SaaS & procurement

The terrain

Enterprise SaaS localization runs on contradiction. The product ships weekly. The translation pipeline was designed for quarterly releases. Engineering uses string keys like invoice.approval.pending_review; the translator sees a blank field with no context. The legal team wants “submit” rendered as a term with contractual weight; the UX team wants it to feel like a button.

Procurement software adds a layer on top. The vocabulary (purchase order, requisition, approval workflow, three-way match, goods receipt) has specific legal meaning in different markets. “Invoice” in Germany is a Rechnung with specific regulatory requirements about what must appear on it. “Approve” in a Japanese procurement context carries a formality expectation that “承認する” handles differently than “許可する”. These are not translation preferences. They are product requirements.

The release cadence is the other pressure. A product shipping to 30 locales every two weeks cannot run each release through a four-day round-trip with a traditional LSP. The pipeline has to be fast by design, not fast by exception.

What we have learned

High-repetition UI copy (save, cancel, confirm, delete, filter) creates excellent translation memory leverage on SaaS products. After the first release, 40–60% of new strings in a standard product update are TM matches. That leverage compounds. By year two of a well-maintained TM, the marginal cost of a new locale version of a standard release is a small fraction of the first-release cost.

The long tail is harder. Terminology-dense sections (contract management, compliance reporting, approval hierarchies) don’t benefit from TM leverage because they change frequently and are never repeated verbatim. These sections need a different approach: termbase-injected LLM translation with human review at the term-boundary level, not segment level.

Register varies by market in ways that are not obvious from English. German enterprise software users expect formal register consistently. Informal language in a legal context is not charming, it is unprofessional. Japanese interfaces require honorific choices that are invisible in the English source. French legal vocabulary has Anglicism sensitivities that your generic MT output will violate.

A typical engagement

We start with a terminology audit. If there is an existing TM, we assess its consistency before we use it. A polluted TM is a liability, not an asset. We build or extend a termbase covering the product’s core vocabulary and legal terms, with context and forbidden-alternative columns.

From there, we design the pipeline: TM pre-translation, termbase-injected LLM translation for net-new segments, human review at the terminology boundary level. We validate locale formatting (numbers, dates, currencies) as a separate QA pass from linguistic review, because they fail in different places.

For ongoing work, we integrate with your CI/CD pipeline. String exports on merge, translation within the release window, delivery before the next deploy. We have done this at two-week cadences for SaaS products with 40,000+ active strings across 15 locales.