Treat the World as One Market
Default to one global version; localize only where law or platform forces you to
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 6
- Confidence
- 92%
Every new market tells you it is different and that you must adapt. Gotthilf's counter-position from internationalizing Duolingo across dozens of countries: cultural differences are the last 5% of understanding people, while forking your product per market multiplies code, process, personnel, cost, and — worst — the time every future experiment takes. The default is one global version, tested once and rolled out everywhere, with exceptions carved only for regulation, platform reality, and hard blockers.
Origin
Gina Gotthilf's experience launching Duolingo across Japan, China, Korea, Turkey, Spain, France, Brazil, Mexico, Chile, Argentina, India and more while leading growth and internationalization.
Core principles
- 01Humans are far more similar than they believe; people enjoy believing their country is special.
- 02Cultural difference is the final 5% of understanding a user, not the first 50%.
- 03Every localized fork must be maintained, tested, and re-tested for every future hypothesis, forever.
- 04Time is almost more valuable than money to a startup — and forking is a tax paid in time.
- 05The exceptions that are real: regulation, censorship regimes, and hard product blockers. Not colors, mascots, or 'we don't like free things here.'
- 06An 80/20: take the 80 globally; buy the 20 only when you have the luxury.
How to run it
- 1
Adopt one global version as the default
Build and ship a single version, and only deviate when you have a hard, named reason. The burden of proof sits on the localization, not the default.
- 2
Test once, roll out everywhere
When you run an A/B test or try new marketing, treat a win as a global win: if it worked here, presume it works there, and try it everywhere.
Pro tip This is also what makes your test velocity survivable — one version means one experiment, not N.
- 3
Discount 'here it's different' feedback by default
Gotthilf got the identical message in every launch market: here it's different, people are different, you don't understand, this won't work, change it. Examples she was given: green is a negative color here; owls are seen as bad animals here; here people think free things are low quality; here people love free things. The claims contradicted each other across markets.
Pro tip When two markets give you opposite versions of the same 'cultural truth', both are noise.
Watch out Discounting the feedback is not ignoring the market — it is refusing to fork the product on unverified folklore.
- 4
Carve exceptions only for regulation, censorship, and hard blockers
China required watching for specific words and phrases. Regulated industries and B2B contexts genuinely need per-country legal and financial handling. That is the exception list.
- 5
Verify the mechanics of local access — that IS the real work
The Duolingo India failure was not cultural, it was mechanical: because Hindi typing is hard and India has many languages, most people set their phone UI to English. Duolingo used phone UI language as the learner's base language, so it offered English speakers French, Spanish, and German — while most Indians were there to learn English. Nobody found what they came for and they left.
Pro tip Get on a plane. They only discovered the UI-language assumption when they finally went to India.
Watch out The distinction is sharp: do not localize taste, DO validate assumptions about devices, defaults, and access.
- 6
Buy the local 20% only once it's a luxury
When you have surplus resources and the global 80% is exhausted, then invest in market-specific work — with clear eyes about the permanent maintenance cost you're accepting.
Pro tip Anything you add is something you maintain forever — in life, and in code.
In the wild
Duolingo inferred a user's native language from their phone's UI setting. In India, because typing in Hindi is hard and the country is multilingual, most people set their phone UI to English — so Duolingo offered them French, Spanish and German lessons from an English base, when what they overwhelmingly wanted was to learn English. The team only discovered this by physically going to India.
→ Users didn't find what they were looking for and left — a mechanical assumption failure, not a cultural one, and one of the two genuine exceptions to the treat-the-world-as-one rule.
Gotthilf launched Duolingo in China, hit a million downloads on the first day, and then the app was blocked by the government. It stopped working, users rated it one star en masse, and the team could not work out what to do.
→ Recovering from the resulting rating damage was a prolonged struggle — and China became the other market where genuine local adaptation (watching for specific words and phrases) was required.
In every market someone told her the product had to change: green is a negative color here, owls are badly regarded here, here people think free things are low quality — and elsewhere, here people love free things.
→ Duolingo largely ignored it, kept one global version, and shipped every experiment worldwide — with the two exceptions above.
Common mistakes
Forking the product per country too early
Too many changes too soon means you learn very little, while adding enormous code and organizational complexity.
Underestimating the experiment tax of forks
Every fork means running every future test in every version. The cost is paid in time — and time matters more than money to a startup.
Trusting local folklore over data
'People here are different' is delivered with total confidence in every market and is frequently self-contradictory across markets. It is a feeling of specialness, not a growth insight.
Confusing cultural localization with mechanical validation
Duolingo's India failure had nothing to do with culture and everything to do with an assumption about device settings. Skipping the mechanical check while obsessing over cultural nuance gets it exactly backwards.
Is it for you?
Best for
Consumer product and growth teams expanding into multiple international markets with constrained engineering resources
Not ideal for
Regulated industries, fintech, or B2B products where legal, financial, and compliance requirements differ materially by country
From the transcript
“as a startup you focus too much on those marginal differences between groups of people you can run the risk of making big mistakes which…”
“we just found that by treating the world as one and whenever we deployed you know AB test or or tried something new in terms…”
“I got the same feedback whatever country it was here it's different people are different”
“now every single experiment or or or hypothesis is going to take you way longer to be able to deploy”
From the episode
Scaling Duolingo, embracing failure, and insight into Latin America’s tech scene
Gina Gotthilf (Latitud, Duolingo)