Conquering a country every week

There are 193 countries in the United Nations.¹ Not long ago, the consumer insights platform I work at in Manchester was opening five new ones a year. At that pace, covering the list would have taken forever until the 2060s.

Then we changed how we do it and went to one a week. Fifty new countries in a year. Roughly a quarter of the UN.

One year of opening countries

0 new countries

3% of the 193 UN member states, in one year

One box is one week. The totals are real: five new countries in a year, then fifty. Which weeks they fell in is only there to draw the picture.

Not a switch

There is a famous version of this story. In January 2016 a streaming service went live in 130 new countries in a single moment, in the middle of a keynote, and jumped from about 60 countries to 190.² One slide, one click, most of the planet.

I love that story, and it has nothing to do with us. A film does not need anybody on the other side of it. Our product is people answering questions. Companies come to us to find out what real consumers think of a product before they bet on it.

So to open a country you need the people. A panel that actually lives there, is big enough to answer quickly, and is who it says it is. Then the app those people use has to speak to them properly. Then you have to check that the method still measures what it measures at home.

When each of those is a project, you get five countries a year. So we stopped running projects and built a system, the same six steps every time. People do the parts that need judgment. AI does the parts that need volume.

Opening a countrystep 1 of 6

Find the people

Pick a third-party panel that actually has respondents in the country, enough of them, and real ones.

The real sequence. How long each step takes varies by country, so the picture shows the order, not the clock.

The fourth step is the one that changed the math. Translating an interface is easy now. What our localization does is harder: it makes sure the questions match the version of the methodology that country uses. Scales are the obvious example. The agree-to-disagree scale a country runs on, and the exact words on each of its points, have to match that country’s version of the method before the first answer comes in. Otherwise the answers mean something different from the ones at home, which is the whole problem of the next section. It also picks the right language in the first place, which in a lot of countries is a decision before it is a translation.

Then a day of real testing, and the country goes live for every customer at once.

Real people

My own teams look after data quality and the mobile app, so my part of every launch is the unglamorous question. Is the person answering a person, and do they mean it?

That is hard everywhere. When a large American polling organization tested the kind of online panels anyone can sign up to, 4 to 7% of respondents turned out to be bogus. 84% of the bogus ones passed the trick question that was supposed to catch them. Their tell was that they said yes to everything.³

Then it got worse. In a 2025 paper, an AI built to take surveys passed 99.8% of 6,000 attention checks.⁴ The old defense was a line like “select strongly disagree to show you are reading.” The robot reads.

So a country is not open when the panel is connected. It is open when the answers coming out of it can be believed.

A four is not a four

Suppose the people are real. The next problem is that honest people in different countries use the same scale differently.

In 2006 a researcher put the same kind of five-point questions to students in 26 countries. In Japan, 39% of all answers landed on 4 or 5. In the Netherlands, 48%. In the United States, 54%. In Mexico and India 60%, and in Malaysia and Taiwan 61%.⁵ Nobody was lying. That is simply what “quite good” sounds like in each place. A bigger study across 19 nations found the same thing and tied it to culture.⁶

Your product scores 52% “4 or 5 out of 5”

Japan39%better than usual

Netherlands48%better than usual

USA54%about normal

Mexico60%worse than usual

India60%worse than usual

Malaysia61%worse than usual

Taiwan61%worse than usual

The gray bars are real: how often people in each country chose 4 or 5 across many unrelated questions (Harzing, 2006; 26 countries, 1,581 students). Your product and its score are made up.

The detail I enjoy most is the flat 5. Only 8% of answers in the Netherlands went all the way to the top, and 10% in Japan. In Mexico and India it was 28%. And the language of the questionnaire moved things as well. Asked in English, people drifted to the middle. Asked in their own language, they went to the edges.⁵

This is what “making sure the methodology works” means in practice. A score has to be read against the local habit. Otherwise you launch a mediocre product in Malaysia and kill a great one in Japan.

Not just language

Everyone expects translation to be the hard part. It is certainly the part with the best stories. The car that flopped in Latin America because its name meant “doesn’t go.” The cola whose slogan promised to bring your ancestors back from the grave.

Both are myths. The car sold perfectly well, and the slogan is filed under urban legends.⁷

The real traps are duller, and they are not about words at all. Take the weekend.⁸

Mon

Tue

Wed

Thu

Fri

Sat

Sun

The weekend you probably assumed.

Rest days by country. Ask about “last weekend” on the wrong morning and you are asking about a day at the office.

Or take “which language?.” India has 22 scheduled languages and no national one. South Africa has twelve official languages, since sign language was added in 2023. The world has 7,170 living ones.⁹ “Translate the app” is not a task. It is a list of decisions.

Which is why I keep telling anyone who will listen that localization is not about the language. The language is the bit a machine can do.

A country a week

I called this piece “conquering,” which is how it looks on a dashboard. It does not feel like that from the inside. Every week a new place teaches us one more thing we had assumed was universal: what a weekend is, what a four means, who counts as a real person.

Fifty countries in a year. If nothing breaks, the rest of the UN is under three years away, give or take the ones we already had. That is not a sentence I expected to write in a job about surveys.

Sources and further reading