India’s insurance market continues to grow across motor, health and term insurance, but the country’s financial progress is yet to translate into adequate financial security for households, according to Santosh Bhat, Chief Data Scientist and Head of Advanced Technology at Policybazaar.
Speaking to Brij Pahwa, Editorial Lead, BW Businessworld and exchange4media, and Editor, MartechAI.com, at the Global Fintech Fest in Mumbai, Bhat said the data showed that insurance activity had remained broadly on expected lines.
“If I’m looking at the data and broadly what we see in our numbers and so on, the numbers have been consistently going up,” he said. “Whether it’s motor insurance or health insurance or term insurance, it’s not necessarily slowed down.”
However, Bhat said rising insurance activity should not automatically be interpreted as evidence that Indian households are financially secure. The larger challenge, he said, was financial literacy.
“I think there is still a long way for India in terms of financial literacy,” Bhat said. “People have misconceptions about how financially secure they are or what they need to become secure and how they need to secure their family.”
According to him, Indians often do not have a clear understanding of the level of protection they need, the risks they face or the role insurance should play in securing their families.
The data quality problem
Bhat also pointed to the quality of data as one of the biggest reasons why artificial intelligence systems sometimes make inaccurate predictions about consumers.
A model can only be as reliable as the information on which it is trained. If the data is incomplete, outdated or incorrectly interpreted, the model can produce biased or irrelevant outcomes.
“It is predominantly lagging behind because of the quality of data that is being fed to the models,” he said. “If you have the wrong set of data and you train the model on that, then obviously the model is going to create a bias.”
Bhat said organisations need to examine data for accuracy and authenticity before using it to train AI systems. Data enrichment and efforts to remove bias are equally important, he added.
His comments are relevant beyond insurance. Across sectors, businesses are using AI to predict consumer behaviour, personalise communication and recommend products. However, inaccurate targeting can undermine customer trust, particularly when automated systems make assumptions that do not reflect a person’s actual financial circumstances.
AI handles 60 per cent of chats
At Policybazaar, AI is being used extensively in customer service. Bhat said the company’s AI systems now handle approximately 60 per cent of customer interactions across voice and text chats.
The aim, however, is not to replace human advisers entirely. Instead, AI is being used to handle routine queries and remain available to customers at all times.
“The AI agent’s role is to understand and solve whatever possible, in whatever possible ways, he could solve a customer’s problem,” Bhat said. “The role of an AI agent is to be available 24-7 whenever the customer needs him.”
He described the AI agent as a “digital twin” of a human adviser. It can respond to simple queries, such as requests for a soft copy of a document or basic questions about a policy and its benefits.
More complex issues are transferred to human advisers. Bhat said payment-related problems, refunds, cancellations and other complicated concerns were deliberately escalated to people.
The company’s data, he said, showed that the increased use of AI had not negatively affected customer satisfaction.
“The chats, whether it’s through voice or text chats, it’s more than 50 per cent, it’s possibly about 60 per cent now,” he said. “That number has slightly gone up compared to last year, but the satisfaction pretty much has remained the same.”
This suggests that customers may be willing to engage with AI when the problem is simple and the system resolves it quickly. However, human intervention remains important when the issue involves financial stress, uncertainty or a more complicated personal situation.
The challenge of explainable AI
Bhat has previously written about the black-box problem in AI, referring to the difficulty of understanding how a model arrived at a particular answer or recommendation.
He said explainability would be essential for building trust as AI becomes more deeply embedded in financial services and insurance.
“Any large language model is predominantly a black box,” Bhat said. “You’ve given certain inputs and certain prompts and so on, it’s forced to answer, but it won’t tell you how it got to that reasoning.”
The issue becomes particularly significant in insurance, where automated systems may influence underwriting, customer recommendations, fraud detection or claims-related processes. A customer may reasonably want to know why a system has classified them in a certain way or recommended a particular product.
Bhat said the industry would have to work towards making these systems more understandable and transparent.
AI, automation and the distance from AGI
The discussion also touched on whether current AI systems represent genuine intelligence or are primarily sophisticated automation tools.
Bhat said today’s AI systems were capable of performing tasks that conventional software could not handle effectively, including reading documents and extracting information from calls. However, he did not believe India or the world was close to achieving artificial general intelligence.
“If that question refers to artificial general intelligence, then I think we are some way away,” he said. “We are not anywhere close. Whether it is three years away or five years away, we’ll wait and watch. It is very difficult to answer that question.”
He added that large language models were not equally effective across all tasks. While they could process text and conversations efficiently, they were not necessarily as capable when dealing with numbers and interpreting complex numerical relationships.
Faster motor insurance issuance
One of the practical areas where AI has improved Policybazaar’s customer experience is motor insurance.
In cases where a motor policy has expired and the customer wants to renew it, insurers may require a video inspection of the vehicle. Earlier, the process could take two or three days.
Bhat said AI could now assess the video and examine several parameters almost immediately. In suitable cases, a policy could potentially be issued within five minutes.
The same approach is being used in pay-as-you-drive insurance, where customers may need to record specific parts of their vehicle, including the odometer.
“It’s a significant uplift for the customer,” Bhat said.
The company is also working on smaller, specialised language models designed for specific use cases. Bhat said fine-tuned models were already being used for certain tasks, while a broader model for Policybazaar’s business and insurance-related questions was still under development.
India’s foundational AI gap
Despite India’s large population, extensive data and linguistic diversity, Bhat believes the country has fallen behind in the foundational part of the AI race.
“We’ve lagged in the foundational part of the AI race,” he said. “We don’t have any foundational model which has been built in India, nor are we controlling the compute.”
He noted that India did not yet control either the hardware or all the software required to compete at the highest level of AI development. However, he argued that the race was still at an early stage and that India could catch up.
India’s advantage, according to Bhat, lies in its scale and diversity. The country has a large population, multiple languages and a vast number of real-world problems that require practical solutions.
“We are sitting on just a volcano of data,” he said.
For Bhat, India’s AI opportunity may not depend only on creating the world’s largest model. It could also emerge from building systems that understand Indian consumers, languages and business contexts, while solving specific problems across sectors such as insurance, healthcare, banking and public services.