence for product design: the structural assumptions of the development team become the structural defaults of the product. The product ships with a cultural fingerprint — casual, egalitarian, low-context, low-uncertainty-avoidance, individualist, left-to-right. This fingerprint is not labelled. It is not declared. It is not acknowledged. It is experienced by every user who does not share it as a vague but persistent foreignness — a feeling that “this product was not built for me.”

The feeling does not generate a bug report. It generates abandonment.

The Walmart Principle

In 1997, Walmart entered Germany. The world’s largest retailer, armed with a model that had conquered the United States, arrived in a market of eighty million consumers. They acquired two German retail chains. They applied the Walmart model. They wore the costume: German-language signage, local currency, local products on the shelves.

The structure was American.

Walmart greeters — employees stationed at the entrance to smile and welcome customers — were perceived as intrusive by German shoppers who value efficiency and privacy in retail interactions. The mandatory employee cheerleading sessions, a staple of Walmart’s American corporate culture, were considered bizarre and demeaning by German workers. The first head of German operations did not speak German and declared English the official management language. The company cycled through four CEOs in four years.

Walmart lost over one billion dollars and withdrew from Germany in 2006.

The surface was correct. The language was German. The currency was euros. The products were local. Every visible element was adapted. Every structural element — the greeting culture, the employee relations model, the management communication style, the assumptions about customer service, the retail philosophy — was American.

Walmart did not fail because of a bad translation. Walmart failed because of a structural misalignment between the American retail cultural system and the German retail cultural system. The costume was perfect. The body underneath did not fit.

The Surface-Structure Test

For any company deploying a product across cultural borders, the surface-structure test separates the costume from the adaptation.

Surface question: Can this change be made by a translator, a graphic designer, or a format conversion script?

If yes, it is surface. It is necessary. It is not sufficient.

Structure question: Does this change require understanding how the target culture perceives hierarchy, establishes trust, communicates uncertainty, relates to time, and navigates formality?

If yes, it is structural. It cannot be delegated to a translation team. It requires cultural knowledge — not about the language, but about the system in which the language operates.

The surface-structure test is not an abstraction. Applied to a specific product entering a specific market, it produces a concrete list of structural adaptations. Consider an AI customer service tool being deployed from the Netherlands to Japan.

Surface adaptations: translate to Japanese, adjust date and number formats, convert currency, use Japanese-style visual elements.

Structural adaptations: redesign the greeting sequence to establish relational context before transactional efficiency. Recalibrate the confidence language — eliminate hedging, present answers with authority. Implement hierarchical information presentation — attribute information to institutional sources. Replace the binary feedback mechanism (thumbs up / thumbs down) with an indirect evaluation method that does not require the user to make explicit negative judgments. Adjust response length — Japanese high uncertainty-avoidance users prefer comprehensive answers. Design the entire interaction flow to respect the high-context communication style — provide contextual information without being asked, because in high-context cultures, having to ask is itself a signal of system failure.

The surface adaptations take days. The structural adaptations take months. The surface adaptations are on the checklist. The structural adaptations require a redesign of the interaction architecture — and a team that understands the cultural system well enough to redesign it correctly.

Structural Adaptation as Design Discipline

Structural adaptation is not a sensitivity exercise. It is not a diversity initiative. It is a design discipline with frameworks, methods, and evaluation criteria.

The frameworks exist. Hofstede’s six dimensions provide a quantitative map of cultural variation across power distance, individualism, uncertainty avoidance, masculinity, long-term orientation, and indulgence. Trompenaars’ seven dimensions map universalism versus particularism, individualism versus communitarianism, specific versus diffuse relationships, neutral versus emotional expression, achievement versus ascription, sequential versus synchronic time orientation, and internal versus external control. Hall’s high-context/low-context spectrum maps the degree to which meaning is carried by explicit words versus implicit context.

These frameworks are not perfect. They generalise. They are based on national averages that obscure regional and individual variation. They are starting points, not destinations. They are, however, vastly superior to the alternative — which is no framework at all. Which is what most localisation projects use.

The methods exist. Cultural research — not desk research, not blog posts, but structured application of the frameworks to the specific product and market. Interaction pattern design — redesigning how the product communicates, not just what it says. Trust signal calibration — matching the product’s trust architecture to the culture’s trust model. Testing with cultural insiders — not with the localisation team, not with a cultural consultant, but with people who live in the target culture and use the product on real tasks.

The evaluation criteria exist. Not “is the translation correct?” but “does this tool feel like it was built for me?” The second question can only be answered by a user in the target culture. It cannot be answered by the development team. It cannot be answered by a checklist.

The discipline is harder than the costume. It is slower. It costs more upfront. It produces products that work — not in the abstract, not in the progress report, but in the hands of the user in Osaka, in Riyadh, in Munich, in Sao Paulo, who opens the product and makes a judgment, in two seconds, about whether this thing understands their world.

The AI Localisation Problem

The structural problem is acute in AI products.

A traditional software product has a fixed interface. The surface elements — buttons, labels, menus — can be translated. The structural elements — layout, hierarchy, interaction flow — can be redesigned per market, once.

An AI product generates dynamic content. Every response is new. Every response carries cultural assumptions — in its tone, its formality, its confidence level, its response length, its approach to uncertainty, its assumptions about the relationship between the tool and the user. These assumptions are embedded in the model, not in the interface. They cannot be fixed with a translation layer.

A chatbot that responds in fluent German with American casual register is a chatbot wearing a costume. The German is correct. The cultural performance is American. The user in Munich receives a grammatically perfect response that feels wrong in a way they cannot articulate — and that failure of articulation is precisely why structural misalignment is so dangerous. The user does not think “the register is wrong.” The user thinks “I don’t trust this tool.”

For AI products, structural adaptation means calibrating the model’s cultural behaviour — not just its language. The formality register, the confidence calibration, the response architecture, the relationship model between tool and user — all of these must be designed per cultural context. This is not a post-processing step. It is a design requirement that must be embedded from the beginning.

At Bluewaves, this is the work. Meridian, the cross-cultural design publication, exists because the problem is structural, not linguistic. Every Gizmo we build for a client operating across cultural borders is designed with the structural question first: not “what language does this user speak?” but “what cultural system does this user operate in?”

The language follows the structure. Never the other way around.

The Costume Removal

The costume is comfortable. It is familiar. It produces deliverables that management can track. It creates the feeling of progress without the substance of adaptation.

Removing the costume means accepting that cultural adaptation is design work, not translation work. It means investing in cultural research before a single string is translated. It means hiring people who understand the target culture at the system level — not as a list of dos and don’ts, but as an interconnected set of assumptions about how the world works. It means testing with people who live in the culture, not people who have read about it.

It means accepting that the product you built for one cultural context may need to be substantially redesigned — not retranslated, not reformatted, but redesigned — for another.

This is expensive. This is slow. This is the discipline.

The alternative is cheaper and faster. The alternative is a product that speaks twelve languages and understands none of them. A product dressed in the colours of a dozen cultures and built on the assumptions of one. A product wearing a costume and calling it localisation.

Most companies choose the costume. The ones that choose the structure are the ones whose products work in Tokyo, in Riyadh, in Munich, in Lagos. Not because the translation is correct. Because the architecture is right.

The costume is easy. The structure is the work. The distinction is everything.