Use cases · Selling across languages

How it works

You export the full set of policy and terms pages — returns, warranty, shipping, privacy — in every market language you publish. The large model reads them all together in one overnight batch and reasons about meaning across the whole set: where a returns window is 14 days in one language and 30 in another, where a liability sentence is softer in German than in Dutch, where a warranty carve-out appears in French but not Italian. This is not the same job as checking one translated page against its source. It compares every language against every other and reports where they no longer say the same thing. It runs on ml2, the machine with the bigger card, overnight rather than while you wait. The model that can hold your whole policy estate at once keeps its bulk in slow system memory and does only a little work per word — enough to finish a reading job by morning, not enough to answer someone sitting at the screen.

Data you need

The full policy and terms text in every market language, as text you can export — not screenshots, and not a CMS you cannot get plain text out of. It also helps to mark the differences that are meant to be there: a returns window that genuinely differs by country because local law requires it, a market that really does offer a longer guarantee. Without that list the model flags every divergence, including the deliberate ones, and the review time goes on dismissing them.

What to expect

It raises differences for a person to judge; it does not decide which are wrong. Expect two kinds of error. It will flag differences that are only stylistic — a clause reordered, a synonym chosen — as if they were drift, so a real meaning change and a harmless rephrase land in the same list. And it will miss drift that hides inside agreement: two pages that both say 30 days but attach different conditions to that deadline can read as matching when they are not. The sharpest failure is the deliberate difference read as a mistake — a returns window that is longer in one market on purpose, flagged as an inconsistency to "fix", which if acted on without thought would break a term you are legally required to offer. Treat the output as a reading aid, not a compliance check.

Where people stay involved

Whoever owns your policies reads the flag list and decides what is real drift and what is intended. A change to a legal term — a refund deadline, a warranty limit — is drafted and signed by that person, never edited automatically from the model's output. This finds candidates for a human to check. It does not certify that your policies are consistent or lawful, and a clean run is not sign-off.

• Translate your product catalogue with your own word list locked

  Draft catalogue translations into EU languages with your brand, material and category terms applied from your own glossary, ready for review.

• Let your support team chat with customers in any EU language

  Live translation between agent and customer in chat — the customer writes in their language, your agent reads and replies in their own.

• Draft replies to customer emails in the customer's language

  Reads a customer email in any language, pulls the relevant passages from your policies, and drafts a reply in their language for an agent to send.

• Read all your reviews, in every language, as one weekly digest

  Translates and tags every review from every market, then writes a short weekly digest in your language with translated quotes attached.