Q1. Could you start by giving us a brief overview of your professional background, particularly focusing on your expertise in the industry?

I've worked in CX transformation, quality, risk, and digital operations in fintech, payments, banking, and BPO for nearly 20 years. With an emphasis on AI-driven quality assurance, fraud analytics, digitising onboarding, and enhancing operations, I have overseen significant transformation projects for both consumer and merchant enterprises. In order to reduce expenses, bolster risk controls, and enhance customer satisfaction, I have experience integrating AI techniques like speech analytics, LLM-based QA, and hiring automation. My area of expertise is Lean Digital Transformation, which improves compliance, governance, and agent capabilities while streamlining processes. My work has continuously produced scalable models for various sites, reduced fraud, improved SLA performance, and cost reductions.

Q2. With the Digital Personal Data Protection (DPDP) Act now in focus, how has the implementation of GenAI changed to ensure that PII (Personally Identifiable Information) is scrubbed before it reaches the LLM for analytics?

GenAI systems have adopted a privacy-by-design strategy after the Digital Personal Data Protection Act, 2023, went into force in India. PII redaction procedures are now added by businesses before data is sent to the LLM. This entails anonymising data using NLP models while maintaining context, utilising automatic tokenisation, and hiding specifics like names and account numbers.

More banks now run LLMs on their own servers or in private cloud environments instead of using public ones. They also improve prompt logging, access controls, and human review processes. The main change is that LLMs no longer see raw production data. Instead, only curated, masked, and approved datasets go through secure systems, which helps meet regulations and keeps analytics accurate.

Q3. How much capital can a bank/Fintech save in 'Risk-Weighted Assets' by having a superior, AI-driven fraud/quality detection system?

An advanced AI system for fraud and quality detection doesn’t directly lower RWA on its own, but it does improve asset quality and reduces the need for unexpected loss provisions. For banks following Basel rules, cutting fraud and early delinquencies by 10–20% can lead to real savings in the capital set aside for risk. In large retail portfolios, this can mean saving tens to hundreds of crores each year. More importantly, AI makes it easier to accurately estimate the Probability of Default (PD) and Loss Given Default (LGD), which helps optimize IRB models. The main benefit is that better detection reduces bad loans and improves how capital is used over time.

Q4. What is the 'Time-to-Maturity' for a GenAI CX project before it can be scaled across a global multi-shore operation?

For a GenAI CX project like auto-QA or AI-assisted call summaries, it usually takes 6 to 9 months to reach the point where it can be scaled globally. The first three months are for stabilizing data, tuning models, controlling hallucinations, and checking compliance. The next few months focus on integrating with workflows, managing change, and tracking key metrics like AHT, CSAT, and QA consistency. Full maturity comes after monitoring for model drift, adjusting for different languages, and aligning governance across regions. Scaling too soon without a good feedback system can hurt compliance and performance. Running controlled pilots with clear ROI goals is essential before rolling out worldwide.

Q5. When implementing advanced solutions like GenAI-driven auto-QA, how much of the rollout time is spent in 'Regulatory Sandboxing'?

In banking and payments, about 20 to 35% of rollout time goes to regulatory sandboxing, validation, and aligning risk governance. This covers model checks, bias testing, tracking data sources, setting up audit logs, and explainability reviews. The time needed depends on whether the solution affects customer communication, credit decisions, or fraud checks. High-impact uses like automated complaint handling or fraud detection need stricter sandboxing than internal tools like auto-summarization. Experienced organizations save time by involving compliance teams early in the design process instead of waiting until after development.

Q6. How has this 'Complexity Compression' impacted the Total Cost per Human Interaction, and what is the specific delta in agent training time required to stop burnout in this new AI-first environment?

AI-driven complexity compression usually cuts the total cost per human interaction by 12 to 25% by automating simple tasks like documentation, searching, summarizing, and QA sampling. But as AI handles routine work, the tasks left for agents get harder, which can increase their mental workload. Training now shifts from process-heavy onboarding to focus more on judgment and critical thinking. While AI can shorten training time by 15 to 20%, there needs to be more focus on resilience and decision-making skills. Without proper AI training, agents are at higher risk of burnout. The change is not just about time, but about redesigning training to build stronger problem-solving abilities.

Q7. If you were an investor looking at companies within the space, what critical question would you pose to their senior management?

If I were an investor, I would ask: “Is your AI actually creating lasting operating leverage, or just making things look better on the surface?” I’d look at whether AI adoption truly lowers costs in a sustainable way, strengthens compliance, and improves risk-adjusted returns, or if it only boosts dashboard numbers. I’d also check how mature the company’s governance is, how they manage model drift, their reliance on outside LLM providers, and whether they can show real gains in capital efficiency. What really sets companies apart is not just trying new things but scaling up with strong compliance and lasting margin growth.