Consumer Behaviour Research

Mainstream digital commerce was designed around a buyer who reads product pages, compares prices, and trusts institutional systems. EVAA is studying whether that model fits the majority of the world's consumers, or whether purchasing in much of the Global South is mediated through relationships, recommendation, and earned trust. This brief frames the research questions we are asking and the published literature our thesis builds on. It is a thinking document, not a results report.

The problem we are studying

Digital commerce was designed around a very specific type of buyer: someone who is literate in the platform's language, has reliable internet access, trusts institutional systems, compares products independently, and makes decisions based on information alone. That buyer exists, but represents a minority of the world's population.

For much of South Asia, Southeast Asia, Sub-Saharan Africa, and Latin America, the literature on relational commerce, base-of-the-pyramid markets, and informal economies suggests that purchasing decisions follow a different logic. Consumers often buy through people they know and rely on recommendation, relationship, and earned trust rather than product pages and star ratings. EVAA's research question is whether that pattern holds robustly across categories and markets, and what it implies for how AI agents should be designed.

What we are studying

How trust is established between buyers and sellers in markets with low formal-platform adoption.
The role of language, dialect, and cultural familiarity in purchase intent and conversion.
Whether AI agents can plausibly approximate the trust dynamics of a known human recommender.
The gap between digital-commerce adoption rates and underlying consumer readiness to transact online with strangers.
How voice-first interfaces may change purchasing behaviour in low-bandwidth and lower-literacy environments.

What we expect to find, and why

Trust precedes information. We hypothesize that personal recommendation from a known contact outweighs product specifications and price comparison for most Global South buyers. This is consistent with Hofstede's work on collectivist cultures and Prahalad's observations on bottom-of-the-pyramid markets, where in-group trust appears to be a stronger purchase signal than abstract product information.

Language is not just translation. We hypothesize that delivering product information in a buyer's native dialect, with culturally appropriate framing rather than literal translation, raises purchase intent. The premise builds on Donner's research on mobile internet adoption and Arora's work on the next-billion-user experience, both of which point to belonging cues and not only intelligibility as drivers of digital trust.

Relationships are the commerce layer. We hypothesize that in many Global South markets the intermediary is the actual commerce infrastructure. Local shop owners, WhatsApp contacts, and community figures appear to carry the trust layer that mainstream digital platforms try to skip. Jack and Suri's work on M-PESA and Donner's After Access both describe this intermediary-as-infrastructure pattern.

AI agents may become trust proxies. We hypothesize that culturally tuned AI agents, speaking in familiar language and tone, can over time approximate some of the trust dynamics of a known human recommender. This is the core open question. The hypothesis builds on Bickmore's work on relational agents in health, which suggests that continuity, memory, and consistent persona are mechanisms by which software can develop something like rapport with users.

How we plan to study this

Qualitative interviews with consumers in selected Global South markets to map purchase decision logic.
Analysis of decision patterns in chat-first commerce environments such as WhatsApp and Telegram.
Comparative testing of multilingual AI agents against generic agents across selected market segments.
Longitudinal trust-building measurement over multi-week interaction windows.
Cross-referencing observed patterns with the published literature on relational and informal commerce.

What this could mean for AI commerce

If the hypotheses above hold, the implication is that there is a large underserved market for AI agents designed not as search or comparison tools, but as relationship proxies. An AI that speaks a buyer's language, remembers their preferences, and earns their trust over time would behave less like a chatbot and more like a digital equivalent of a trusted local recommender.

The broader question is whether trust is a property that can be built by software at all, or whether it is intrinsically human. Our working position is that some components of trust, including consistency, memory, language register, and non-judgmental tone, are mechanisable, while others may not be. The point of this research is to find out which is which.

Why 4 Billion People Don't Buy the Way the Internet Thinks They Do