Sourabh Chatterjee joined Kotak Life Insurance in December 2025. Coming from a nimble startup environment at Oona Insurance, he immediately began shaping the company’s Artificial Intelligence strategy, which focuses on building minimum viable products (MVPs) for AI use cases rather than proof of concepts (POCs).
“We undertake functioning minimum viable products (MVPs), not proof of concepts (POCs), at Kotak Life. AI projects are evaluated from two perspectives,” said Chatterjee. “The way forward is to work on AI use cases that are vertical (‘Go-Deeper’) and horizontal (‘Go-Wider’).”
To build a robust AI foundation, Kotak Life is strengthening its data warehouse infrastructure while improving data quality. The company is also experimenting with multiple AI models within the necessary governance guardrails.
Edited Excerpts:
How was your experience working in Singapore at Oona Insurance ?
Oona was an ‘unlearn and relearn’ experience, where I had to start from scratch, go back to basics, inherit the technology stack which came in with the acquired insurance entity and then build multi-country, multi-tenant, multi-currency, multi-language scalable platforms for multiple markets in Southeast Asia.
As far as work culture was concerned, the onus was on me to understand and adapt to the working cultures across Southeast Asian markets and it was again an exciting learning experience.
How much do you plan to reduce the expense ratio using technology ?
Technology spending (both Capex and Opex) is seen as an expense today in all organizations. Our endeavour is to change this measurement paradigm by positioning technology as a value creator and delivering outcomes which drive revenues, operational efficiencies, enhance CX (customer experience), distributor experience (DX) and IX (internal functions/departments experience). In the age of AI, technology teams can be at the front and center of value creation and then the discussion shifts from reducing expenses to driving value.
Have your digital servicing platforms measurably improved the 13th and 25th month persistency. Steps undertaken to improve persistency ratio ?
Persistency is a very important metric for the life insurance business. The 13th month persistency for Kotak Life is in the higher 80’s (in percentage) and the 61st month persistency is mid 60’s (in percentage).
From a technology standpoint, we have established systems that directly contribute to persistency improvement. In the payment ecosystem, nearly 90% of our renewal collections happen through digital channels - close to 70% is through Electronic National Automated Clearing House (eNACH) and 20% through online payment gateways.
Additionally, we have embedded eNACH within ‘Boost’ - our distributor app and other customer platforms. This has enabled Electronic Clearing Service (ECS) activation for nearly 35,000 policies, enabling customers to digitally set up an automatic bank-debit mandate directly within the app, without requiring physical paperwork or separate banking steps.
What kind of POCs are you undertaking in the space of Artificial Intelligence (AI) ?
AI Beyond POCs
We undertake functioning MVPs not POCs at Kotak Life. Also, AI projects are evaluated from two perspectives:
First is ‘go-wider’ projects for awareness, adoption and productivity gains within different functions, and the other is, ‘go-deeper’ projects for measurable impact on topline or bottom-line across the value chain.
Horizontally, we have successfully piloted around 20–25 AI use cases, across various departments like Operations, Marketing, Finance, Technology and others. For instance, the Marketing team is using AI for content creation, while Product and Actuarial teams are using it for competition product research and insights, compliance teams for summarising regulatory notices and generating insights, and finance teams for reconciliation activities. All of this is being done within organisational boundaries to ensure we are compliant with the applicable Information Security guidelines.
Even at an individual level, AI is helping improve productivity. For example, I personally use Copilot to summarise meeting minutes, draft notes, and write emails and also an AI Agent to organize my mailbox, aggregating important emails based on my own rules, sentiment analysis of received emails, and sending a concise weekly summary every Monday morning at 9 am.
Vertically, we go deeper into specific business processes and have shortlisted three to four strategic use cases where AI can drive a meaningful impact. These include using AI for front line sales productivity improvement, voice for Customer servicing and assisting underwriters and actuaries to better mitigating risk and fraud.
We are currently experimenting in these areas and our models are in a learning phase. While the horizontal productivity use cases are already proving effective, the deeper vertical implementations are evolving rapidly.
Success from AI depends heavily on data quality and quantity. What’s your data strategy for AI ?
Building the Data Foundation for AI
Ultimately, the success and return on investment from AI depend heavily on the quality, volume, tagging, and annotation of the data used to train and fine-tune the models. Whether we use commercially available LLMs or develop proprietary models of those, the outcomes will only be as strong as the data that feeds and trains and retrains those models.
Beyond building a central data warehouse, our focus is now on improving data quality through proper data tagging and metadata management. Different business functions generate different types of data — underwriting data, claims data, financial data, and so on. A significant effort is underway to organise and structure this data so that it can serve as a reliable foundation layer for AI models.
As a life insurance company that has been serving over five crore customers for more than 25 years, we possess a vast amount of historical data. However, much of this data resides across different systems and silos. Bringing it together, structuring it effectively, annotating it, and ensuring its relevance for AI models is a complex organisational exercise.
However, beyond the technology layer, we are also focused on identifying cost-efficient AI models that can be effectively integrated into our organisational workflows. At present, our AI deployments operate in an assisted mode, with a human in the loop, which is important because AI models can produce inaccuracies or hallucinations. Given that we operate in a highly regulated life insurance environment, strong governance, controls, oversight and a HITL (human in the loop) framework is essential.
I believe another big challenge needed for the success of AI projects is organizational change management. Moving from manual processes to automation, assisted AI, and eventually fully autonomous decision-making requires operational controls, people management skills and maturity. That is the journey we are currently navigating.
