Shipped by Lokalise − Autumn 2025

Full transcript

Welcome and introductions

Marta (host): Welcome, everyone. The webinar today is called Shipped by Lokalise — the Autumn 2025 edition. Our agenda is simple: first I'll explain what Shipped by Lokalise is. It's a new series we're starting, and today is the first edition. We're bringing you the new things we've been developing and building in product at Lokalise, plus some sneak peeks of what's coming. Across the webinar you'll see some demos, and at the end we'll do a Q&A.

Our speaker today is Luke, Principal Product Manager at Lokalise. He has many years of experience shaping products across different teams, has led B2B SaaS teams at startups and scaleups, and has turned ambiguous problems into shipped value. At Lokalise, he oversees product delivery across the product teams, aligning strategy and execution to deliver additional value for customers.

Luke: Welcome, everybody. I'm Luke, Principal Product Manager at Lokalise. Thank you for joining our quarterly product updates. Through these Shipped by Lokalise sessions, we aim to keep consistent touchpoints with our customers and share updates on a regular basis. Today I'll walk you through the key features we released in Q3, show short demos of each, and close with a sneak peek into what's coming next. Stay tuned especially if scaling your localization to more content across other departments interests you.

We'll cover four main topics from Q3: our new Figma Buzz integration, self-serve custom AI models, AI translation quality and smart routing, workflow improvements, plus some new integrations and file format support.

Where Lokalise is investing

Luke: We want our AI solutions to offer high-quality translations you can trust. We started by launching our own AI orchestrator to deliver the highest quality among the various LLMs and technologies out there, layered on automatic translation scoring, and are now working on custom AI profiles and more. We track metrics like acceptance rates closely, and the results show for themselves: our efforts are resulting in acceptance rates of up to 90%, equivalent to human translation acceptance rates and much higher than vanilla LLMs without this additional investment.

Automation is another high priority. As AI enables localization to scale further across organizations, the right foundations need to be in place to support that scale — which is why we've invested so heavily in workflows, and soon in new AI agents and even a Lokalise MCP server for more effective end-to-end automation.

Integrations are another important theme — we want to cater to the various content sources our customers use. We've doubled our investment in integrations over the past few months, so expect more updates there too. And we're working on a new value proposition for localization teams to scale beyond just software content — more on that later.

Live demo: the Figma Buzz integration

Marta (host): Sorry, Luke, I need to take over for a second — we just got a message from our demand gen team, we have to go live tomorrow in five markets, and there are brand assets in Figma that need to be localized, but our design team is too busy to help. Please tell me you have a solution, because otherwise I'm stuck.

Luke: Let me show you our new Figma Buzz integration. To localize this asset, all you need to do is find the Lokalise plugin in Figma — this is the new Figma Buzz interface — select the assets containing the content that needs translating, and export the content directly from Figma into Lokalise. That kicks off the process.

In the background, I've already set up a Lokalise project with the integration configured so that any content sent from Figma Buzz is automatically translated with AI. As a designer, all you need to do is export the content, then import it once the AI translation is ready — in this case, French and Italian.

Figma Buzz is Figma's product to help brand designers and marketers create on-brand assets for social media, ads, one-pagers, and more. Content designers can now translate content directly inside Figma Buzz — no copy-paste, no context loss — with instant AI translations that stay on-brand, using the same translation assets and context used elsewhere in your organization. For higher-stakes content, you can kick off a full Lokalise workflow with human review without ever leaving Figma. The result: marketers and brand designers keep their Figma localization operations, get full context inside the Lokalise TMS, reduce back-and-forth, speed up iteration, and critical brand assets still go through your localization workflows.

AI translation quality: custom AI profiles

Luke: Investment in AI translation quality is a top priority — we want to help customers scale localization to more content through AI while protecting translation quality. Our goal: human-level quality, assisted by humans, so nuanced translations get the review they deserve.

One solution we've been building is custom AI profiles, built with retrieval-augmented generation (RAG). Before translating, Lokalise AI retrieves recent translations from your translation history or other translation assets you select, identifies similar content already translated, and applies the same styling to generate new AI translations — domain-aware AI that mirrors your tone and terminology.

During Q3 we ran a closed beta with a select group of customers, and soon you'll be able to self-serve and create your own custom models on enterprise plans, including multiple custom models for different content types (UI, help center, legal, marketing, etc.) so content routes appropriately. In early results, these models achieve roughly a 90% acceptance rate without post-editing — comparable to average human quality for many use cases — based on tests across content from over 100 customers.

If you're interested in this topic, check out our recent webinars: one from Sasho, one of our engineering leads on the AI/ML team, going deeper into how this technology works, and one from Alessia on AI scoring. Both are available on demand.

Luke (demo): Within your team settings, there's a new custom AI profile section where you can create new AI profiles — either based on your translation memory (the RAG technology mentioned earlier) or custom profiles for specific datasets, for example a model trained to translate from English to Italian using RAG on tagged content. You'll need at least 500 examples per language for optimal quality. Then you select the projects you want the custom AI profile activated on, and that's it — every project in your Lokalise CMS then uses that custom model automatically when you translate with AI. It's self-serve, so localization teams don't need machine learning support to tune the model, and you get higher first-pass AI translation quality, fewer handoffs, and lower review costs.

Workflows + AI scoring automation

Luke: On lower review costs and self-serve options — customers used to need engineering involvement to build custom workflows to automate their localization processes, which is why we built workflows and kept investing in them through Q3. Our most recent investment targeted power users with small changes that save real time: defaulting contributors to the selected translation review language instead of manually selecting contributors per language, adding local time support (no more UTC calculations), and adding more scheduling options so workflows can run more than once a day or less than once a week.

The real power of workflows showed up in another Q3 beta where we layered AI scoring on top: workflows that first translate using Lokalise AI, automatically score the AI-generated translations, and then create human review tasks only for content given a low quality score. Feedback during the beta has been positive, and we're seeing teams reduce their review effort by up to 80%. We plan to launch this combination to general availability over the coming weeks.

Luke (demo): Here I've uploaded a JSON file with a source language of English, translating to Polish and Russian, using a workflow with an AI translation task followed by an automated human review task for low-scoring content. Once the workflow runs, the AI task translates everything and scores each translation — green scores are high, red scores are low. Once the AI task closes, a human review task is created automatically covering only the words that scored low — in this run, 35 words for Polish and 9 for Russian — along with information on why each translation got a low score.

In summary: workflows take you from reviewing everything to reviewing only what matters. Layering AI scoring on top gives you a hyper-automated, technology-backed localization process producing on-brand, trustworthy translations fast, while reducing manual coordination and keeping translation cycles tight — especially for teams shipping continuously.

New integrations and file formats

Luke: We've doubled our investment in integrations, and in Q3 specifically we released updates to our Webflow integration, aligning it with Webflow's public API and adding support for Webflow components. We launched a beta supporting XCStrings, a file format used by software engineering teams working on Apple mobile applications. And for Contentful users, we added support for connecting multiple Lokalise projects to the same Contentful instance, giving you more control over which workflows apply to different content in Contentful.

Sneak peek: Lokalise Autopilot

Luke: Localization is going through a lot of change right now — AI is revolutionizing the industry, which creates an opportunity for localization teams to increase the impact they deliver to the business. Using AI, teams can localize more content, meaning localization teams are increasingly asked to increase coverage across the business, and they need the right tool stack to do it.

So far we've focused on strengthening our offering for product and software use cases. The next generation of Lokalise will be designed and purpose-built for localization teams to cover more content localization across teams in the enterprise — fitting smoothly into the workflow of everyone from product, marketing, content, and design to eventually HR and data. The aim is for Lokalise to be so seamless you barely know it's there, leveraging context-rich AI that works behind the scenes without much configuration.

We're starting by focusing on long-form content, an area that needs some care: high-quality, in-context previews covering many content types and file formats, plus a new way of requesting translations through the Lokalise translation request portal — no repetitive setup, no new platform to learn, no handholding required, so even a new team member can localize on day one without slowing anyone down. We're aiming to kick off a beta for Lokalise Autopilot in the New Year, and we're looking for early testers if this is of interest to you.

Live Q&A

Q: Where can I see what's new about the new Figma plugin compared to the previous version?
The Figma Buzz integration is an extension of our existing Figma integration — it required investment to make sure it works with Figma Buzz specifically, since that's a newer product from Figma. It's the same underlying functionality, extended to Figma Buzz. Relevant help docs are available on request.

Q: If I translate with a custom AI profile and use scoring, does it deduct the amount of words twice from the AI words included in my subscription?
No — scoring does not deduct AI words twice.

Q: Is the AI feature (scoring and custom models) included in the enterprise plan? Is there a word limit?
Yes, AI scoring and custom models will be available on the enterprise plan — they're currently in beta, which is why some customers don't see them yet; they'll be released shortly. There isn't a realistic word limit: even with a very large translation memory, RAG looks for relevant, more recent translations rather than processing the entire dataset, which keeps things consistent and accurate.

Q: Would it be possible to see which engine(s) run behind the custom AI feature, and adjust them depending on content type?
Not at the moment — today it's based on our orchestration choosing the best model for the target and source language. It's on the AI/ML team's radar as a future initiative, so stay tuned; your CSM, if you have one, is a good way to stay updated on this.

Q: Will AI translation be included in project automations, as well as machine translation?
Yes — both AI translation and MT are available in project automations today.

With no further questions, the session closed with a reminder that the recording and beta program sign-ups would follow by email.