Weekly UI batches for an AI/SaaS platform: zero-string-drift across 18 months

The client

A mid-market AI/SaaS platform expanding into the Czech and Slovak markets. Web-based product, dense in-app copy, frequent release cadence. The client had no internal localization function — strings were “owned” by a product manager who had four other responsibilities.

The challenge

Every release sprint added or modified UI strings. Onboarding flows, error messages, in-app notifications, help center articles. The client had attempted machine translation with light review but was seeing customer support tickets that traced back to clumsy or ambiguous Czech phrasing — particularly in error states and onboarding microcopy where tone matters most.

The work had no consistent owner, and the client knew it. They needed a single point of contact who would build institutional knowledge of the product, not just deliver translated strings.

The approach

I proposed a weekly batch cadence with three commitments: (1) every batch lands within 36 hours of receipt, (2) I maintain the glossary and the TM, (3) I read the strings in context via the staging environment, not just as flat segments.

Week 1 (kickoff). Audit of existing translations. Built a 320-term glossary covering product nomenclature, action verbs, and locale conventions (formal “you” register; date and currency formatting; capitalization rules). Documented a one-page style guide.

Steady state. Each Monday: source strings exported as XLIFF from the client’s TMS, delivered to me by 10:00 CET. Tuesday EOD: translations returned, with a per-batch change log noting any source-side issues and any new terminology added to the glossary. Wednesday: client engineering pushes to staging; I review in context and submit corrections if anything reads wrong in situ.

Quarterly. A 90-minute video call with the client’s product team. We review what landed, what got escalated, and any platform-level patterns in the strings (e.g., “we’re adding a lot of permission-related copy this quarter — let’s preempt the terminology discussion”).

Tools used

The client’s TMS (Smartling), memoQ for offline review of larger batches, Loom for asynchronous walkthroughs of contextual issues, a private Notion glossary mirrored to the TMS termbase, Resend for batch delivery notifications.

Outcome

Across 78 release batches over 18 months, zero terminology regressions reported by the client’s QA team. Czech and Slovak signup-completion rates landed within 3% of the English-language baseline — the client’s previous expectation had been a 10–15% drag on translated locales.

The engagement evolved: starting in month 9, the client began routing all loc-related questions through me — not just translation requests, but also questions about i18n architecture, locale formatting in their date pickers, and best practices for internationalization in their React codebase. By month 12, we converted the project-based engagement into a monthly retainer with expanded scope.

This is the canonical example of a Tier B engagement evolving into a Tier C embedded engagement.