Somewhere in your codebase, a developer typed “Sign up” directly into a button component. It took two seconds and cost nothing. Multiply that by a few thousand strings and three years of shipping, and it becomes a project that can delay your entire international roadmap.
Brian McConnell has seen this pattern from the inside. Before founding Localization Technology Partners, he led localization at Lyft, Medium, and Notion. Today, Brian helps startups and scale-ups set up internationalization before the bill comes due. In his conversation with Stefan Huyghe on The Agile Localization Podcast, he made the case that the cheapest time to prepare for global growth is long before you think you need it.
Listen to the new episode on:
Two shortcuts that come back to bite
Ask Brian which early decisions cause the most pain later, and he names two.
The first is hard-coded English text. Almost every team does it, because it creates no friction at first. The trouble starts once there are thousands of strings to extract before anything can be translated.
The second is choosing a CMS or help center without thinking about multiple languages. Those platforms are usually picked long before anyone mentions going global. When localization finally comes up, the company often learns its marketing site and support content have to move to a new system first. That’s a migration nobody budgeted for.
Neither problem requires translating anything early to avoid. Building with clean internationalization habits costs almost nothing up front, even if at first you only support British English.
There’s no harm in doing it soon, because that forces you to follow design and coding practices that you’ll need to do anyway.
Trust usage over promises
How do you know when it’s time to actually translate? Brian’s favorite signal is people using your product anyway.
When he arrived at Notion, the app had no Japanese or Korean interface, yet both markets were growing quickly. Notion was a tool for building wikis and knowledge bases, so nothing stopped people from writing in their own language. That organic traction made the case for localization almost by itself.
If people are starting to use your product in spite of it not being adapted for their market, that’s a signal that there’s strong demand there.
He is more skeptical of a different kind of demand, like when the sales team insists that a German version would close a batch of deals. Sometimes it’s true, sometimes it isn’t, and chasing those promises leads to reactive, scattered expansion. Competitor launches are worth watching too. If a rival is moving into a market, it’s probably seeing the same signals.
Waiting costs time, and time costs markets
Getting the product ready to run in multiple languages and regions took Lyft almost 18 months.
Money is less of a problem than the time to market.
While Lyft cleared its backlog, Uber kept expanding. Localization wasn’t Lyft’s only misstep. Uber adapted its model for markets like Germany by dispatching local taxis through the familiar app, and its food delivery business carried it through the pandemic. Even so, Brian sees the delay as a big reason the two companies ended up so far apart in value.
Think plumbing
To a startup CTO, Brian frames localization infrastructure as an integration problem. App strings, the CMS, the help center, and lifecycle emails all need to connect to one translation system.
The trouble often starts with who gets the job:
It’s, “Hey, Alex, you speak Spanish, right? Oh, great, you’re in charge of localization.”
That person searches Google or ChatGPT and lands on whatever solution comes up. Engineering-led efforts make a similar mistake from the other side: they pick a tool that handles UI strings well but struggles once the CMS and help center come into play. Brian unpacks this pattern in more detail in his guest post for the Crowdin blog, Growing Pains: Challenges Faced by Startups and Emerging Companies.
For content, Brian favors CMS platforms like Contentful or Storyblok, which are designed with localization as a primary use case. For translation, he looks for ready-made integrations, since engineering time is always scarce:
One of the things I like about Crowdin is it has a pretty extensive ecosystem of system integrations, so you don’t have to waste a lot of time building bespoke custom tooling or software to connect your CMS to the translation platform.
Keep the team small
With those pipes in place, teams stay small. When Brian left Lyft, it had about 3,000 employees and two people in localization. The goal is full automation:
The ideal state you want to get to is nobody really needs to know how the translations are done or how to request them.
A writer publishes a new help center article and gets an AI placeholder almost immediately, with vendor review a few days later. The localization team stops shipping files back and forth and focuses on priorities and vendors. For those vendors, Brian suggests a mid-sized LSP – one “big enough to serve your needs, but not so big that your business doesn’t matter to them.”
Speak in revenue
To get the budget approved, Brian shows leadership how much revenue localization can bring in. He puts the pitch in simple terms:
If you spend a million dollars on this, you’re going to get 10 or 20 million back. That’s always the kind of investment they’ll want to make.
Notion now earns close to 70% of its revenue outside the US, which he says it couldn’t have reached without localization. Before that kind of revenue exists, he points to peers like Atlassian or to Spanish speakers at home. At Lyft, his team found that among Spanish-language users in Los Angeles, Uber was about three times more likely to be installed than Lyft.
He also suggests renaming the team international experience. Executives grasp it more easily than localization, and it covers the full scope of the work.
Let go of old rituals
Brian calls translation memory a dinosaur. It’s still useful for exact matches, but it has become background infrastructure, like caching, that nobody outside the team thinks about.
He also questions the focus on quality scores. If churn and engagement in a language look like the rest of the product, the translations probably aren’t holding anyone back. Users who are used to imperfect AI tend to be glad the content exists in their language at all. Medical and legal content are the clear exceptions.
Human judgment still matters most where AI lacks context.
If it [AI] doesn’t have good context, it just becomes a guessing machine.
Short UI strings are the classic case, where one English word can be either a noun or a verb.
The window is shorter now
Companies used to spend five years proving themselves at home before thinking about international markets. Now a product like Lovable can attract users worldwide in its first year.
Brian works with Replit, whose AI already manages many languages even though its interface isn’t fully localized. Global users can show up before the roadmap expects them, and the teams that prepared early are the ones ready to welcome them.
Brian’s background
Brian McConnell is the Founder and General Manager of Localization Technology Partners, a consultancy that helps startups and scale-ups design internationalization and localization programs built to keep pace with modern software development. Over more than two decades in the field, he has led localization at Lyft, Medium, and Notion. At Notion, he helped turn early international traction into a major share of revenue. Brian is also Fractional Head of Localization at Replit.
The views, thoughts, and opinions expressed in this podcast are the speaker’s own and may not necessarily align with those of Crowdin.
Listen to the new episode on:
Localize your product with Crowdin
Yuliia Makarenko
Yuliia Makarenko is a marketing specialist with over a decade of experience, and she’s all about creating content that readers will love. She’s a pro at using her skills in SEO, research, and data analysis to write useful content. When she’s not diving into content creation, you can find her reading a good thriller, practicing some yoga, or simply enjoying playtime with her little one.