Glopal · Ecommerce localisation
Ecommerce translation: what actually drives revenue
Translation changes your words. Localisation changes the buying experience, and only the second one produces growth.
In live 50/50 A/B tests across 40+ merchants and 50+ site and country combinations, translated stores lifted conversion by an average of 40% at launch, individual results 15% to 100%.
See which markets pay backboutique.example.de / de-delive
Runner Low, Damen
129,00 €
inkl. MwSt. und Zoll
boutique.example.jp / ja-jp日本
ランナーロー ・ レディース
¥19,800
税込・関税込
boutique.example.ae / ar-aeالإمارات
حذاء رنر لو للسيدات
480 د.إ
شامل الضريبة والرسوم
Trusted by brands selling in 200+ markets
The short version
01Translation quality has been commoditised. That is precisely why translating alone stops producing growth.
02The measurable lift comes at launch and it is large: 40% on average in controlled tests, with a wide spread.
03The language-preference statistic most articles quote is from a 2007 study. The current number is 76%.
04Native platform tools genuinely work, up to a documented boundary. Knowing where it sits saves a wasted migration.
05The hard part is not launch. It is the fifty product drops after launch.
06Producing a good translation is nearly solved. Finding the bad one is not.
07Your default markets are probably your least efficient ones.
Evidence
What does the data actually say?
Two things get conflated: what shoppers say they prefer, and what changes when a store is translated. The first is survey data. The second is an experiment, and far rarer.
The spread is the honest part: a brand whose category is already understood sits near the bottom of that range, one whose product needs explaining near the top.
The language-preference statistic most articles quote is from a 2007 study. The current number is 76%.CSA Research, Can't Read Won't Buy: B2C, 2020 edition. 8,709 consumers, 29 countries.
One correction
The “72.1% of consumers” figure quoted across this topic is from the 2007 Common Sense Advisory edition, not a current study.
Primary result · first-party
+40%
Average conversion lift at launch in controlled 50/50 A/B tests.
15%40% average100%
40+ merchants, 50+ site and country combinations, results 15% to 100%. A launch measurement against an untranslated experience in the same market, not a permanent multiplier.
Glopal platform data, 2026.
Definitions
Translation, or localisation?
Language gets the shopper to the cart. Everything after it decides whether they complete.
Translation covers
- Product copy and descriptions
- Category pages and navigation
- Transactional email
- Policy pages
Localisation adds
- Everything in translation, plus
- Currency and price formatting
- Sizing, address and phone conventions
- Payment methods buyers recognise
- Delivery promises in local terms
- Duties and landed cost, before checkout
The failure runs both ways: local currency and payment bolted onto machine-made copy fails just as reliably. Premium brands feel it hardest, because they sell an experience rather than a product. Shoppers detect rushed translation immediately and discount the brand for it.
Four routes
How should you choose an approach?
Four genuinely different products, not tiers. Most merchants choose on price per word: the one axis on which all four look alike, and the one that predicts nothing.
The evaluation question is who owns the ongoing work after launch.
| Route | Best for | Who owns the work after launch | Real boundary |
|---|---|---|---|
| Native platform tools | Testing one or two markets with no budget | Your team, manually | Automatic translation is capped; outside content uncovered |
| Translation app or widget | Indexable multilingual pages, no dev work | Your team runs glossaries, reviews strings, manages updates | Quality control and upkeep sit with you |
| Translation agency | Brand-critical copy in a few languages | Nobody, until you commission the next batch | An agency translates a snapshot; your catalogue keeps moving |
| Managed localisation platform | Multi-market selling where language, currency, duties and checkout must all work | The provider | Images and video stay yours; commitment is not self-serve |
Scroll the table sideways to see every column →
That third column predicts whether the project still works in twelve months, and nobody raises it during procurement. Translating a store is a project. Running one is not.
Where the native option stops
Translate & Adapt auto-translates two languages free from a set of 21, and Shopify provides 33 pre-translated checkout languages. Real capabilities; do not pay to replicate them. Past two languages translations are added by hand, URL slugs and tags are untranslated, and anything outside Shopify's own objects is uncovered.
Related reading
Search
Will translated pages get indexed?
Yes, when they are served server-side on their own URLs. Established apps including Weglot do exactly that and their pages get indexed. Anyone claiming otherwise is selling against a weakness that is not there.
01
Dedicated URL
02
Server rendered
03
hreflang, canonical, sitemap
04
Indexed
The real failure mode is narrower: tools that switch language only in the browser leave one page to crawl, and free tiers are sometimes configured that way. Ask whether your French page is a server-rendered URL with hreflang on the plan you are buying.
Then ask the harder question, because it is a separate commercial problem. Local keyword demand decides whether those indexed pages earn traffic or merely exist, and a literal translation of your English keyword set is not what anyone types in Munich. That is where providers actually differ.
The product
How does the localisation system work?
Machine translation is confidently wrong about register. Human translation cannot keep up with a weekly catalogue. Both, in sequence.
The loop
Coverage first, quality continuously
- Source changes · New or edited content is detected
- Immediate translation · Served with the page. Coverage, not quality
- Context-aware improvement · Style guide, glossary, do-not-translate list
- Continuous upkeep · Every later page follows the same path
The design choice most tools get wrong is that a shopper never waits for the best version to exist. Tags, prices and currencies stay untouched. Without this running a store decays quietly: products land, collections retire, and a French shopper meets a half-English page.
Proofreading
Finding the bad translation
Producing a good translation is close to solved. Finding the one wrong sentence in ten thousand, when nobody on your team reads the language, is not. Proofreading flags the likely errors with a suggested correction, so reviewers are pointed at candidates instead of reading everything. What that changes is not only speed: corrections stop being disposable, because every edit is reused wherever that wording appears. Quality compounds rather than resetting each season.
Brand voice
What makes it sound like you
Five inputs, agreed before launch and enforced automatically: a tone-of-voice and style guide, a do-not-translate list for brand and collection names, a glossary, a blocklist of words that must never appear in a market, and in-market reference material. Premium brands get a dedicated team maintaining it per language.
Source · EN
Crafted from brushed Italian suede.
Live · FR
Fabriqué en cuir suédé italien brossé.
Suggested · glossary match
gratté, approved for this brand, reused in 214 places
Measurement
How quality is scored
Automated daily checks flag wording that has drifted from versions already approved for your brand. Separately, native linguists score representative 1,500-word samples out of 100 using MQM, weighting four error types: terminology, accuracy, linguistic conventions and style, each graded neutral, minor, major or critical. Three independent translators score every sample, so the result reflects the language rather than one person's taste.
Review scope
How much human review
Review is scoped rather than spread evenly. The lightest covers the homepage and the key static elements of your main templates; the fullest is defined page by page with you and includes product names. Launch runs from roughly a week to a month. Right-to-left is the hardest case, and we run it in production on a live flagship storefront.
Yours to own
- Brand and terminology decisions
- Campaign imagery and video
- Market and pricing strategy
- Commercial and legal positions
- Judging whether it sounds like you
Absorbed
What the system absorbs
- Chasing what changed
- Re-running strings
- Reconciling glossaries
- Finding the half-translated page
Two honest limits. Localisation removes friction from demand that exists; it does not create it. And paid converts from launch while organic builds over six to twelve months, so a provider dating a ranking cannot know it.
Evidence
What does a real launch look like?
The honest question behind “how does it work” is how long until you know whether it worked.
| Brand | Markets | Time to launch | Result |
|---|---|---|---|
| Sabre Paris | 5 | 2 weeks | German revenue doubled |
| Rodier | 8 | 2 months | Organic revenue +22% |
| Mac Duggal | 28 | Under 3 months | International DTC from 8% to 16%, including a full right-to-left Arabic storefront |
Scroll the table sideways to see every column →
Fashion and premium brands, where a shopper's judgment of translation quality transfers onto the brand. Across all three, launch is measured in weeks and the compounding shows up in organic revenue.
Sequencing
Which markets first?
Not the ones you are about to pick. Germany, France and the UK is the shortlist everyone follows, which is what makes it expensive.
The Western Big Three converts better than any other cluster, at 1.72%, and is still the least efficient.
Its funnel is not broken and its storefronts are not the problem. Its score fell from 1.31 to 0.67 across five quarters because cost per click moved from 15% below the global average to 22% above it while same-store growth turned negative. Conversion and efficiency are different things. Localising a German store may lift conversion, exactly as the A/B data says. It will not make German traffic cheap.
Cluster averages also hide the two most interesting markets. Inside Southern Europe and Benelux, Spain grew 34.2% year on year, the fastest individual market in the panel, with Italy behind it at 27.6%. Inside the Anglosphere, whose 1.04 average reads as flat, Australia grew 28.4% while Canada contracted 16.9%. Both are underweighted by most operators, which is why attention there is still cheap.
+34.2%
Spain, year on year
The fastest individual market in the panel, with Italy behind it at 27.6%.
+28.4%
Australia, year on year
Inside the same 1.04 Anglosphere average, Canada contracted 16.9%.
Growth-per-Dollar Index · Q1 2026
1.73×
Top-to-bottom gap in growth per dollar, the widest in the dataset.
Growth-per-Dollar Index, Q1 2026. Same-store GMV growth divided by cost per click, normalised so the global average is 1.00. 54 direct-to-consumer merchants, 17 countries, five consecutive quarters, observed transactional data.
| Cluster | Same-store GMV YoY | Cost per click | CPC vs global | Growth per dollar |
|---|---|---|---|---|
| Nordics and Ireland (IE, SE, NO, DK, FI) | +9.0% | $0.367 | 18% below | 1.16, highest |
| Southern Europe and Benelux (IT, ES, NL, BE, CH) | +30.8% | $0.452 | roughly flat | 1.13 |
| Anglosphere (AU, CA, US, NZ) | +0.1% | $0.374 | 7% below | 1.04 |
| Western Big Three (DE, FR, GB) | −5.1% | $0.554 | 22% above | 0.67, the only reading below 1.00 |
Scroll the table sideways to see every column →
So localise a market when three things hold at once: demand exists, attention is affordable, and language is suppressing conversion. Rarely the biggest economy on your list.
The opportunity now is the markets that didn't make the old shortlist.
Questions
Frequently asked
How many languages should I start with?
Two or three done properly beats eight done thinly, and start where your analytics show demand. Our Q1 2026 benchmark found Germany, France and the UK the least dollar-efficient of four clusters, at 22% above global cost per click with negative same-store growth.
Will translating my store hurt my existing SEO?
Not if implemented correctly. Each language version needs its own crawlable URL, hreflang, canonicals and sitemap inclusion, and your source-language pages are unaffected. Risk enters during migrations, when URL structures change.
How do you know the translation is any good?
Automated daily checks flag wording that has drifted from versions already approved for your brand. Separately, native linguists score 1,500-word samples out of 100 using MQM, weighting terminology, accuracy, linguistic conventions and style. Three independent translators score each sample.
Do product images get translated?
No. Text inside images, banners and video is untouched, so campaign visuals stay yours per market. Brands with strict art direction generally prefer that.
Where to go next
Which markets actually pay back
The index ranks all four clusters, shows the five-quarter trajectory behind the 1.73 times spread, and gives you the CPC and conversion bands to test your own campaigns against. Free.
Get the Q1 2026 indexOr see your own store localised, priced in euros with duty and tax included
Growth per dollar · Q1 2026
Nordics and Ireland1.16
Southern Europe and Benelux1.13
Anglosphere1.04
Western Big Three0.67
Same-store GMV growth divided by cost per click, normalised so the global average is 1.00.