Learnings From Elsewhere: Why AI Research Tools Are Failing 85% of the World

By Insight Platforms

• article

• Behavioural Insights
• Market Trends
• AI
• Artificial Intelligence
• Accessible Research

I learned a new expression this year: WEIRD. And no, it’s not for describing Star Wars superfans. It’s an acronym for Western, Educated, Industrialised, Rich, and Democratic, and it explains why so much of the AI research technology we celebrate in London and San Francisco simply doesn’t work for most of the planet.

As a Brazilian, I’ve always noticed that technology showcased at Insight Platforms events doesn’t quite translate here. The Portuguese sounds off. The generated data doesn’t reflect how we think. The infrastructure assumptions don’t hold. But I assumed these were isolated problems.

They’re not.

Over the past year, I’ve hosted Insights from Elsewhere, a podcast featuring senior insights leaders from Latin America, Africa, Asia, Eastern Europe, and the Middle East. What began as curiosity about research beyond traditional power centres revealed something more urgent: researchers in emerging markets aren’t just adapting global AI tools. They’re exposing fundamental flaws that limit effectiveness worldwide.

The numbers tell the story: 85% of the world’s population lives outside the US and Western Europe. That’s 7 billion people whose reality doesn’t match the assumptions baked into most research technology.

When researchers in the US and Europe talk about “global insights”, they imagine methods and tools travelling smoothly across continents. But the insights professionals I spoke with are driving innovation through necessity, creativity, and deep cultural understanding, and their experiences offer a masterclass in what the global research community and AI tool developers urgently need to learn.

This article is a summary of topics discussed on the Insights from Elsewhere Podcast. Listen to the episodes below:

The WEIRD Problem: When “Universal” Tools Aren’t Universal

The term WEIRD, coined by behavioural scientists, describes populations that are Western, Educated, Industrialised, Rich, and Democratic. The problem? Most psychological research, and by extension most research technology, has been developed for and tested exclusively on WEIRD populations.

Sushma Panchawati from India put it bluntly: “LLMs are trained primarily on English content from Western countries. Multiple research papers have proven this bias. Somebody sitting in Nairobi would know a lot more about America than Americans would know about Jakarta, Nairobi, or Mumbai. LLMs replicate this asymmetry.”

How WEIRD Bias Manifests

This bias creates barriers across emerging markets in predictable patterns:

Language: More Than Translation

In Brazil, less than 2% of the population speaks fluent English, as Rosana Martins noted. Yet many research platforms are designed in English with poor Portuguese translations, not just linguistically, but conceptually.

The challenges multiply across languages:

• In China, Sami Wong explained that “one sound can mean thousands of different words with thousands of different meanings”.
• AI struggles with the metaphors, ambiguity, irony, and sarcasm that characterise Argentine Spanish, according to Constanza Cilley.
• In Poland, Agnieszka Górnicka notes that Hungarian semantic structure simply eludes large language models.

Infrastructure: The Invisible Barriers

High internet penetration statistics mask critical realities. In Kenya, Nyambura Mambo from Safaricom revealed that whilst internet penetration exceeds 90%, most devices are feature phones, not smartphones. SMS-based surveys work, but sophisticated video platforms don’t.

Constanza Cilley warned from Argentina: “High penetration stats don’t mean people can easily use sophisticated online research tools.” Internet literacy and bandwidth limitations create barriers that aggregate numbers don’t capture.

In China, Sami Wong can’t use ChatGPT without a VPN, and WeChat isn’t optional, it’s essential for business. In South Africa, Sheila Akinnusi noted that “the cost of data is prohibitive” even when the internet exists.

Cultural Blind Spots: Assumptions That Break

Japan’s ageing population defies Western stereotypes. Madoka Suganuma shared that 30% of Japan’s population is over 65, but “life starts at 65”. These consumers are tech-savvy, adventurous, and redefining what ageing means. The standard research cut-off of “65+” simply doesn’t work.

Sushma’s cross-market privacy study revealed surprising patterns: Germany and the US showed far more concern about digital privacy, whereas people in Brazil, Kenya, and India were excited about technology and weren’t limiting usage due to privacy concerns.

Cultural taboos vary dramatically and require local expertise:

• In Lesotho, Sheila Akinnusi learned never to ask women to discuss finance in front of men.
• In Brazil and Costa Rica, as Stephanie Vincent explained, asking someone’s income is deeply offensive.
• In Morocco, Soufiane Alkhatiri found face-to-face interviews essential for building trust and getting genuine responses.
• In rural Vietnam, doors open wide for foreign visitors, a level of openness unimaginable elsewhere.

The Monolith Trap

Regional labels distort reality. “LATAM”, “Eastern Europe”, “MENA”, “Africa”, “APAC” are PowerPoint categories, not lived realities. I remember being grouped into “LAPAC” in a past role, as if Latin America, Asia, and the Pacific naturally belong together.

The truth: Brazil has little in common with Costa Rica. Kenya is nothing like Morocco. Japan and China might as well be on different planets. Even within single countries, regional behaviours, foods, and traditions vary dramatically.

Countries behave as countries, not as clusters.

How Researchers Are Actually Using AI

Researchers everywhere – including myself – love AI for saving time on laborious tasks: data processing, coding, summarisation, organisation.

But outside English-speaking countries and Western culture, the representativeness of LLMs and their ability to predict behaviour decreases dramatically. They’re trained on data that doesn’t reflect elsewhere. They miss the implicit irony in Argentine and Brazilian speech. They stumble over regional dialects in India, Morocco, and China.

One insight united every interview: AI is a co-pilot, not the pilot. Human interpretation and contextualisation remain irreplaceable.

Despite challenges, researchers across emerging markets are optimistic about AI, but their approaches differ significantly from the “press the button” mentality common in Western markets. They use AI to gain speed, overcome local constraints, and make sense of massive datasets – but the value added to insights comes from humans.

Practical Applications That Work

Fraud Detection: In Costa Rica, Stephanie Vincent’s AI catches sophisticated two-step fraud (human passes initial checks, bot completes survey) in real time.

Time to Insights: In Morocco, Soufiane Alkhatiri validated AI ad testing by running parallel studies. Traditional methodology versus AI produced identical results in 5 minutes versus 2 weeks.

Transcription: Paul Gebara in Poland noted that AI reduced 1–2 days of manual transcription to one hour, freeing researchers for actual analysis – which is far more valuable.

Synthetic Data for Niche Targets: Soufiane explained, “Small sample bases can be expanded through simulations. You can run it many times.” This proves particularly valuable in emerging markets where niche segments are harder to reach.

Predictive Analytics: Elvys Nunes on his global role at Reckitt sees the most exciting frontier: “Understanding new consumer needs even before consumers articulate them. Finding white space opportunities where no one is looking.”

The Critical Human Element

Every researcher I spoke with emphasised that AI is a tool, not a replacement.

Sheila Akinnusi from Nedbank South Africa put it perfectly: “AI cannot make the decision at the end of the day. It’ll give you a playbook, but you, as the individual, still have intuition, gut feel, and common sense. That’s where the power lies.”

Sushma Panchawati warned of two critical AI pitfalls: Western bias in training data and the “single story problem”. LLMs rush to conclusions and discard everything that doesn’t match the dominant narrative. “We have to train these models to sit with contradiction because human nature is contradictory.”

What Researchers Actually Need from AI

All of them identified clear gaps that represent massive opportunities:

1. True Localisation

Not just translation, but genuine linguistic sophistication:

• Understanding Moroccan and Chinese dialects for social listening and qualitative analysis.
• Hungarian semantic structure understanding.
• Identifying Regional variants: Costa Rican Spanish ≠ Argentine Spanish ≠ Mexican Spanish.

And the list goes on… you get the gist!

2. Context-Aware Training

Sushma Panchawati shared the approach of her company, Pre-Data AI’s model: train digital twins on specific consumer data for particular personas. As she said: “If you’re selling acne solutions in Jakarta, you don’t want an American perspective. You want that Jakarta person’s point of view.”

3. Integration Across Methods

Rosana Martins in Brazil noted that most platforms focus on a single method: panel OR social listening OR qual. “Few blend multiple data types seamlessly.” Researchers need tools that triangulate qual + quant + behavioural + social in one place.

4. Simplicity Over Sophistication

Constanza Cilley’s insight from Argentina: When targeting middle and lower-class consumers, Zoom outperformed sophisticated platforms. Why? “It’s what participants are used to using. Feel familiar, feel confident.” Implementation matters more than features.

5. Infrastructure-Appropriate Solutions

Tools must work within local realities and adapt to local legislation, digital literacy, data availability and internet reach and bandwidth. While this is not accounted for, the barriers for technology usage will remain insurmountable.

The Bottom Line

AI research tools don’t need to be less sophisticated to work in emerging markets, they need to be more sophisticated in different ways. They need to:

• Handle linguistic complexity, not just translation.
• Understand cultural context, not just text.
• Preserve contradiction and nuance, not rush to summary.
• Work within infrastructure realities, not assume unlimited bandwidth.
• Support human interpretation, not replace it.
• Enable triangulation across methods, not silo data types.

Researchers are already innovating: combining ethnographies with AI analysis, using WhatsApp for diary studies, building self-service platforms on limited budgets, validating synthetic data against real respondents, and training local language models.

The question isn’t whether AI research tools can work in non-WEIRD markets. It’s whether tool developers are willing to learn from the researchers who know these markets best.

The Untapped Opportunity

The business case for getting this right is compelling:

Poland is the second-fastest growing economy globally over the past three decades, second only to China. Yet Paul Gebara and Agnieszka Górnicka report that Poland is consistently dropped from country lists because it’s mistaken for “that French country somewhere.” Poland just entered the G20. The CEE research market is worth $1 billion.

Central America weighs as much as Colombia in economy and purchasing power, yet Stephanie Vincent notes it’s “hugely underrepresented in research and business strategy.”

Morocco sits at the crossroads of Europe, Africa, and the Middle East, with a 25% youth population and World Cup 2030 creating massive momentum. Soufiane’s message to multinationals: “If you want growth within the EU, it’s here, it’s in Eastern Europe. If you want to understand the future of Africa, start in Morocco.”

In Kenya, Nyambura stated that local companies using M-PESA are pioneering digital community fundraising that Western fintech is only now discovering.

In China, Sami Wong observes that consumers are “extremely open-minded towards new innovations, very adaptable to change, and happy to be inspired” if brands understand that emotional value matters more than quality claims.

If AI research tools remain WEIRD, they’ll miss the markets with the highest growth potential and keep fighting over super-saturated Western English-speaking markets.

What “Elsewhere” Teaches Us

If you’re in HQ, the lesson from Insights from Elsewhere is simple and slightly uncomfortable: the future of research technology and AI will be decided as much in São Paulo, Nairobi, Johannesburg, Warsaw, Mumbai, Beijing, San José, Buenos Aires, Tokyo, and Casablanca as in London or San Francisco.

And if you’re in one of those “elsewheres,” these conversations are a reminder that your day-to-day constraints and ingenious workarounds aren’t footnotes. They’re exactly what the rest of the industry needs to learn from.

Rosana Martins told me in our first episode: “Look outside. Try to create a stronger repertoire. Learn from the broader frame of reference, not only your category or business. That will make you a better insighter.”

Advice that applies to all of us: whether we’re building AI tools, commissioning research, or trying to understand consumers in an interconnected world that’s far more diverse than our training data suggests.

This article is a summary of topics discussed on the Insights from Elsewhere Podcast, a podcast hosted by Cynthia Portugal, Growth Director at Insight Platforms, featuring senior insights leaders from Latin America, Africa, Asia, Eastern Europe and the Middle East. Each episode explores how research happens beyond traditional power centres and what the global community can learn from local experts. Listen to the podcast below:

Author