UI localisation: designing one B2B interface for several languages

Design one B2B interface for several languages: room for text expansion, layouts that flex, a glossary, CLDR formats, right-to-left and pseudo-localisation.

  • Design Systems
  • B2B SaaS and workflow software
  • B2B SaaS & Workflow Software
Insights

Design one B2B interface for several languages: room for text expansion, layouts that flex, a glossary, CLDR formats, right-to-left and pseudo-localisation. To make one B2B interface work in several languages, design it to change before anyone translates it.

To make one B2B interface work in several languages, design it to change before anyone translates it. Leave room for short labels to double or triple in length, let layouts reflow rather than truncate, keep sentences whole, take formats from locale data, use start and end instead of left and right, cover every script in your fonts, and test each build with pseudo-localisation.

Nearly all of this lives in the design system. If its components and tokens assume English, each new language is a redesign. If they assume change, a new language is a translation job.

This guide is for product managers and design leads taking a B2B product into new markets. It covers interface design only, not translation vendors, machine-translation quality or marketing-site localisation.

Localisation is a design system decision

It is tempting to treat localisation as a content task that starts when the strings go out for translation. By then, the expensive decisions have been made: a button with a fixed width, a table header that truncates at 12 characters, a margin set to the left, a date assembled from three variables, a sentence glued together from fragments. Each one is cheap to get right in a component and costly to find across hundreds of screens.

That is why we treat localisation as part of interaction and UI design from the first component, not a phase at the end. The argument holds for layout, direction, type and formatting, which the design system can solve once for every screen. It holds less well for language itself: terminology, string authoring and translator context belong to product and engineering as much as to design. A good design system makes those easier, but it cannot write a glossary for you.


How much room to leave for text expansion

The W3C's article on text size in translation says text translated from English will normally expand, and that the shorter the string, the more it grows. It reproduces IBM's averages for English into European languages: a string of up to 10 characters can reach 200 to 300 per cent of its English length, 11 to 20 characters 180 to 200 per cent, and anything over 70 characters around 130 per cent. The same article shows the Flickr label "views" running to three times its length in Italian.

Microsoft's pseudo-localisation guidance, used in the test recipe below, gives a working rule: lengthen English text by 40 per cent when testing, while noting that real translations can occasionally be 200 or even 400 per cent longer, and that one or two word strings grow proportionally more than longer ones.

Three complications matter for a B2B interface:

  • Compound words. German, Dutch and Finnish join words into one long word. The W3C's example turns "Input processing features" into "Eingabeverarbeitungsfunktionen", which will not wrap on its own the way the English does.
  • Wider characters. Chinese, Japanese and Korean characters take more horizontal space each, so a translation with fewer characters can still be wider. The W3C's example is "desktop" becoming デスクトップ.
  • Taller lines. The W3C lists Arabic, Chinese, Devanagari, Japanese, Korean and Tibetan among the scripts that need more height than Latin text, and notes that Thai is set with more space between lines than Latin text, so text grows downwards as well as sideways.

Translations can also be shorter, which leaves awkward gaps if a layout was tuned to the English length. And the W3C warns against abbreviating to make text fit: many languages cannot abbreviate the same way.

A practical rule: size components to their content with minimum widths, never fixed widths, and design the short labels (buttons, tabs, column headers, chips) as if they will double.


What breaks first, and how to fix it

The elements that break first are the small, constrained ones, because short strings grow most and sit in the tightest spaces.

ElementWhat breaksFix in the design system
Buttons and button groupsLabels clip, wrap unevenly or push the group off screenSize to content with a minimum width; let groups wrap or stack; no fixed widths
Tabs and top navigationItems overflow; long compound words do not wrapAn overflow menu for items that do not fit; hyphenation on (it needs the page's lang attribute, per CSS Text Level 3); a vertical navigation option
Form labels beside fieldsLabels squeezed into a fixed columnLabels above fields, so they can grow without shrinking the input
Table headers and columnsHeaders truncate; numbers misalign once separators changeHeaders that wrap to two lines; numbers formatted by locale and aligned to the end edge
ChartsAxis labels and legends overflow; dates and numbers in the wrong formatLegends that wrap; locale formatting on axes and tooltips; a documented rule for time axes in right-to-left layouts
Status chips and badgesText clipped inside a fixed pillPills that grow with text; the full text on hover where space is truly fixed
Sentences with values in themFragments in the wrong order, wrong pluralsWhole sentences with named placeholders and plural rules (see below)
Emails, exports and notificationsBuilt outside the interface, so never localised or never testedThe same string files, formatting functions and tests as the interface

Truncation is a last resort, not a layout strategy. When it is unavoidable, the full text must still be reachable.


Terminology, glossary and strings

B2B products live on precise terms: workspace, account, seat, approval, ledger. If those terms drift between screens in English, they will drift further in each translation. Give the product its own glossary, owned by the product team, that lists:

  • each domain term, its definition and where it appears;
  • terms that are never translated, such as product names and feature names you treat as brands;
  • the approved translation for each language, and who approved it.

How strings are written decides whether translators can do their job. Microsoft's guidance on message formatting sets out the rules:

  • Never concatenate. Write each sentence in full. Word order differs between languages, and nouns in German, French and Russian carry a gender that changes the words around them, so fragments cannot be translated reliably.
  • Use named placeholders, such as {startTime} rather than %s, so translators can see what each value is and move it. Microsoft's example is "Selected {0} out of {1}", where a Japanese translation puts the second value first.
  • Handle plurals with plural rules, not an added "s". The CLDR plural rules give English two categories (one, other), Polish and Russian four, Arabic six and Japanese one.
  • Do not reuse a string in two contexts. Microsoft's example is "Bookmark", which is a noun in one place and a verb in another and may translate differently.
  • Give translators context for short strings and for every placeholder, including the range of values it can take.

Dates, numbers, currency, names and addresses

Never hand-build a format. The Unicode CLDR (Common Locale Data Repository) holds the formats for dates, numbers, currencies, units and plurals, and is built into all modern operating systems and browsers; its latest release, CLDR 49, came out on 22 October 2026. On the web, the ECMAScript Internationalization API (Intl) exposes this data. Formatting 5 November 2026 and the number 1234567.89 with a current browser's locale data gives:

LocaleShort dateNumber
English (US)11/5/261,234,567.89
English (UK)05/11/20261,234,567.89
German (Germany)05.11.261.234.567,89
French (France)05/11/20261 234 567,89
English (India)05/11/2612,34,567.89

The same date reads as 11 May in one market and 5 November in another, and India groups digits in lakhs. Currency symbols move too: the same locale data puts the euro sign before the amount in English and after it in German and French. For a B2B product, the rules that follow are:

  • format every date, number, currency amount, percentage and unit through locale-aware functions, including in charts, exports and emails;
  • keep the locale (language and formats) separate from the currency, since a user who reads German may be billed in another country's currency;
  • even a percentage is locale-specific: Microsoft notes that French puts a non-breaking space before the % sign and Turkish puts the sign before the number.

Names and addresses break forms. The W3C's guidance on personal names around the world advises a single full-name field where possible, no assumption that the given name comes first (Chinese, Japanese, Korean and Hungarian put the family name first), no required family name, and fields long enough for long names with spaces, hyphens and apostrophes. For addresses, the fields and their order vary by country; Google's libaddressinput, which powers address entry in Android and Chromium, holds per-country address metadata you can build a form from, rather than reshaping a US form for every market.


Right-to-left scripts

Arabic and Hebrew read right to left, and the interface should mirror with them. The groundwork is technical but it starts in design:

  • Set direction in markup. The HTML standard strongly encourages the dir attribute rather than CSS for text direction. The W3C Internationalization guidance on the dir attribute adds the practical uses: dir="rtl" on the html element for the page, and dir="auto" on forms and inserted text such as user input whose direction you cannot know in advance.
  • Use logical directions everywhere. CSS Logical Properties let layout follow start and end rather than left and right, so a correctly built page mirrors without a second stylesheet. The same must be true of design tokens and of the design files: "padding-start", not "padding-left".

What mirrors and what does not is a design decision, and the major guidelines agree on most of it. Cloudscape, the open-source design system built for AWS products, mirrors all layout and text and any element that shows direction, progress or the passage of time, but not untranslated text, code, telephone numbers, email addresses, or icons of physical objects such as a magnifying glass. Apple's right-to-left guidelines add detail a B2B product needs:

  • the digits inside a number never reverse, though a sequence of numbers that shows progress, such as a slider's scale, does;
  • Hebrew uses Western Arabic numerals, while Arabic may use Western or Eastern Arabic numerals depending on the country;
  • a paragraph aligns to its own language, so an English paragraph inside an Arabic interface stays left-aligned;
  • a control that means a physical direction, "to the right", keeps pointing right;
  • Arabic and Hebrew have no capital letters, so beside all-capitals Latin text they can look small; Apple suggests about 2 points larger.

Data-heavy screens need explicit rules the guidelines leave open, such as whether a chart's time axis runs right to left. Decide them once in the design system and document them. Then have a native speaker of each right-to-left language review the result, which Cloudscape also recommends.


Fonts and scripts

A font that covers English may cover nothing else. Before adding a language, check that the interface typeface, or a defined fallback, covers its script in every weight you use; missing glyphs appear as empty boxes. The Noto font collection covers more than 150 writing systems and more than 1,000 languages, so it can serve as the fallback for scripts your brand typeface does not support.

Type tokens need per-script values. Because the scripts listed earlier need more vertical room, line height set for Latin text will clip or crowd Thai, Devanagari or Arabic. Avoid using capitals as the only form of emphasis, because Arabic, Hebrew and the Chinese, Japanese and Korean scripts have none.


Testing with pseudo-localisation: a recipe

Pseudo-localisation fakes a translation so you can find layout and string problems before paying for a real one, and keep finding them on every build. A recipe based on Microsoft's method:

  1. Add a pseudo-locale to development and staging. ICU defines en-XA (accented) and en-XB (bidirectional); Windows defines qps-ploc (base), qps-plocm (mirrored) and qps-ploca (East Asian).
  2. Transform every string automatically. Swap letters for accented ones, lengthen each string by at least 40 per cent (more for short labels), pad with characters from Greek, Cyrillic and Asian scripts, and wrap each string in delimiters such as [!!! ... !!!].
  3. Walk the critical flows. Navigation, settings, the densest table, the longest form, charts, empty and error states, emails and exports.
  4. Read the symptoms. A missing closing delimiter means truncation. Several delimited fragments in one sentence mean concatenation. Plain English text means a hard-coded string. Empty boxes mean a font gap. Overlaps mean fixed sizes.
  5. Repeat with the mirrored locale to check right-to-left layout, icons and alignment.
  6. Switch region formats separately, for example to German, Indian English and Arabic, to check dates, numbers and currency in tables and charts.
  7. Keep it running. Capture screenshots of key screens in each pseudo-locale on every release so regressions show up in review.

Pseudo-localisation does not replace review of the real translation in context; Microsoft says as much. It removes the mechanical failures so that review can focus on language.


What the design system must provide: a checklist

If localisation is a design system decision, these are the parts of the design system that carry it:

  • Components that size to content: minimum widths rather than fixed widths, a wrapping rule and a truncation rule for each component, and overflow patterns for tabs and navigation.
  • Direction-aware tokens: spacing, alignment and position named by start and end, so mirroring is a setting rather than a redesign. Our guide to design tokens explains why tokens are where these decisions belong.
  • Icons marked for mirroring: each directional icon flagged to mirror or not, following one documented rule.
  • Type tokens per script: font stacks with fallbacks, and line heights for each script you support.
  • Formatting, not formats: shared functions for dates, times, numbers, currency, units and plurals, used by every component, chart and export.
  • Form patterns for names and addresses that work across countries.
  • String rules: whole sentences, named placeholders, no reused strings, and context notes for translators, alongside the product's glossary.
  • Test locales built in: pseudo-locales and a right-to-left locale available in the component library and in staging, so every new component is checked before it ships.

With these in place, adding a language is mostly translation and review. Without them, each language reopens design decisions across the whole product.

About the authors

  • Rishabh Choudhary

    Product Designer

    Rishabh Choudhary is a Product Designer at Tcules who writes about design systems and about designing for people the average interface leaves out. Rishabh's articles look at ten age-inclusive products, inclusive design for older and disabled people, eLearning platforms from the side of both learners and administrators, and software for lab and life-science teams. On design systems, the pieces cover components that several brands can reuse and B2B SaaS design systems worth studying. Rishabh has also co-written on fintech UX, healthcare UX and choosing a fintech design agency.

    More from Rishabh
  • Yash Gidwani

    Product Designer

    Yash Gidwani is a Product Designer at Tcules who writes about getting people into a SaaS product and keeping them there. Yash's guides cover SaaS onboarding strategies, onboarding and engagement software, and redesigning a live product without upsetting the people who use it. Other articles go into specific workflows: lending and loan origination apps, insurance quoting and claims, staffing and recruitment software, and a search-first create flow that stops duplicate records. Yash has also co-written the SaaS UX audit checklist and a 20-question checklist for legacy modernisation readiness.

    More from Yash

Planning a product in more than one language?

Before your next language ships, we can check what your interface needs.