By Suzanne-Rose Griveau — Shortened version of her article published in MultiLingual (December 2025, pp. 21–24)
In the world of software-as-a-service (SaaS), speed is not a luxury — it’s survival. New features can drop daily; content flows from multiple teams in multiple formats; and localization teams must instantly deliver translations that are accurate, on-brand, and deployment-ready. AI and automation are now rewriting workflows, roles, and priorities, pushing teams to rethink processes rather than stack new tools. Below is a concise selection of the author’s key points and best practices — presented exactly as written in the original article.
Recognize That Governance Comes First
“There’s no localization team. If someone needs a translation, they post translation requests in a Slack channel called ‘translation,’ and product managers do the translating themselves in Phrase.” Without ownership, quality faltered. In one infamous case, “We used the French word collaborateurs — meaning ‘employees’ — and someone translated it into English as ‘collaborator.’ That triggered panic from the UK team.”
To stabilize quality, governance was introduced through:
- A Google Sheets glossary with a Chrome plugin
- Quarterly updates
- Accessible, approved terminology
Best Practice: Don’t let localization run on ad hoc goodwill. Define ownership, document standards, and centralize terminology before scaling up.
Automate Where It Hurts Least
Machine translation (MT) now often serves as the first step — a safety net if content must go live fast. As Teresa Toronjo explains, “We also apply MT as a first step, so there’s always a fallback if content needs to go live quickly. Translators then have 48 hours to review and edit the MT.”
Some teams go further, using AI-driven quality prediction scores to decide which segments need human review, allowing linguists to focus on high-impact work.
For another senior localization specialist, MT runs through XTM — DeepL for most European languages, Microsoft Translator for Asian ones — but “several steps still require manual work: pre-processing, post-processing, tagging, quality assurance (QA) after review, and final implementation.”
Best Practice: Let automation handle low-risk, high-volume segments.
Make AI Part of the Plumbing
At Yango and InDrive, Yana Kolesnikova built pipelines that batch-schedule user interface (UI) translations, integrate Crowdin with GitHub, and run AI-assisted QA to catch issues. “It became clear that we either had to hire more people or automate parts of the process to reduce repetitive effort,” she said.
AI now supports:
- Source preparation to rewrite unclear keys
- Quality control through inconsistency detection
- Tone optimization for different markets
Best Practice: Integrate AI into your infrastructure so it becomes a stable, invisible part of your localization pipeline.
Approach Localization as a Design Feature
For some teams, localization is no longer the final step; it’s embedded in product design from day one. At Yango, for instance, the localization team works closely with developers, designers, and product managers to ensure that every feature is designed with multilingual markets in mind.
- Glossaries integrated directly into CAT tools
- Annotated Figma screenshots for visual context
- Continuous localization pipelines batching strings twice a week
- Automation that detects mismatched variables, incorrect characters, or Cyrillic letters
Tone of voice is also treated as a design parameter.
“I’d push for a culture that sees localization as part of the product design, not a final step.”
Best Practice: Treat localization as a product feature. Involve linguists from the design stage, provide full visual context, and embed glossaries into tools to ensure consistency, relevance, and efficiency from the first line of code.
Review by Risk, Not by Habit
Treating every string equally is a fast track to burnout. Leading teams prioritize review based on business impact, visibility, and legal stakes. Toronjo applies a similar system at Malt:
“Regarding quality, I care deeply — but only as far as it impacts the user experience. ‘Best quality’ depends on the content type; a help article doesn’t need the same quality as a legal document or a product feature. We have a tiered system. Tier 1 features are either AI-related, like our AI-powered search, or revenue-related, such as quote approval workflows. These get prioritized in our quality checks.”
Best Practice: Map your content to risk tiers.
Start Scaling Before You Grow
Every expert interviewed warned against waiting until growth breaks the system. One senior localization specialist summed it up: “Unfortunately, the localization stack is fragmented.”
Marketing runs a content management system (CMS) integrated with translation workflows; technical communication uses a separate CMS without integration. With languages split between in-house staff and language service providers (LSPs), “internal teams might update a file while the vendor is still working from an earlier version, leading to inconsistencies.”
Her proof of concept boiled the essentials down to four points: automation, cultural adaptation, early involvement in product development, and a mindset that sees localization as strategic.
Best Practice: Future-proof now. Design processes that can handle twice the volume you currently have — without doubling the chaos.
Design for Internationalization Early
Scaling localization starts with strong internationalization — designing software so it can support multiple languages and cultural formats without rework. This means separating translatable strings from code, using variables instead of hardcoding text, and planning for differences in date formats, currencies, and writing systems.
Toronjo saw firsthand how neglecting this step inflated work:
- “We had 15 languages — but many were ‘fake.’”
- “For example, in Spain, some users preferred English UI but needed Spanish-specific information like local contact emails or legal disclaimers.”
- “So we duplicated the English strings and created a language called ‘English (Spain),’ just to localize 250 keys manually.”
By working with engineers to replace duplicated strings with dynamic variables, she deleted 10 “fake” languages, reducing the string count from 1.5 million to 600,000.
Best Practice: Bake internationalization into development from the start.
SUZANNE-ROSE GRIVEAU is a global content manager and former certified German teacher. She recently completed her master’s degree in Technical communication and Localization (TCLoc) at the University of Strasbourg.
