Build Small. Think Big. Designing Scalable AI for Emerging Markets
- Contributor - Prashant Subhedar
- 6 min read
Share:
Somewhere in a rural town near Patna, India, Pushpa Devi’s phone lights up with an alert from her microloan app. She doesn’t read English well and struggles to navigate the interface. When she finally reaches the AI-based helpdesk and types a simple question in Hindi—“मुझे पैसा क्यों नहीं मिला?” (Why didn’t I get the money?)—she’s greeted with silence, or worse, gibberish.
Meanhwile in a dusty suburb of Jakarta, a 17-year-old named Rizky is arguing with a chatbot. He’s trying to reschedule a government test prep session offered through a national education app. The AI assistant, designed to automate replies, responds with maddening loops: “I’m sorry, I didn’t understand that. Did you mean schedule a new test?”
He didn’t. And now he’s back where he started, two hours later, slightly more cynical about the future of artificial intelligence.
So much for inclusive technology.
These aren’t edge cases. They’re the rule in much of the world, where AI is arriving with lofty intentions but missing the plot in translation. Designing scalable AI for emerging markets isn’t about compressing power into smaller packages. It’s about reimagining the entire equation.
The Scale Fallacy
In startup mythology, “scale” often means building something once and deploying it everywhere. But in real life, scale that ignores specificity ends up being a masterclass in tone-deafness.
A truly scalable AI system is one that grows into new geographies not by replication, but by adaptation.
Brazil’s favelas, Kenya’s urban peripheries, Vietnam’s rural school districts. Each bring different constraints: unstable networks, cultural nuance, limited digital literacy, and data sparsity. These aren’t bugs to be squashed. They’re the terrain.
Lessons from the Fringes
In the Philippines, when Typhoon Odette tore through Visayas in 2021, an AI-powered disaster alert system failed to reach thousands in time. Why? Because the app it depended on was too heavy for older phones, and relied on real-time connectivity in areas that hadn’t had power for days.
In Peru, a health diagnostics app trialed in rural communities was abandoned after users complained that the AI “felt cold,” refusing to accept local health idioms like “el estómago frío”, a common expression for indigestion.
In Uganda, AI-driven agri-assistance tools designed to help smallholder farmers failed to gain traction when voice prompts arrived in British-accented English, reading out plant disease names most farmers had never heard. Meanwhile, word-of-mouth advice from neighbors, though often less accurate, felt more trustworthy.
This isn’t about nostalgia for the analog. It’s a call to question AI’s default assumptions.
Why the Cloud Isn’t Always the Answer
Much of today’s AI depends on cloud infrastructure: data sent back and forth, servers whirring in distant countries, and models constantly retraining. But in regions where 3G is still the norm and data is a monthly budget line item, this just doesn’t fly.
Edge computing, where data is processed locally on the device, offers one path forward. In Cambodia, a startup building AI tools for early malaria detection switched from a cloud-based model to a solar-powered Raspberry Pi that ran basic inference locally. Not only did this reduce data costs, but it also built trust: users saw the diagnosis happen in real-time, right in front of them.
The Invisible Users
Let’s talk about language.
Despite over 400 million people speaking Hindi, most AI chatbots still fumble the language, especially regional dialects like Maithili or Bhojpuri. In Nigeria, Pidgin English, spoken by more than 75 million is barely present in NLP datasets. Tagalog, Bengali, Yoruba, and Quechua? You’re lucky if the AI recognizes them at all.
Localization is not only about translating interface text but also about respecting idioms, the cadence, the cultural shorthand that makes communication human.
If your AI sounds like a confused tourist, it probably won’t be invited back.
UX: The New Infrastructure
In many emerging markets, the interface is the infrastructure. A poorly designed UI isn’t just annoying; it can be exclusionary.
- When AI apps ask for account logins but don’t offer SMS verification, users with feature phones are locked out.
- When chatbot inputs assume perfect spelling, users typing in phonetic scripts get error messages.
- When icons look sleek but abstract, older users, like the sari-clad grandmother in Jaipur trying to access her pension are left squinting in confusion.
The best design in these contexts is often not flashy. It’s forgiving.
Modular Design, Cultural Fit
One of the smarter approaches we’ve seen comes from a Brazilian fintech called Conta Zap. Designed for favela residents often ignored by formal banks, their AI infrastructure is built modularly: identity verification via WhatsApp, AI-assisted credit scoring based on mobile recharges and utility bills, and a simplified, icon-based interface.
This isn’t minimalism for minimalism’s sake. It’s contextual elegance; removing everything that distracts, keeping only what enables.
AI and the Ghost of Bureaucracy
In many countries, public services have adopted AI to reduce workload. But AI can unintentionally mimic the worst traits of bureaucracy: opacity, rigidity, a weird obsession with forms.
In Bangladesh, residents applying for COVID aid via an AI-enabled government portal reported that if a name had a spelling error or missing prefix (Md. for Mohammad), the system rejected the request. Human clerks would’ve understood. The AI did not.
Similarly, in Pakistan, biometric facial recognition tools used to verify pensioners would fail for women wearing traditional head coverings or older individuals with changing facial features. A fix required in-person re-verification, ironically the very thing the AI was meant to eliminate.
What Users Actually Want
Meanwhile the secret: Most users don’t want to “talk to an AI.” They want to solve a problem, pay a bill, diagnose a crop disease, get a loan, talk to a doctor. If the AI helps, great. If it gets in the way, they’ll uninstall it faster than you can say “large language model.”
This is a design lesson, not a tech one. People want tools, not tutors. Empathy, not efficiency. Trust, not throughput.
Small Stories, Big Lessons
In Vietnam, an education NGO built an AI-powered reading companion for kids in underserved communities. Early versions read aloud in perfect Hanoi-accented Vietnamese. But students in Hue laughed at the voice; it sounded “posh,” “stuck-up,” like a rich cousin from the capital. So, the team recorded local teachers instead. Engagement doubled.
In Colombia, a social benefits app saw usage spike not after adding new features, but after changing the AI assistant’s name to something more “relatable.” Turns out, people trust María more than “PolicyBot.”
A Few Unsolicited Rules for AI Builders
Let’s get a bit prescriptive. If you’re designing AI for emerging markets and…
- You haven’t tested it offline? You’re not done.
- You need a login but haven’t offered Facebook, WhatsApp, or SMS sign-in? Try again.
- Your chatbot can’t handle typos, mixed languages, or voice input? That’s not AI. That’s AI lite.
- You think the user’s patience is infinite? Bless your heart.
Designing for constraint doesn’t mean dumbing things down. It means sharpening your empathy until it’s as precise as your code.
Why Bang Design Cares About All This
Because we’ve seen what happens when good tech meets bad context.
At Bang Design, we’ve worked with clients building healthcare apps, credit platforms and AI assisted robots. And the patterns repeat: the best outcomes come not from the smartest code, but from the sharpest design questions.
- Who are we excluding without realizing it?
- What assumptions are we baking into the system?
- Does this work when the power goes out?
- Does this work when someone’s grandparent tries it?
Design isn’t the layer you paint on top. It’s the conversation you begin with.
If You’ve Read This Far…
You already care. You understand that the future of AI is more than silicon and syntax; it’s sweat, soil, signal bars, and the small, strange ways people use technology when nobody’s watching.
And if you’re building something for them, the overlooked, the offline, the newly online, we’d love to help.
At Bang Design, we go beyond building products. We build pathways. Context-aware, culturally intelligent, scalable-by-default design for the people most designers forget.
Let’s make it work where it matters most.