Yixiang Xu, Rupalee Ruchismita, Ganesh Iyer


FINO Payments Bank

Anand Bhatia

  • Chief Marketing Officer

Atul Choudhari

  • Assistant Manager – Analytics


Retail Finance Distribution (ReFinD) Research Initiative
Center for Equity, Gender and Leadership, UC Berkeley
Clausen Center for International Business and Policy, UC Berkeley
Fisher Center for Business Analytics
Experimental Social Science Laboratory, UC Berkeley

Researcher Assistants

Yisen Du

  • Chinese University of Hong Kong, Shenzhen

Nikhil Goyal

  • Indian Institute of Technology-KGP

Varun Rao

  • Hindu College, University of Delhi

In emerging markets, local merchants often act as financial agents, providing crucial access to banking services for underserved communities. By significantly reducing the cost of financial access, these fintech-enabled merchants offer a promising path for financial inclusion.

However, this channel suffers from poor merchant retention rates. A range of factors, including the high cost of sourcing both firm and market information impede merchant profitability. Indeed, of the 3,000 surveyed merchants, more than 50% found it difficult to acquire necessary know-how to provide reliable banking services. For the fintech firms, providing timely assistance to a large number of dispersed merchants has proven to be financially unsustainable, creating a demand for low-cost technologies for network support.

To mitigate these challenges, we leverage Generative AI to design personalized, interactive and digestible ‘information journeys’ covering a wide range of business-relevant content for merchants. To further enhance the information acquisition experience for merchants, we use Large Language Models (LLM) to create multimedia content with text, audio and video content journeys in the local language.

Research Methods
In this research, we undertake Randomized Controlled Field Trials (RCTs) in AI. Through running a large-scale field experiment involving 80,000 FINO merchants in India, we causally estimate the direct impact of AI-augmented interactive content on merchant engagement and the indirect impact on enhancing merchant performance and advancing financial inclusion at large. 
For this experiment, we use WhatsApp as the channel of delivery, which has over two billion users and a popular means of communication in developing countries as it avoids data carrier SMS costs and works in low data connections. Our experiment assesses a co-pilot model of human and AI for content personalization, by tailoring content for specific merchant needs.
AI in emerging markets

This study offers insights on General-purpose AI’s ability to undertake content personalization, especially in data-scare emerging markets facing cold start problems.

Oversight for equity and quality

We investigate conditions where human-AI collaboration versus AI-delegation more effectively aligns AI content personalization with merchant needs. In addition to  equity concerns and ongoing AI quality concerns, and emerging market infrastructure limitations were factors which supported a decision to avoid a direct AI supported merchant chatbot, but one intermediated by humans/researchers.

Business and Policy Implications

For the fintech firms, we hope this experiment has insights on contexts where merchant support services can be handled by AI supported solutions. We also offer insights into the extent to which the mainstream Generative AI is generalizable to emerging market contexts and the necessity of developing localized GAI solutions for these markets.