Introducing Lokalise Workflows: 5 steps to automate localization the right way

Full transcript

Introductions

Maria (host, Lokalise): Let's kick it off and start. Today with me are Luke and Paula. Luke is a Staff Product Manager here at Lokalise and has been working on our Workflows feature, and Paula Fernández Quesada is Manager of Product Localization at Navan. Luke and Paula, can you introduce yourselves?

Luke Vella: Hey Maria, and hey everybody — really great to be here. As Maria said, I'm a Staff Product Manager at Lokalise, and I'm currently leading the Workflows initiative. I've been working on this for the past few months now, and I'm really happy to be here.

Paula Fernández Quesada: Thank you everyone for coming, and thank you to Lokalise for inviting me. I'm Paula, a manager on the product side at Navan. We're a tech company in the business travel and expense management world, and I've been working together with Lokalise for the past few months on testing the workflows. I'm looking forward to sharing some tips, tricks, and ideas.

Maria: Awesome, thank you for joining. We can start by defining whether we really need automation, or whether it's something optional. Luke, the stage is yours.

Why automate your localization now

Luke: Let's kick things off by talking about why you should start thinking about automating your localization processes. This applies if you're new to localization or if your processes aren't as refined right now, but it also applies to more experienced localization teams who set up their processes some time ago and are considering whether they're scalable enough for future needs. I'm going to walk you through some of the trends we're observing that support prioritizing localization process automation as soon as possible.

The first clear trend is that our users are adding content at a significantly faster rate. We're seeing content addition growth rates increasing by 20% year on year — the rate at which content grows every year is itself growing, an exponential growth trend. So regardless of whether you're new to localization or have been doing it for some time, you can expect the demand for localization to increase. Customers who prepare their localization processes for scale will likely see less interruption in their day-to-day operations even as the amount of content to be localized continues to grow.

This growth in demand is being further inflated by new technologies such as AI reducing the perceived friction and cost of localization for stakeholders outside of localization departments. We see this firsthand at Lokalise: machine-assisted translations, which include AI translations, now account for 70% of all translations, up from 59% in 2023. AI-powered translations specifically grew 533% on their own in 2024. But here's the catch — some of our most successful companies are not replacing human translations, but strategically combining both in a hybrid model. Localization processes need to be set up in a way that supports this hybrid model, and because these processes can be more taxing to manage than purely machine-driven ones, supporting them without automation is going to be a challenge for your localization team's efficiency and scalability.

We're also noticing interesting trends around repeated content. Companies are increasing their reuse of previously translated content by 150% over last year, which highlights the power of translation memory. More teams — from marketing to product and legal — are generating content that often contains repeated language. This creates a need for structured workflows that manage volume and consistency while also reusing existing translations, which can significantly reduce cost and accelerate the delivery of new translations to market.

In addition, while machine translation is becoming the norm, we're seeing a drop in the number of translations initiated through our API endpoints. We view this as a signal that it's becoming increasingly important for customers to localize efficiently not just by leveraging machine translations, but also by avoiding reliance on expensive engineering resources to tailor localization software. Built-in machine translation and no-code workflow automation in translation management systems make setting up automated workflows accessible even to non-technical localization teams. Automated workflows remove bottlenecks, empower non-technical teams, and ensure that continuous localization can be implemented at scale.

Finally, features that enable more automated processes are indeed being used more and more: companies doubled their use of automated translation workflows in 2024, making it the most popular localization approach. Automated workflows are no longer a nice-to-have — they're evolving into a necessity for companies scaling their localization efforts. This matters because companies that automate their workflows see faster turnaround times, lower costs, and the ability to scale more efficiently. This is why we've been working on Workflows: to help companies better prepare their localization processes with minimal effort. But before you leverage the new Workflows feature, it's worth taking a moment to analyze your current ways of working and identify how to use workflows in the most optimal way. I'll walk you through five steps you should consider before fully automating your localization workflow.

Step 1: Define clear goals and outcomes

Luke: Our first recommendation is to start by defining clear goals and outcomes for your localization processes. Picture yourself in the future and think about how you'd like your localization processes to look and feel, and, more importantly, how these processes help the business achieve its goals. Are you after faster time to market, cost reduction, better translation consistency? Based on this, think about how you can measure the impact of your workflows — which metrics and KPIs will show in a quantitative manner that you've successfully achieved what you set out to achieve: turnaround time, quality score, cost reduction? And of course, make sure you can actually measure these KPIs. This will help you report on the efficiency and effectiveness of your localization processes and get further buy-in for added investment in localization from senior stakeholders. What you really want to avoid is jumping into automation without aligning it with real business needs.

As an example, you might have different process needs for different tiers of languages. Say you mostly serve customers in Germany, France, and China — you consider these translations business-critical, so you might want quality-focused workflows for those languages. You might have other languages that are less critical, where the business is okay with slightly lower quality if it results in better cost savings — in that case you might define separate workflows for them. You may instead realize that all languages should get the same treatment, but different content sources need different handling: for example, your design-stage localization in Figma requires speedy translations, while your mobile app translations should prioritize quality. Once you have a good understanding of the goals and objectives, it becomes easier to identify the localization workflows that best fit your needs.

Step 2: Design your workflow steps and involve the right stakeholders

Luke: Once you've mapped out the goals of your processes, start designing and implementing your workflow steps and ensure the right stakeholders are involved at the right time. You can form different workflows for different groups of content and languages with different goals. If you want to optimize a particular process for speed, you're likely better off creating a workflow based on machine-driven or AI translations. On the other hand, if you're optimizing for quality, you might want workflows with multiple steps that layer on additional human reviews.

Some key points to keep in mind when selecting workflow steps: use no-code automation for repetitive tasks — translation memory, task assignments, status tracking, and AI or machine translations can all be automated. Balance AI and human review: use AI and machine translation for UI strings, help docs, and bug content where speed matters most, but keep marketing, legal, and brand-sensitive content under more human control. Set clear ownership, actions, and permissions, defining who is responsible for each step — especially when there are humans in the loop — to avoid bottlenecks. Automate handoffs and notifications so tasks move forward without waiting on manual updates like emails and ad hoc Slack conversations. And finally, track team performance to identify bottlenecks and optimize your workflows.

Step 3: Make your workflows cost-efficient

Luke: Now you have your first workflows taking shape, and you're confident they'll translate into tangible results because they're mapped to objectives. But we haven't talked about a very important topic: costs. Having automated workflows is great, but you'll still hear questions from stakeholders if they aren't cost-efficient. We've already spoken about leveraging AI and translation memory to reduce redundant work. If you're using some form of human translation or review through a third party, it's also worth understanding your language service providers' terms, conditions, and fees, and staying on the lookout for other providers who might offer an improved service for your specific usage.

One way to improve cost efficiency is by bundling translations together and avoiding a large number of smaller files. This is especially true if you use the services of LSPs, who typically have a minimum order cost. By setting up your workflows on schedules that bundle key updates together into periodic tasks, you're less likely to incur those minimum order costs, making your workflows more cost-efficient in the process.

Step 4: Build in quality from the start

Luke: By now you probably have your workflows set up and are almost ready to go, but there are a couple more things to keep in mind — especially to ensure your workflows prioritize translation quality from the start, even when you're also optimizing for cost efficiency. Technologies like AI typically increase translation quality as more context is added, so it's worth investing time upfront to make sure your style guides, glossaries, and similar resources are set up to maintain brand consistency in your translations. Skipping this could impact the quality of your translations, which can result in a deterioration of trust in the localization process as a whole.

Step 5: Keep iterating — and the key don'ts

Luke: Don't forget that tweaks to workflows will be needed. This is another reason to adopt no-code workflow solutions, so you avoid becoming overly reliant on expensive engineering resources for even small tweaks. When setting up your localization processes, design workflows to be scalable, and avoid too much variance and too many different workflows so you don't add the burden of managing multiple slightly different versions of your processes. Continuously analyze performance so you know when you need to tweak, dedicate time to adjusting workflows based on the historical trends and bottlenecks you see, and avoid sticking to a rigid process that doesn't evolve with your localization needs — consider revamping processes from time to time. As your organization evolves and the demand for localized content increases, so will the need to revise your workflows to ensure they remain relevant as your team scales.

To close off this topic, some key don'ts. Avoid not setting clear objectives — this can lead to unnecessary complexity and wasted automation. Avoid too many manual steps — they slow down processes and add human error risk. Avoid a lack of ownership and visibility — this causes confusion and delays, whereas if people know what's expected of them and are ready to action incoming requests, you're more likely to set your processes up for success. Avoid skipping tools like style guides and glossaries — as mentioned, context is important for higher-quality translations. And finally, don't fail to leverage AI and translation memory, because that increases costs unnecessarily. I hope this has been helpful and that you've taken away some insights for developing your own scalable workflows. I'll hand back over to Maria, who will continue by discussing workflows further with Paula.

How Navan automates localization: a conversation with Paula

Maria: Thank you, Luke — very insightful. Lots of good advice, and now is a good chance to hear from someone who has applied all of this in reality. Paula, you introduced your role at Navan earlier, but can you explain a bit more about what the company does and what your role is there?

Paula: I joined Navan almost two years ago. I'm a one-person localization team, in constant collaboration with product, engineering, design, and marketing teams. We focus on the business travel and expense management industries — an all-in-one platform for mid-to-large companies and also some smaller companies. Localization became core for us as we expanded more and more into the EMEA region; the company is originally from the US, and there was a need for an established localization program, which is where my role came in. We've been working with Lokalise for years now, and Workflows was something I was looking forward to even before it was announced. The fact that it's a no-code solution helps me a lot because I'm not dependent on the engineering teams — as Luke said, it's really good for ownership on the localization side.

Maria: A one-person localization team feels like a lot of work. How was your life before automating anything with workflows? What were your main challenges at that point?

Paula: It was a resource issue, and definitely a time issue. A lot of our projects are connected to GitHub automatically, so we have a constant influx of keys in and out of Lokalise. Even if I sent something to the translators once or twice a week, there was constant asking about when translations were happening — there was no visibility on when things were happening. The idea behind introducing workflows was visibility for me, for the engineers, and for product, and making sure I had the time to invest in developing a proper strategy for the localization program instead of just checking every day in Lokalise which new keys we have.

Maria: And why did you decide to give workflows a chance? Was there anything in the feature itself that resonated with you?

Paula: On the one hand, the ownership of setting it up myself — understanding how the translators and the different teams work, and being able to set it up without depending on engineers to change scripts or GitHub Actions. And our company works a lot with AI, so we want to explore that more for regional adaptation. Workflows are a huge opportunity to create a program with something I call hands-off internationalization, or hands-off regionalization. That's something we're investigating a lot — we're doing small tests for now — but it's really the ownership, I would say: moving towards full program ownership for localization.

Maria: I'm curious about those use cases you mentioned. In reality, how are those workflows set up, and what steps are you following?

Paula: For now I'm testing a double workflow setup in one of our projects. One workflow is set up on a daily schedule: it filters out specific keys — English (UK) only for now — and applies Lokalise AI. It comes to me first to check everything, and the idea is to then assign it to a reviewer if needed; for now, as a test, we're only doing the Lokalise AI step. I've seen it working very well — it runs every day, I can see right away what has changed, and it's really hands-off, so I don't have to worry about it. The other one is a purely TM-based translation workflow — the double step Luke mentioned in his slide about not forgetting the power of TMs. On the same schedule (although I've changed it to three times a week while I'm still testing), all the other languages run through translation memory applied at a 99% match, which for us is almost like 100%. It clears the task, assigns everything to me, I do a quick check, and then assign it to the translators as I see fit. So it's a double workflow within one single project, and it's really useful to see how all the different filters and setups work, because you can define exactly what you want to do for each one.

Maria: You're one of our beta testers. Have you seen any results, any proof or numbers you can share?

Paula: I'm still in the very early stages, working with a lot of the teams to set things up properly, but it has improved over time. I easily save three to four hours a week by not having to click around the different projects and find out what's new — I can see it right away on the task creation panel, including whether any tasks were created at any given time. It's great for ownership and visibility of what's happening, and also for communication with the different engineering teams, because I'm aligning all the workflows with the different sprints and working with them instead of chasing translations. They understand exactly how I work, and it sets expectations on when things are happening. So I'd say it's a time-saving tool, and it helps a lot with communication, because I can very clearly explain how things work and clarify the expectations for everyone.

Maria: Lots of people here haven't started with workflows yet. Can you give them some advice, or lessons they should consider before trying it out?

Paula: My main point would be one Luke already mentioned: really define what you want to do with the workflows. Something I always wanted to do since I joined Navan was to automate a lot — create automatic workflows for translators, create tasks for them, and make sure I assign automatically. That was my objective from the get-go, and now, having this tool, that objective is more precise for every team. So my main suggestion is to really define what you want to use workflows for, and work with the other teams within your company. Don't do everything on your own — go to the other teams and talk to them so you can set it up together, aligned at least on the timelines for all of the teams.

Maria: Very insightful — thank you, Paula. I'll use the chance to remind everyone that you can ask questions to Luke or Paula in the Q&A panel; we'll go to that section in a bit. Thank you, Paula, for sharing your experience with us today. It was a pleasure having you.

Paula: Thank you too — it was really exciting. And thank you for creating Workflows, actually.

What's next from Lokalise

Maria: Before we jump to the Q&A, a few last items. As you've seen, localization is somewhat more complex than just one step, and Lokalise is working very hard on improving not just one part of the localization process, but the whole localization lifecycle. We discovered a bit about Workflows today; we also have Lokalise AI, which some of the workflow templates can integrate; we will explore analytics in our upcoming webinar; and we improved permission management last year for better management of your users.

A reminder that we have an upcoming webinar happening in March, covering the localization metrics that matter: you can track a lot, but how do you know what you should track, what you should present to stakeholders, and how to link it to business decisions? You can register via the link we'll be sharing — we hope to see you there. And before we jump to the Q&A, we have a short poll for you: we want to know what you're missing to take the next step in your localization process.

Q&A

Q: Which workflow templates are available in Lokalise, and for which use cases can they automate localization?
Luke: We're starting off with quite a large number of templates. As a reminder, Workflows has been available to everybody for about a month, so it's a very young product and we're very much invested in improving it further. There are templates for fully automated workflows — as simple as a single AI or machine translation step using Google Translate, or even just translation memory — up to more complex ones that pass your content through translation memory, then AI, and then a human review. There are also a few templates with a double review step, in case you want two rounds of reviews — quite useful for content that's specific to a particular profession or has very context-specific language. So there's a varied number, and they can automate machine-driven translations, but they also remove the need for you to manually create translation and review tasks for humans.

Q: What happens with workflows that include an AI translation step if the quality is not good and that step is already automated?
Luke: Firstly, the AI quality results we're getting are actually very, very promising — we're at a stage where it's very close to human-level translation. But you can never be too sure, and this is why we spoke about the hybrid model, and why one of the first templates we launched also includes a human review step. I'd really recommend it for content where you're more quality-conscious.
Paula: It's also very important, as mentioned during the presentation, to make sure you leverage all your reference materials — style guides, glossaries, and so on — because that goes hand in hand with AI quality. If you're using Lokalise AI, it ties in with your style guides and glossaries, and together with a human review step the output keeps improving. I can understand if it's not 100% what you'd expect at the beginning, but that happens regardless of whether you have automation or not. So make sure you leverage reference materials, style guides, and glossaries, and keep a clean TM as well.

Q: Can we expect that workflows can be copied or duplicated from one project to another?
Luke: This is a very good point. One of the reasons we started with templates was specifically to make it easy to set up a workflow. Workflows are at the project level, and there are a lot of project-level settings, so it was important for us to retain that granularity, at least to start with. Copying existing workflows from one project to another is a bit trickier — that's why we didn't tackle it from the start — because not all settings, like tasks and contributors, can easily be replicated across projects if you don't have the same settings in each project. But we are committed to making workflows work better in multi-project environments. It's a very young feature that we're investing in heavily, and this kind of feedback is very helpful and will help us prioritize the right things.

Q: How do you find the task menu when working with workflows? Any tips for organizing it when so many daily tasks are created?
Paula: That's a very good question, and I don't think I have a perfect answer — it's something I've carried from job to job. On top of the tasks view in Lokalise, where I assign all tasks to myself and the translators and keep that view in the side panel, I have my own trackers: my own spreadsheets for costs and budgeting, and a Jira board for timelines, epics, and connections to bugs and other teams. Lokalise is my quick view to see what has been done and the status of every language. That part is a little bit manual, regardless of whether tasks come from workflows or are created manually — so for me it's multi-tool task tracking, not just Lokalise.

Q: You mentioned having different tiers for languages. In the future, will it be possible to select different target languages for different workflow stages in the same workflow — for example, apply translation memory to all target languages but select specific target languages for the review task?
Luke: That's a great question. We've received this feedback from a couple of other sources as well, and it's something we're investigating at the moment. We're looking into more flexibility within workflows; whether this specific feature will be included isn't something I can answer directly right now, but let's get in touch and have a more detailed conversation about the rationale behind it.

Q: I see there's a Status tab coming under the overview. What can we expect from that?
Luke: The reason behind the status page is to have a centralized place where you can view the status of each of your workflows — which of them are running, since every workflow can handle multiple runs (every day, multiple times a week, and so on). The idea is to give you more granularity into exactly what's happening under the hood — for example, where a key is in a given workflow run and why — so you can avoid having to reach out to our support team with those questions. It's just a more helpful way to understand what's happening under the hood.

Closing remarks

Maria: I see other questions mostly related to the roadmap. In those cases we'll answer separately, because we want to understand your use case and the exact goal of the question — and as Luke said, that will also help us better shape what's coming next in Workflows.

Luke: Since we've talked a bit about tasks and I see more questions about them: tasks are another area of interest for us. Do reach out to us and make sure your feedback comes across to your CSM, if you have a dedicated CSM, or to our support channel — this is an area we'd love to hear more about as well.

Maria: We're almost on time, so I'll take these last two minutes to thank you all for joining us today. A reminder that Workflows is available in your Lokalise plan — you can jump into your project, go to the Workflows tab, and start automating today. Make sure to send us your feedback, and see you in the next webinar.

Luke: Thank you, everyone.

Paula: Thanks, everybody — and thanks, Maria, for hosting.