How the Global South Is Accelerating AI Adoption: Finance Sector Insights

Contents

Executive Summary

This panel discussion explores how the Global South—particularly India—is accelerating AI adoption in the financial services sector. Rather than focusing on AI capability breakthroughs, the conversation emphasizes that institutional trust, regulatory frameworks, and responsible deployment are the critical barriers and enablers for scaling AI in finance. Speakers highlight how regulatory innovation, infrastructure gaps, and inclusive AI design can unlock financial services for hundreds of millions of underserved individuals in emerging markets.

Key Takeaways

Institutional Trust > Raw Capability: The financial sector's success with AI stems from risk management culture, auditability frameworks, and regulatory alignment—not from superior models. These lessons are exportable to other regulated sectors (healthcare, insurance, utilities).
India's AI Opportunity Is Uniquely About Inclusion, Not Optimization: While the West uses AI for margin enhancement, India can use AI to overcome structural barriers (thin files, language, accessibility) and include 1+ billion people in formal financial systems. This is both a massive business opportunity and a development imperative.
Principles-Based Regulation Outperforms Prescriptive Rules for Emerging Technology: The RBI's tech-neutral, outcome-focused approach (the "Seven Sutras") enables experimentation while managing material risks. This contrasts with technology-specific regulations that quickly become obsolete.
Voice & Conversational AI Are Not Luxuries in the Global South—They're Necessities: Designing apps for English-speaking, digitally-savvy users excludes billions. Conversational, voice-first, multilingual AI is the primary mechanism to unlock the next wave of commerce, finance, and services adoption.
The Next Regulatory Frontier Is Deployer Accountability, Not Developer Regulation: RBI cannot regulate AI developers directly. Instead, it holds regulated financial institutions accountable for transparency, bias mitigation, model auditability, and customer protection—making deployers the primary custodians of trustworthy AI.

Key Topics Covered

Institutional AI vs. Frontier AI: Shift from capability-focused to trust-focused deployment in regulated financial services
Regulatory Approaches: India's Reserve Bank (RBI) principles-based, tech-neutral regulatory framework for enabling responsible AI adoption
Use Cases in Finance: Fraud detection, payments, risk assessment, compliance, and customer engagement through conversational AI
Financial Inclusion: Using AI to address "thin file" problems, voice-based banking, and accessibility for underserved populations
Infrastructure Gaps: Data residency requirements, lack of cutting-edge model deployment in India, compute accessibility for smaller fintech firms
Risk Management: Model auditability, bias mitigation, hallucination concerns, and human-in-the-loop safeguards
Agentic Commerce & Conversational AI: Voice-first, multilingual payment and commerce systems designed for non-digitally-native populations
Public-Private Partnerships: Regulatory sandboxes, AI sandboxes, and collaborative frameworks between government and industry
Unit Economics: How AI can reduce operational expenses (3–5% of OpEx) and improve productivity in financial services
Data Residency & Sovereignty: Deployment challenges around geopolitical data governance and the emergence of Indian foundational models

Key Points & Insights

Trust, Not Capability, Is the Constraint: Financial institutions absorb only systems they trust; regulators can only enable what they can supervise. As capabilities commoditize, legitimacy becomes the scarce attribute determining competitive advantage.
RBI's Principles-Based Approach Enables Responsible Innovation: Rather than prescriptive technology regulations, India's central bank focuses on outcomes (consumer protection, transparency, auditability) and principles that remain agnostic to the specific technology used. This has allowed banks to experiment widely while managing proportionate risk.
AI Addresses Structural Problems in Indian Finance:
- Unit economics: 60–100 billion dollars spent annually on OpEx (3–5% of India's $2 trillion credit market) can be significantly reduced
- Thin files: Unstructured data + AI enables underwriting for ~1.3 billion individuals lacking formal credit histories
- Reach: Voice-based, conversational banking removes accessibility barriers for non-English-speaking, low-literacy populations
The "Thin File" Problem Is India's Biggest AI Opportunity: India's large unbanked population lacks sufficient data points for traditional underwriting. AI's ability to process unstructured data (call records, transaction patterns, location data) can "thicken" files and unlock formal credit access for hundreds of millions.
Human-in-the-Loop Is Non-Negotiable for High-Risk Financial Decisions: Best practices emerging from fintech firms include keeping humans in decision loops for underwriting, lending, and high-value transactions, while automating lower-risk tasks like document summarization or fraud detection support.
Hallucination & Opacity Pose Fundamental Risks for LLM Deployment in Finance: Even 1–2% error rates in financial transactions create massive liability. LLMs by design can generate false information; until models guarantee near-zero hallucination rates, most regulated financial companies cannot deploy them on customer-facing high-stakes decisions.
Data Residency & Infrastructure Governance Are Underestimated Barriers: Many cutting-edge AI models from the West don't meet India's data residency requirements. Open-source models come primarily from China. This creates a deployment gap for regulated financial companies that cannot use cloud-hosted global models. Emergence of Indian sovereign language models is critical.
Agentic Commerce Unlocks the "Next Billion" Online Consumers: Only ~10 million of 300–400 million Indians on UPI do online shopping (and they represent 70% of e-commerce volume). Conversational, voice-first, agent-based commerce—designed for how Indians actually shop (via dialogue with retailers, not self-serve apps)—could unlock 290+ million consumers currently excluded by Western UI/UX patterns.
AI Dramatically Improves Financial Inclusion & Consumer Protection: Personal AI agents can protect vulnerable populations (elderly, less educated) from fraud, misselling, and predatory financial products by providing real-time, intelligent advice. Simultaneously, AI can reduce collection friction (empathetic agents vs. frustrated human collectors), improving both customer experience and recovery rates.
Compute & Talent Access Are Bigger Constraints Than Regulation: While regulatory frameworks matter, smaller fintech firms lack affordable compute infrastructure and access to datasets needed for model training. Proposed "AI sandboxes" (providing shared compute, data, and benchmarking tools) are critical to democratize AI innovation beyond large banks.

Notable Quotes or Statements

John Tass Parker (JP Morgan Chase): "In finance, trust is not a feature. It's actually the business model... Increasingly, those that can demonstrate reliability, auditability, resilience, not just model performance will be the ones that are rewarded."
Suvendu Party (RBI): "We are not exactly regulating AI but enabling responsible adoption of AI... Our whole approach is optimistic. We want people to experiment, adopt it responsibly but think creatively in terms of liability framework."
Ashoto (Fintech Investor): "Just 10 million [users] in a country of a billion and a half do 70% of all commerce online... The rest of India needs conversations. Indians shop on retailers where you go and talk... We are conversational in commerce."
Harshel Choksi (Razer Pay): "Imagine a year, two years from now all of us will have an AI agent who is essentially your assistant working with you... having that intelligence available to every person on demand is a massive advantage."
Suvendu Party (RBI): "From a typically black box associated with AI systems... we would like this to be a not a black box but a glass box. Customers should be knowing what they are getting."
Harshel Choksi: "It's okay if the system fails 10% of the time, but it should not be wrong 10% of the time... if [an LLM] gives the wrong analysis and you deliver that information to the customer and say a payment is successful but actually isn't, even if it happens 1% of the time, it creates a massive issue."

Speakers & Organizations Mentioned

Speaker	Role/Organization	Domain
John Tass Parker	Head of Policy Partnerships	JP Morgan Chase
Suvendu Party (Wendu G)	RBI Official, External Expert Committee Member	Reserve Bank of India (Regulator)
Terara	AI/Technology Executive	JP Morgan Chase
Ashoto Mukherjee	Fintech Investor	Investor/Advisory
Harshel Choksi	Co-founder/Executive	Razer Pay (Payments/Fintech)
Barat	Panel Moderator	(Affiliation not specified; appears to be moderator/facilitator)

Institutions/Entities Referenced:

Reserve Bank of India (RBI)
JP Morgan Chase
State Bank of India (SBI)
HDFC Bank
Razer Pay
Google (implied—Chat GPT referenced as available tool)
Indian fintechs and NBFC ecosystem
Government of India
Self-Regulatory Organizations (SROs)

Technical Concepts & Resources

AI Frameworks & Approaches

Frontier AI vs. Institutional AI: Paradigm shift from capability-focused to deployment/trust-focused
Model Risk Management: Oversight frameworks specific to AI systems in regulated financial institutions
Explainability & Auditability: Requirements for "glass box" (transparent) AI vs. "black box" systems
Human-in-the-Loop (HITL): Keeping humans in decision chains for high-risk transactions

Technical Challenges Referenced

Hallucinations in Large Language Models (LLMs): Models generating false/incorrect information, even at 1–2% rates, creating unacceptable liability risk in finance
Model Drift & Degradation: Performance degradation over time requiring periodic validation and retraining
Bias Detection & Mitigation: Systematic removal of demographic/systematic biases in model outputs
Black Box Problem: Opacity in how AI systems make decisions; push toward interpretable/explainable AI

Data & Infrastructure

Thin File Problem: Insufficient historical data for underwriting decisions; AI enables unstructured data processing
Data Residency Requirements: Regulatory mandates that sensitive financial data remain within national borders
Compute Infrastructure Gaps: Scarcity of affordable, accessible compute for smaller fintech firms
Foundational Models: Mention of Indian sovereign language models announced at the summit (to address data residency & vendor lock-in)
Regulatory Sandbox Framework: Since 2019, RBI allows firms to experiment with regulatory relaxation/dispensation in controlled environments
AI Sandbox (proposed): Shared infrastructure providing compute, data, and benchmarking tools to democratize AI innovation for smaller institutions

Emerging Use Cases

Agentic Commerce: Voice-first, conversational, agent-based commerce systems
Conversational Banking: Voice/chat-based payment, account opening, and financial advice
Biometric Payments: Moving from OTP-based authentication to biometric-enabled payments (mentioned in UPI context)
AI-Driven Collections: Empathetic, context-aware collection agents handling early-stage loan recovery (60–70% of first-30-day collections)
Fraud & Scam Remediation: Real-time detection and prevention using AI
Underwriting Models: Alternative credit scoring using unstructured data for thin-file populations

Tools & Systems Mentioned

Mule Hunter.ai: RBI-supported AI system for model validation, implemented across 26+ banks
UPI (Unified Payments Interface): India's foundational digital payment infrastructure; 300–400 million users
Chat GPT: Referenced as an accessible tool for non-technical users (business planning, consumer advice)

Policy/Regulatory Frameworks

Seven Sutras/Principles (RBI): Generic AI governance principles adopted government-wide:
- Innovation prioritized over restraint (with responsible deployment guardrails)
- Principles-based (outcome-focused), not technology-specific
DPDP (Data Protection): Referenced guard rails around sensitive data handling
IT Outsourcing Guidelines: Existing RBI guidance covering third-party dependencies and concentration risk
Consumer Protection Guidelines: Already provide baseline safety frameworks
Finquery & Finteract: Monthly RBI-industry engagement forums with ~2,000+ entities over 1.5 years
Regulatory Sandbox: Framework enabling controlled experimentation with regulatory relief

Research/Assessment Referenced

RBI Dipstick Survey: Survey of ~600 banks/NBFCs + ~75 deep-engagement interviews to assess AI adoption, potential, and challenges
FreeAI Committee Report (RBI): Comprehensive policy recommendations released August 2024

Word Count: ~1,950 | Last Updated: Based on live event transcript (no date stamp provided in transcript)