The fastest-growing AI investment in cross-border retail is translation, and it is the wrong priority. A Passport survey of brand leaders found that 84% have deployed AI for marketing, personalisation, and customer support, while only a third have applied it to inventory, cross-border logistics, or compliance. Brands are automating the layer they can see — the product listing, the support script, the ad copy — while the operational failures that suppress conversion remain untouched.

Machine translation is booming on its own terms. DeepL’s revenue reached $185.2 million in 2024, up 31% year-on-year, and the company is reportedly weighing a $5 billion IPO. The broader AI translation market is projected to hit $8.93 billion by 2030 at a 24.8% compound annual growth rate. The capital is deserved; machine translation works and it scales. The problem is what it lets brands skip.

DeepL’s own Borderless Business report found that 83% of enterprises have not transitioned to modern language AI, and 68% still run multilingual workflows built for a previous era. Enterprise content volume has grown 50% since 2023. The workflows absorbing that volume handle text, not operations. A product listing translated into Japanese still ships in American sizing, still charges in dollars unless a separate pipeline exists, still routes returns through a process designed for the domestic market. The Stord State of AI 2026 report confirms the pattern across retail: 88% of organisations use AI in at least one function, but only 7% have reached full-scale implementation, and 62% remain in early experimentation.

Sizing is the conversion problem translation cannot reach. Fit issues cause 53% of apparel returns globally, according to Prime AI, and the problem compounds across borders: Asian sizes run one to two sizes smaller than American equivalents, and European cuts are slimmer and measured in centimetres. Bold Metrics, a sizing technology firm, reported a 3.8x conversion lift among European shoppers who used its fit recommendation tool. PwC research, as cited by Bold Metrics, found that 32% of consumers stop buying from a brand after a single bad experience; ill-fitting international orders qualify. A brand that translates its product page into six languages but ships one country’s sizing has not localised; it has decorated.

Payment localisation shows a comparable gap. The Passport survey ranked enhanced localisation fourth among brand priorities for 2026, behind delivery speed, market expansion, and shipping costs. Returns widen the divide: cross-border returns add weeks to handling timelines compared to domestic orders at considerably higher cost, and cross-border return rates average 25% across categories, running higher for apparel and footwear. In a 2021 survey, 73% of shoppers in the US, Germany, Australia, and the UK expected cross-border returns to be difficult before they placed an order. That expectation is priced into the decision to buy.

Translation is the AI investment brands make because it is legible, not because it is load-bearing.

The strongest defence of translation-first is its cost structure. Machine translation deploys cheaply, scales instantly, and produces clean before-and-after conversion lifts on product pages — the kind of A/B test that justifies a budget. The counter-argument holds if translation unlocks enough incremental revenue to fund deeper localisation downstream. In practice, it rarely does. The measurability of translation absorbs the optimisation cycle: brands test what they can test, and sizing infrastructure, payment-method expansion, and return-policy redesign do not lend themselves to rapid iteration. Translation succeeds on its own terms while starving the next investment.

If brands continue to allocate cross-border AI spending to the linguistic layer while treating sizing, payment, and returns as manual operations, the conversion gap between domestic and international shoppers will persist even as product pages read fluently in thirty languages. A shopper in Seoul who reads the listing in Korean but faces an unfamiliar checkout, an American size chart, and a return process routed through a US warehouse will close the tab. The question for cross-border retail has never been whether AI can translate a listing. It is whether the order works after the words do.