A product lead based in the US, who I was on a call with last quarter, explained, very patiently, that the team had decided to ship a behavioural change based on Q2 telemetry limited to the US. When I asked whether anyone had pulled the Bangladesh cohort, it turned out nobody on the call had thought to, even though the product had been live there for two quarters with roughly three times the segment variance they were trying to manufacture in their US sample. The cohort they were missing is the one I keep coming back to: a security guard working an evening shift off Mirpur Road, phone still on Android 6 and running Facebook Lite while it charges in the back room where he naps. His nephew belongs to that same cohort, having come up through a BUET intake and now writing Python for a B2B startup, carrying around an iPhone with Claude Code installed on it. Often enough that I've stopped finding it remarkable, the two of them live under the same roof.
Bangladesh is the most expressive A/B testing ground in the world and you can prove it from public numbers. The country sits around 170 million people inside an area smaller than the US state of Iowa. That works out to a population density a touch above 1,300 per square kilometre, denser than the entire United States compressed into a strip a tenth of the country's land area. The language is Bangla on the surface, but the median consumer in Dhaka has functional English for technology, casual Hindi from a lifetime of Bombay cinema, and at least liturgical Arabic from school. The religious split is roughly 88% Muslim with the rest distributed across Hindu, Buddhist, and Christian communities who live in their own concentrations at the level of the neighbourhood. Income spans six deciles that are nearly distinct between a Pathao rider and a Gulshan exporter without leaving a radius of fifteen kilometres. Median age is 27 and the distribution is sharply bimodal, with a cohort under 25 that grew up native to mobile phones and a cohort over 50 still native to cash, sharing the same WhatsApp family groups. Hans Rosling spent most of his Gapminder career trying to get audiences to see this kind of variance inside a single country instead of averaging it across continents, and Bangladesh is the test bench his slides kept implying existed.
The economics here are almost embarrassing. To stand up a panel of 5,000 users in Bangladesh roughly representative at the national level, with consented attribution against bKash transactions and a ping from the mobile network, costs less than a single quarter of a mid sized SF agency retainer. The same panel in the US costs an order of magnitude more and gives you variance only across geography and income, not across the device tier the user is on, their religion, or the gradient between rural and urban at meaningful scale. Daron Acemoglu and James Robinson built most of Why Nations Fail around the argument that institutions decide whether a country can sustain experimentation. Bangladesh's regulators have spent the last decade quietly making the consumer product experiment a permissive activity, while keeping money flow tightly governed. The friction shows up at settlement, and a feature flag ships without much resistance from anyone. That is exactly the regulatory shape a researcher wants.
The country is, by every variance measure that matters to a product team, the most expressive consumer A/B testing ground in the world.
The companies that have figured this out are mostly Asian and mostly quiet about it. ByteDance has been shipping TikTok and CapCut feature variants to Bangladesh ahead of broader South Asia rollouts since about 2022. The creator base on TikTok that speaks Bangla is large, native to mobile, and willing to engage with experimental video formats that India's heavier regulatory drag would slow down. Grab studied the mechanics of Pathao's ride hailing in Dhaka closely before importing several pricing and pooling patterns into Jakarta and Bangkok. None of this is documented in a quarterly slide, but it shows up if you read the engineering blogs of South Asian product teams. The same A/B framework names appear in Dhaka and Jakarta posts within a few weeks of each other. Daraz, now owned by Alibaba, has been running experiments on pricing tiers in Bangladesh ahead of its Pakistan and Sri Lanka markets for years, because the country offers a credible income distribution across six tiers inside one logistics network. None of these teams brag about it, because the advantage of running here only holds as long as your competition keeps not noticing that Dhaka is the panel.
The reason almost nobody else uses Bangladesh has less to do with any real gap in the market than with how few of the teams who could use it have actually looked. Patrick McKenzie has written, more than once, about the countries nobody discusses enough that quietly shape large parts of the SaaS economy, and Bangladesh fits the shape exactly. Part of it is that the country doesn't end up on a Q3 earnings slide; Silicon Valley's mental model of "emerging markets" goes through India, then plateaus, and the analyst note never reaches Dhaka. Part of it is that the teams who do use it have no incentive to evangelise, since an underused research instrument loses its edge the moment someone writes a Substack about it.
And underneath both of those is a habit worth naming on its own: most product orgs optimise for the country with the most articles written about it, rather than the one with the highest ratio of signal to noise. The league table of attention that results runs US first, then India, then Japan, with Bangladesh nowhere on the page. A product team that learns to read this market has a lead measured in years over a competitor that won't.
I don't want to oversell the country, because most of the boosterish writing about Bangladesh is the kind of thing that makes a serious product team close the tab. There are real frictions worth naming. Settlement out of bKash into a foreign bank takes longer than it should, reliable street addresses are still a problem for delivery experiments outside Dhaka and Chittagong, and the regulatory environment for data residency tightened in 2024 in ways that make some experiment designs harder. None of these break the deal on their own. All of them are solvable with a partner who has actually worked in the market, and the central claim survives regardless: the instrument is sitting here in working order, waiting patiently on the shelf for the class of product team that has actually learned how to read it to show up and put it to use.
Both men are inside the same A/B test on at least four global consumer products right now, whether the teams running those experiments know it or not. The question for the rest of 2026 is which product team notices first and builds a panel that accounts for both of them by design instead of by accident.
So far, on every call I have been on, the answer is none of them.
The teams I have raised it with mostly want to know whether we have a Bangla localisation partner, which is a smaller question, and a different one.