AI Masterclass in Financial Services

Abstract

The masterclass opened with a strategic overview of how artificial intelligence is accelerating across the banking, capital‑markets and insurance (BFSI) sector, highlighting the shift from simple chat‑bots to multi‑modal, action‑oriented agents. Babu Unnikrishnan presented market‑level data, a recent TCS customer‑adoption survey and the launch of a new AI‑focused data‑center, before demonstrating a multimodal Voice‑AI solution built for Tata Capital. The session then moved to the practicalities of fine‑tuning, model governance, integration, and scaling AI in a highly regulated environment. Kavita guided participants through an “opportunity‑canvas” framework and a rapid‑prototype workflow that enables business users to design, test and hand‑off AI solutions, with live instructions for a hands‑on lab.

Detailed Summary

• Opening remarks (Babu Unnikrishnan) framed the session as a two‑part journey: first a high‑level view of AI adoption in financial services, then a proof‑of‑value lab where attendees would design, prototype and test an AI solution on their laptops.
• The presenter stressed that AI is entering a hyper‑acceleration phase and that the industry is moving from “chat‑bot” style interfaces to agentic, action‑driven experiences.

2. AI Trends Shaping BFSI

Key observation	Explanation
Multi‑model adoption	Enterprises are expected to combine several foundation models (LLMs, vision models, speech models) within a single workflow.
Context engineering	Building the right context (prompt, data, domain knowledge) before invoking a model is now the most critical step.
Rapid model iteration	Recent releases (e.g., “Codex 5.3” and “Sonnet 4.6”) illustrate how quickly model capabilities evolve—new generations appear within minutes of each other.
Vertical convergence	Investment is no longer limited to hardware; TCS has announced an AI‑focused data center (a “one‑giga‑hour” facility) that will be rolled out over the next five years, linking compute, storage, and specialized models.

3. Survey Insights – Where the Industry Stands

• Scope: Survey of 170 TCS customers (global + India).
• Adoption: ~90 % of respondents have some form of AI in place (banking, capital markets, insurance).
• Maturity: Roughly 60 % describe their AI projects as fairly advanced.
• Model preferences: 50 % demand enterprise‑specific or fine‑tuned models to meet regulatory, security, and accuracy requirements.
• Outcome focus: Majority stress the need for business‑outcome metrics—not just POCs, but measurable ROI.

4. Voice‑AI Demonstration – Tata Capital Use‑Case

• A multimodal Voice‑AI prototype was played out, simulating a loan‑offering conversation with a fictional customer “Omkar Sharma”.

• Highlights of the demo:

  1. Natural, bilingual interaction (English + regional language cues).
  2. Dynamic data capture (loan amount, interest rate, tenure, EMI).
  3. Form‑filling assistance (collecting personal and business details).
  4. End‑to‑end flow ending with a loan‑approval summary.

• The presenter used the demo to illustrate:

  • Agentic, multimodal design (speech + language model).
  • Localised model adaptation (India‑specific “Sarvam” model) to handle linguistic diversity.
  • Scalability through multi‑channel integration (voice, chat, web).

5. Multi‑Model Architecture & Fine‑Tuning

• Fine‑tuning was positioned as the next growth lever for BFSI, expected to lift model accuracy within 6‑9 months.

• Hybrid approach: combine large, general‑purpose LLMs with domain‑specific fine‑tuned models.

• Agentic workflows (partial autonomy) will emerge, but full autonomy is still distant.

• The speaker stressed three pillars for successful fine‑tuning:

  1. Instruction‑set design (clear, domain‑specific prompts).
  2. Iterative evaluation (continuous metrics, validation loops).
  3. Governance & compliance (guardrails for bias, security, regulatory constraints).

• Real‑world anecdotes:

  • Litigation‑prediction – a base model vs. a fine‑tuned model showed markedly higher accuracy after iterative domain‑specific training.
  • Insurance claim processing – a fine‑tuned model helped surface fraud indicators behind the scenes.

6. Operational Challenges & Scaling AI

Challenge	Insight / Recommendation
Accuracy & compliance	Ongoing validation, especially for regulated financial data, is mandatory.
Enterprise integration	AI must hook into CRM, core‑banking, lending platforms; isolated pilots rarely deliver ROI.
Process‑centric ROI	Value emerges when AI is embedded in end‑to‑end workflows, not just discrete tasks.
Scaling beyond POCs	Requires legal, compliance, and governance alignment—technology alone is insufficient.
Business‑team engagement	Early involvement of business users ensures relevance and adoption.

7. Business‑Led AI Design Framework (Opportunity Canvas)

• Goal: Empower business users to identify, scope and prototype AI opportunities without heavy reliance on engineering.

• Steps Introduced by Kavita:

  1. Define the Business Challenge – e.g., portfolio managers spending excessive time compiling quarterly performance reports.
  2. Opportunity Canvas – a structured set of prompts that capture pain points, data sources, expected impact, and solution direction.
  3. Ideation with AI – business users ask AI to research public data (Yahoo Finance, news sentiment) and generate a concise summary.
  4. Rapid Prototyping – using the WIP‑coding design tool (integrating Lovable & Google AI Studio) to convert prompts into a visual UI prototype within minutes.
  5. Validation & Hand‑off – the prototype is shown to stakeholders; once approved, code can be auto‑generated and integrated with back‑end services.

• The canvas approach is positioned as a right‑to‑left outcome‑first methodology, contrasting with traditional left‑to‑right process mapping.

8. AI Workbench & Hands‑On Lab

• The AI workbench is a multi‑tool environment that aggregates voice, language, and vision capabilities plus a low‑code orchestration layer.

• Participants received handbooks pre‑loaded on their laptops, each containing a “challenge card” (the portfolio‑manager use‑case).

• Mentor support was announced: each table had a dedicated TCS mentor to guide through discovery, design, and prototype phases.

• The lab workflow:

  1. Log into the workspace (using the table number).
  2. Complete the discovery stage – capture context, pain points, data availability.
  3. Run the AI‑driven ideation – generate the opportunity canvas.
  4. Create a low‑code UI prototype (via the WIP‑coding tool).
  5. Review, iterate, and prepare for scale – discuss guardrails, model selection, and integration points.

• The session closed with encouragement to experiment, reiterating that the hands‑on experience mirrors a real-world pipeline from idea to production.

9. Closing Remarks & Announcements

• AI Data‑Center – reminder of the forthcoming “one‑giga‑hour” data center, underscoring TCS’s infrastructure commitment.
• Future AI Days – TCS will continue hosting immersive “AI Days” for business teams to explore tools, governance, and best practices.
• Call to Action – participants were urged to apply the outcome‑backward design in their own organisations and to leverage the mentors and workbench for rapid proof‑of‑value development.

Key Takeaways

• AI is moving from chatbot‑style interfaces to agentic, action‑oriented experiences across BFSI.
• Multi‑model architectures and context engineering are now essential; a single LLM is rarely sufficient.
• Fine‑tuning with domain‑specific data will become a primary differentiator, especially for regulated use‑cases.
• 90 % of surveyed BFSI customers have AI in production; 60 % consider it fairly advanced. Yet half of them still demand enterprise‑specific models.
• Voice‑AI can handle multilingual, end‑to‑end loan‑orchestration, demonstrating the practical impact of multimodal agents.
• Scaling AI requires more than technology – legal, compliance, and governance alignment are mandatory.
• Business‑led AI design (Opportunity Canvas) empowers non‑technical users to define, prototype, and validate AI solutions quickly.
• The AI workbench + low‑code prototyping can shrink a typical 2‑3‑week UX effort to a matter of minutes for proof‑of‑value.
• Integration with core banking, CRM, and lending platforms is the critical success factor for real ROI.
• TCS’s new AI data‑center and ongoing AI‑Day programs signal long‑term infrastructure and knowledge‑share investments for the financial sector.