Prof Dipak Gupta, Mehta School of Management, IIT Bombay
SESSION OVERVIEW
A deep dive into AI's fintech evolution, from automation to intelligence, enabling smarter decisions, scalable innovation, risk resilience, personalization, and transformative impact across financial ecosystems.
VIDEO RECORDING
AI Transformation in Fintech: From Automation to Intelligence
Detailed Summary
Dr Prasad Ramnathan opened the session, introducing himself (32 years in industry, CTO of the Technology Innovation Hub, IIT Bombay) and invited each panelist to give a brief self‑introduction.
2. Panelist Introductions
Speaker
Key points from introduction
Harsh Kumar (Poonawalla Fincorp)
25 + years experience; leads HR & AI; notes strong collaboration with the Technology Innovation Hub (TIH).
Prof Ashutosh Gupta (IIT Bombay)
Research focus: verification of AI models; stresses that accuracy alone is insufficient – models must be trusted and verifiable.
Mukund Kannan (Mphasis)
~10 years in AI; runs a Center of Excellence; builds custom AI solutions for clients; describes AI work as “a kid in a toy‑shop”.
Anjini Bharadwaj (HDFC ERGO)
20 years in digital & AI for insurance; leads the “HDPC R” group.
3. Framing the AI‑Fintech Landscape
The moderator positioned the conversation: while public attention focuses on large‑language‑model (LLM) tools (ChatGPT, Gemini, Co‑Pilot), the real fintech challenge is the purposeful deployment of AI for decision‑making, not just for chat‑based tools.
He argued that AI is not new; the sector has been automating manual processes for a decade. The current shift is toward AI‑augmented intelligent decisions.
4. Use Cases & Quantifiable Benefits in the NBFC Space
4.1 “Sathi” – End‑to‑End Document Processing
Problem: Credit analysts spent 1–1.5 hours per loan file manually verifying documents and preparing credit assessment memos (CAM).
Solution: A joint effort of Poonawalla Fincorp, TIH, and IIT‑Bombay creates an AI pipeline that (i) extracts and digitises documents, (ii) summarises content with Gen‑AI, and (iii) applies machine‑learning scoring.
Impact: Reported 15‑40 % productivity uplift. The system, originally piloted on personal‑loan workflows, now serves seven product lines.
4.2 Broader NBFC Benefits
Automation of data extraction, summarisation, and mitigant identification.
Ongoing measurement of efficiency gains as more product lines adopt the solution.
5. AI in Insurance – HDFC ERGO Perspective
Digital Penetration: 84 % of customer servicing occurs via digital channels; ≈10 % of those interactions are AI‑enabled.
Key AI/Gen‑AI Applications
Document Extraction & Summarisation – Used across policy issuance, underwriting, claim handling, and back‑office operations.
Hyper‑Personalisation – AI assists in building rapid‑to‑market products; a newly‑launched AI‑enabled policy‑issuance system reduced product‑launch timelines from months to weeks.
Performance Gains: Reported 20‑30 % efficiency improvements across underwriting and claims processing.
6. Service‑Provider Viewpoint – Mphasis
6.1 Adoption Barriers
Technology–Business Mis‑alignment – Projects often launch due to “FOMO” rather than a clear business case.
User Adoption & Change Management – Lack of stakeholder involvement leads to resistance and scaling issues.
6.2 Illustrative Projects
Digital Risk (Mortgage Underwriting) – Combines Intelligent Document Processing (IDP) with decision‑support nudges; serves the top five U.S. banks and four largest U.S. home‑loan processors.
Healthcare Discharge‑Summary Digitisation – Initial success turned problematic when a model mistakenly identified “hair‑fall” as a diagnosis, triggering negative publicity. The experience highlighted the need for guardrails and human‑in‑the‑loop validation.
6.3 Governance & Guardrails
Emphasis is building behaviour‑monitoring layers, confidence thresholds, and human‑review checkpoints before model outputs hit production.
Regulatory frameworks are increasingly supportive, providing guidelines rather than restrictions, which facilitates faster adoption when proper safeguards exist.
7. Building Trust, Verifiability & Explainability (Academic Lens)
Prof Ashutosh Gupta described two real‑world verification projects:
Multilingual Course Content Translation – Automated translation errors dropped from 1/20 sentences to 1/100, but human review remains essential.
AI‑Assisted Interview Screening – A chatbot extracts candidate data (e.g., marks) and provides source documents as evidence, ensuring auditability.
Key Takeaways
He warned against over‑reliance on large LLMs for simple tasks; simpler, interpretable models often provide comparable performance with higher transparency.
8. Governance, Guardrails & Regulatory Alignment
Mukund Kannan reiterated that post‑deployment, guardrails (behavioral patterns, ethical constraints) are crucial.
He noted a shift where regulators are becoming enablers, supplying frameworks that help organisations embed compliance into AI pipelines.
9. Future Outlook – Panelists’ Vision
Speaker
Outlook Highlights
Harsh Kumar
AI will continue to serve assistant, reviewer, and decision‑maker roles simultaneously; human oversight remains mandatory.
Mukund Kannan
Emphasises cross‑validation (multiple models + domain experts) for low‑risk use‑cases; reviewer role will persist for the foreseeable future.
Anjini Bharadwaj
Choice of AI approach depends on risk & compliance; all three roles (assistant, reviewer, decision‑maker) will coexist, selected per use‑case.
Prof Ashutosh Gupta
Calls for a unified definition of “explanation” tailored to context; sees a research opportunity to standardise explainability metrics.
Moderator (Dr Ramnathan)
AI hype will settle into a “slope of enlightenment”; generative AI will expand beyond chat‑bots to decision‑support engines that cut manual labor dramatically.
10. Audience Q&A
10.1 Is AI a Hype Cycle Like Blockchain?
Consensus: Yes, there is hype, but unlike many blockchain projects, AI already has mass‑market adoption (e.g., daily chatbot usage). The “enlightenment” phase will bring measurable business value.
10.2 Integration of Pre‑Gen‑AI FinTech Products (e.g., Oaken, ULI)
Panelists indicated that many legacy fintech solutions are being retro‑fitted with Gen‑AI layers, accelerating feature roll‑outs.
10.3 Future of Affordable, Accessible Capital
Harsh Kumar forecasted that real‑time credit‑history analysis and instant policy issuance will make capital more reachable within the next 3‑4 years.
10.4 Startup Boom – Will AI‑Centred FinTech Startups Survive?
Mukund Kannan likened the current surge to the dot‑com era: a large quantity of entrants precedes the emergence of quality winners. The market will self‑correct; strong governance and clear value propositions will differentiate survivors.
10.5 Governance for AI‑Generated Code & Liability
Prof Ashutosh Gupta stressed that human developers remain legally accountable for AI‑generated artefacts; organisations must codify responsibility policies and proportionate penalties.
10.6 Offline‑to‑Online Transition for Consultancy Firms
Brief advice: Start with a digital‑first pilot, embed AI‑assisted advisory tools, and ensure human oversight to maintain credibility.
11. Closing Remarks
The moderator thanked the panelists and audience, reiterated willingness to continue conversations after the session, and formally concluded the discussion.