India AI Impact Summit Insights
All sessions

AI for ALL Challenge & Panel on Leveraging AI for Development in the Global South

Contents

Executive Summary

This AI summit panel discusses the critical challenges and opportunities for AI startups building for impact in the Global South, specifically focusing on healthcare, agriculture, and fintech sectors. The conversation reveals that most AI startups fail not due to poor technology but due to inability to cross the "valley of death"—the gap between prototype and commercial adoption—requiring sector-specific strategies, patient capital, and deep ecosystem understanding.

Key Takeaways

  1. Don't Build for the End User; Build for the Trust Layer: In agriculture, farmers trust intermediaries, not direct digital interactions. In healthcare, doctors must be partners in design, not just users. In fintech, different institution types have different requirements. Understand who actually pulls the wallet and design for them.

  2. Real-World Data Access Determines Success More Than Technology: Having access to real patient, farmer, or financial data early is more valuable than a perfect algorithm. Plan your go-to-market to accommodate data limitations (e.g., start with consumers in healthcare rather than diagnostics).

  3. Sector-Specific Strategies Are Non-Negotiable; Generic AI Won't Work: One-size-fits-all AI pitches fail. Healthcare requires regulatory alignment and doctor workflows; agriculture requires hyperlocal contextualization; fintech requires understanding multiple institution types and regulatory bodies. Each sector has its own "valley of death" characteristics.

  4. Patient Capital + Clear Monetization Path > Fast Growth: The most successful startups balance growth ambitions with disciplined unit economics and realistic timelines. Overselling release cadence or taking capital without clear use-allocation breaks trust with both customers and investors.

  5. Government Procurement is the Hidden Scaling Path: Once performance and economic benefit are proven (via assessments like HTA or competitions), government procurement unlocks billion-person scale without the customization burden of enterprise sales. Plan to navigate public procurement policy early.

Key Topics Covered

  • Valley of Death in AI Startups: The critical phase where startups face high cash burn and negative cash flow before generating significant revenue
  • Sector-Specific Challenges:
    • Healthcare: Model accuracy on real-world patient data, regulatory frameworks, user adoption by doctors
    • Agriculture: Hyperlocal context, farmer trust layers, unit economics across value chains
    • Fintech: Real-world data complexity, regulatory uncertainty, talent gaps combining finance + AI
  • Scaling Strategies: From product-market fit to procurement pathways, particularly government procurement
  • Impact Definition: Different metrics across sectors (user base vs. enterprise adoption vs. government service delivery)
  • Data Infrastructure: Access to real patient/farmer/financial data as fundamental blocker
  • AI Product Development Platforms: Infrastructure to democratize product development (Better Labs case study)
  • Investment & Fundraising: Investor evaluation criteria, funding stages, and capital structuring for AI companies
  • Cybersecurity & Innovation: Integration of security into AI platform development

Key Points & Insights

  1. The "Valley of Death" Affects 90% of Ventures: According to Harvard Business Review, 90% of new ventures experience this critical early phase where cash burn exceeds revenue. AI startups must overcome this through clear monetization pathways and strategic capital allocation rather than technology excellence alone.

  2. Real-World Data Access is Non-Negotiable: Most AI startups fail because their models are trained on public/synthetic datasets but fail on real-world data. In healthcare, if the patient population in training data doesn't match the deployment context, the model will fail. This requires either early access to real data or a go-to-market strategy that accommodates this limitation.

  3. Healthcare AI Requires Doctor Investment in Workflow: AI diagnostic tools only succeed if they answer the question: "Why should a doctor pull money out of their wallet to use this?" If doctors aren't invested in the proposed workflow, adoption fails. Regulatory constraints (non-doctors cannot diagnose) further limit addressable markets unless pivoting to consumer wellness (less regulated).

  4. Agriculture is Hyperlocal and Trust-Dependent: Indian farmers hold ~1 hectare average plots with extreme seasonal variation and risk aversion. AI models must be curated per farm/farmer, not generic. Critically, models should be built for trusted intermediaries (village-level entrepreneurs, grassroot organizations) rather than directly for farmers—a "fidgetital" (hybrid digital-physical) approach, not purely digital.

  5. Accuracy is Not the Priority—Good Enough is: In agriculture, 70-80% accuracy is sufficient to begin testing; perfectionism kills adoption. Co-development with farmers improves models faster than lab-based optimization. "Perfection is the enemy of good."

  6. Fintech Must Account for Regulatory Uncertainty: Financial sector startups face constantly changing regulations and high customer risk aversion due to cyber concerns. Models must be built for real-world scenarios (different customer segments, banks, regulatory bodies), not just internal UAT databases. Talent bottleneck: finding individuals with both AI and financial domain expertise is extremely difficult.

  7. Meaningful Scale Differs by Sector:

    • Healthcare personal wellness: user base metrics (most scalable, least regulated)
    • Healthcare enterprise: workflow efficiency improvements for large organizations
    • Healthcare government procurement: infinite scale potential but requires perfect product-market fit, full regulatory alignment, and navigation of public procurement policy
    • Agriculture: 120-150M small-holder farmers + 7-8B consumers = massive scale, but requires solving trust, distribution, unit economics, and data infrastructure barriers

  8. Procurement Can Be a Proving Ground: Government procurement (e.g., India AI Mission competitions, ICMR chest X-ray assessments) guarantees both validation and revenue if companies meet performance + economic benefit benchmarks. This bypasses L1 pricing barriers through health technology assessments.

  9. Weak Data Infrastructure is India's Biggest Opportunity and Barrier: Proprietary Indian data exists (agriculture universities, ICR institutions, weather stations, satellite imagery, mandis) but remains un-harnessed. Open-sourcing data stacks (e.g., Agristack with 70M farmer IDs) could unlock multiple use cases, but data collection, processing, and cleaning remain expensive—best handled by government.

  10. Impact is Tiered: Impact progresses from enablement (easier access) → efficiency improvements → solving previously unsolvable problems (highest impact). Only AI-designed antibiotics and novel drug discovery represent true outcome-changing impact so far; most AI health solutions remain in enablement/efficiency tiers.

Notable Quotes or Statements

  • Dr. Anurag Agarwal (Health): "If doctors who are required to establish any diagnosis are not invested in the workflow that you are thinking of for using your product, it will die."

  • Hemendra Mathur (Agriculture): "Don't chase perfection. Perfection is the enemy of good. 70-80% accuracy is good enough to start testing the waters."

  • Hemendra Mathur (Agriculture): "Don't build for the farmer; build for the people who are trusted by the farmer." (On the "fidgetital" model)

  • Ravindra Misra (Fintech): "Suppose you try to swim, learn swimming in a very small pond... then you are thrown into a sea. You don't know about the winds, the gust, the velocity. That's scaling up."

  • Dr. Kaneka Singh Roa (Bioprinting): "Real technology, if you really want to be an AI startup out of the AI bubble, you need to have patience. You need to develop the technology."

  • Yodra Bajora (Investor): "Strong technology plus clear evidence gives you more leverage than you've ever had before." (On the unique window for AI founders)

  • Jiraj Bharat Patil (Kalari Capital): "What really convinces us fast is clarity and conviction that founders have. Do you have a clear pulse of where you're operating, who you're competing with?"

  • Jiraj Bharat Patil: "It's never a quantitative bet. It's more a qualitative bet... You're basically picking who you're going to be working with rather than who's financing your company."

  • Dr. Bina Ray (Better Labs): "Democratization of innovation—how do you shorten the period of research and development which was largely in large corporates... and make it available to everybody?"

Speakers & Organizations Mentioned

Panel Moderators & Speakers:

  • Magna Bal (Director, ESY Center, moderator)
  • Dr. Anurag Agarwal (Dean of Biosciences and Health Research, Head of Kittita Center for Digital Health, Ashoka University)
  • Hemendra Mathur (Venture Partner, Bharat Innovation Fund)
  • Ravindra Misra (Chief Product Officer, OPPL Innovate)
  • Dr. Kaneka Singh Roa (Founder & CEO, Sel Private Limited)

Investors:

  • Yodra Bajora (Founder, C Experts)
  • Jiraj Bharat Patil (Partner, Kalari Capital)

Entrepreneurs:

  • Dr. Bina Ray (Founder, Better Labs) — AI-powered product development platform for physical products
  • Sonel Khana (Co-founder, Secure Bling) — Cybersecurity for APIs/applications

Institutions/Initiatives:

  • Niti Aayog (Government of India planning body)
  • Ashoka University
  • International Labour Organization (ILO)
  • ICMR (Indian Council of Medical Research)
  • Kalari Capital (VC firm with women founder focus: "CXO by Kalari")
  • India AI Mission (government competition & procurement initiative)
  • Agristack (70M farmer IDs issued)
  • NIFTIM (National Institute of Food Technology & Management)
  • Akare (health tech company, cited as scaling example)

Technical Concepts & Resources

Data & Infrastructure:

  • Agristack: Open-source agricultural data stack with 70M farmer IDs; opportunity to open-source for multiple use cases
  • Satellite imagery + spectrometry: Used for fair quality assessment in agriculture (grades: physical, chemical, biochemical)
  • Public health data: ICMR assessments, IMD (Indian Meteorological Department) weather data
  • Single Source of Truth (SSOT): Banking/fintech data centralization to eliminate subjective lending decisions

Frameworks & Models:

  • Health Technology Assessment (HTA): Evaluates both performance and economic benefit; used by ICMR to certify chest X-ray solutions
  • FRAME Framework (for investor evaluation):
    • Flow: Money flow in market, willingness to pay
    • Risk: Risk assessment
    • Access: How investors reach founders
    • Market: AI stack positioning and margins
    • Environment: Regulatory/ecosystem context

Product Development:

  • Better Labs Platform: Modular lab access (formulation, compliance, manufacturing) for food/wellness products; enables founders to design from compliance, not post-hoc
  • Regulatory compliance labs: Indian regulatory data currently available; European/global expansion planned

Regulatory & Policy:

  • Key Fact Statement (KFS): Required transparency disclosure for fintech customers (APR/cost)
  • RBI AI Banking Strategy: Recently released; model for sector-specific AI strategies (Health Minister launching AI in Health strategy)
  • Small Business Innovation Research (SBIR): US initiative ensuring startup access to government procurement (cited as example for India to replicate)

Investment Stages:

  • Angel/Pre-Seed: Product MVP, early revenue, early go-to-market
  • Seed: Product risk complete, alpha/beta ready, prototyping done, market understanding established
  • Series A+: Product complete, team in place, established markets, capital for acceleration

Metrics & Evaluation:

  • Repeatability: Sales repeatability across customers; degree of customization required
  • Unit Economics: CAC (Customer Acquisition Cost), retention, gross margins
  • Month-on-Month (MoM) / Quarter-on-Quarter (QoQ) Growth: 50% growth signals traction
  • Gross Margins: Software businesses require healthy margins; high compute costs (e.g., continuous transcription) break margin models

Key Risk Areas:

  • AI Fatigue in Healthcare: Radiologists seeing false positives from AI tools → both ignoring AI and over-relying on AI ("rubber stamping")
  • Bias in Financial Models: Unstructured data leading to biased lending decisions; objective models required for trust
  • Cybersecurity: APIs and AI models as new attack vectors; continuous scanning needed (Secure Bling model)

Conclusion

The summit reveals that AI impact in the Global South is not primarily a technology problem but an ecosystem, data access, and market understanding problem. Success requires: (1) deep sector expertise and stakeholder mapping, (2) patient capital aligned with realistic timelines, (3) access to real-world data, (4) focus on user adoption incentives over technical perfection, and (5) navigation of regulatory and procurement frameworks. India's demographic dividend and proprietary data assets present a unique opportunity to build globally competitive AI solutions, but only if startups and capital holders adopt sector-specific, customer-centric approaches rather than generic AI narratives.