We are implementing an AI model to assess claims, particularly focusing on fraud detection: Sunil Jain, CTO, Pramerica Life Insurance

In an insightful interaction with Express Computer, Sunil Jain, CTO, Pramerica Life Insurance, discusses the integration of AI and GenAI technologies into their operations. Moving beyond proof of concept, Pramerica is implementing AI models for claims assessment and fraud detection, resulting in significant reductions in claim settlement times, sometimes up to 50%. Jain shares their efforts in developing self-service platforms and leveraging data to enhance customer experience. As they explore future innovations, including voice and AI-driven chatbots, Jain emphasises the importance of user experience and the evolving landscape of insurance technology.

As AI and GenAI technologies continue to evolve, how are you incorporating them into your operations? Are there specific use cases where you see these technologies providing the most significant value? Could you share any insights or examples with us?

We’ve moved beyond the POC stage for AI and GenAI. For instance, in collaboration with our claims team, we’ve started implementing an AI model to assess claims, particularly focusing on fraud detection. Initially, there was some resistance to the AI-driven insights, but as we worked with the team, they’ve started accepting the results. The model highlights signals, like a recent bank account or nominee change, that can indicate potential fraud. Over time, we’ve expanded from identifying 10 to now around 18-19 key fraud indicators. The claims team is now more aligned with the AI, and it has led to faster claims processing, when the risk is flagged as low.

On the GenAI side, we’re feeding back claims data to improve model accuracy. We’re also working on a lead propensity model and using Vision AI in underwriting to replace traditional OCR, which is more expensive. We’ve started pilots here, and while the results will take time, we’re optimistic about the benefits.

Would you say that for these reasons the claim settlement time has got significantly reduced?

Yes, the claim settlement time has significantly reduced. While AI and ML play a key role, other factors contribute as well. For example, we’ve integrated multiple channels like SMS and email for claim intimation, eliminating the need for customers to visit a branch. Once a claim is initiated, the workflow is automated. We also use QR scanning to verify documents like death certificates, ensuring authenticity. Additionally, we’ve integrated Abha for health record verification and are collaborating with IAB to check for any previous claims, making the process more efficient and transparent. In many cases, the reduction in settlement time has reached up to 50%.

However, some challenges remain with adoption. There are individuals who are sceptical, believing their judgement is better than AI models. It’s more of an adoption issue than a model problem, as some people easily accept the new system, while others are more resistant, struggling to trust that a system can be more intelligent than them.

Pramerica has enhanced customer access to policy management and claims tracking through self-service platforms. What challenges did you face during the implementation of these platforms, and how did you overcome them?

There were two key areas of focus for us when implementing the self-service platforms: one for customers and the other for agents. While agents prefer a closer connection to customers, high-net-worth individuals (HNIs) often prefer a direct self-service approach. We automated 40 out of 64 services, covering about 95% of the requests we receive. 

The main challenge was handling legacy systems, especially since insurance companies still rely on mainframes, which aren’t built for real-time data processing. We had to extract data, create a warehouse, and ensure compliance with processes like bank account changes, which require manual verification. Compliance and fraud concerns also slowed down certain updates, such as changes to phone numbers or addresses. While technology moves fast, we had to ensure we adhered to regulations.

What can be the use case potential of GenAI in the insurance industry?

There are two parts to consider: AI and GenAI. AI is already playing a significant role in underwriting, which helps reduce risk. Underwriting data is much larger than claims data, with claims being less than 1%. AI engines process this vast amount of data quickly, identifying patterns and correlations we might not foresee, allowing policies to be adjusted more accurately.

Another area with great potential is medical underwriting. We are exploring partnerships with AI giants like Google to see if medical underwriting can be enhanced. For example, AI has shown promising results in predicting diseases like breast cancer years before onset, which could transform the industry.

However, it’s not just about us; even re-insurers need to accept these models. Currently, we aren’t fully automating the process but are using AI to flag potential risks. Additionally, we’re working on propensity models, combining data from bureaus, Vision AI, and financial records to assess whether a customer can afford a policy and what their persistence rate might be. The system continually evolves as it processes more data, providing better insights.

With hybrid cloud becoming a popular approach, how do you see its impact on cloud providers, particularly in terms of pricing, and what factors should organisations consider when choosing between private and hybrid cloud, especially when evaluating total cost of ownership?

Hybrid cloud is definitely the way forward, and it will put pressure on cloud providers to offer better rates. However, it depends on the economy and the size of the organisation. For large banks, having an engineering team to manage a private cloud can be more cost-effective and scalable. Some big banks are already using private clouds, which give them greater control and savings at scale.

On the other hand, smaller companies, with a footprint of around ₹10 crores a year, need to carefully evaluate the total cost of ownership. Saving a crore here or there might not be worth the additional engineering costs. Hybrid cloud is a good option if you have the budget to invest in an engineering team for the long term, but for startups or growing companies, cost management is crucial. As seen with Ola’s switch from Google Maps to their own Ola Maps, companies often hit a ceiling on costs and must reevaluate their cloud strategy based on total cost of ownership.

Given recent allegations of data hacks and breaches at certain insurance companies, what security and compliance measures does your organisation have in place to ensure the integrity of digital transactions and protect customer data throughout the entire insurance process?

It’s a complex situation. While it’s not as difficult as for banks, who deal with highly sensitive data that can move quickly between accounts, organisations still face strict regulations. Banks, for example, have to comply with very stringent RBI guidelines. Similarly, we face strict audits from UIDAI, such as GRCP audits, cybersecurity audits, and VPAT compliance. Security has become non-negotiable. 

We work closely with providers like Airtel, handling everything from VPAT to cybersecurity audits and internal reviews. As a US-based company, we also have to meet additional security standards. It’s a continuous process. A cybersecurity expert once said, “If even Microsoft gets bugs, how can you blame me?” That’s the reality we live in. 

UIDAI has introduced stronger controls, and compliance is mandatory. Last year, there were around 80 controls, and now there are 126. These controls go beyond technical aspects; they even cover HR policies, ensuring we hire the right people and manage fraud risks effectively, especially when dealing with sensitive data. It’s an evolving and challenging landscape. Let’s see how it continues to unfold.

How are you leveraging AI for threat detection and filtering out false positives, especially given the large number of incidents reported daily? Additionally, with the growing sophistication of attacks, how do you see AI playing a role in staying ahead of these evolving threats?

There are two aspects to this. First, we are using AI for threat detection in collaboration with partners. I receive between 500 to 800 incidents daily, sometimes even up to a couple thousand. It’s impossible to treat all of them with the same priority manually. We’re working with Google, where they do the first level of scanning and filter down to the top three or four incidents for us to focus on. Their AI models assess the incoming traffic and determine whether the threats are genuine or just noise. This is still in the early stages—we started using it on October 1st—and we’ve seen a 25% reduction in incidents, but it’s too early to say whether all real threats are being caught.

On the second point, as attacks become more sophisticated, AI is crucial in staying ahead. AI is now being used to write malicious code that’s far more advanced than before. It’s like a battle of wits between two AIs, constantly learning and countering each other. It’s a war out there.

Pramerica is building a data-driven Direct-to-Customer (D2C) ecosystem to enhance cross-selling and outreach. Can you share the strategy behind this ecosystem and the role data plays in shaping it?

The D2C model is certainly beneficial as it helps save costs, but it’s not easy to navigate due to the complexity of the products, especially in a market like India where insurance penetration, apart from term insurance, is only around 4%. To succeed, we need to be agile. We collect and analyse leads, focusing on existing policyholders’ data, like payment history, lifestyle, and credit scores. We’re building a Customer 360 model, incorporating data from various sources like Vahan and social media. This helps in better targeting for cross-selling and upselling. While we’re optimistic, it’s still early, and we need to refine our models further. Ultimately, the success of this approach depends on how much data we can afford to acquire and use efficiently to close sales.

Are there any emerging technologies or trends in the insurance industry that you plan to capitalise on?

Yes, I believe the biggest trends in the insurance industry focus on enhancing the customer journey and performance engineering. With account aggregators, IEPs, and bureaus in place, the question is whether we can inform individuals about their insurability at the outset. Ideally, we want to minimise the number of fields they need to fill out while automating the process as much as possible.

There is a continuous emphasis on user experience, which will always evolve. As one company improves, others will follow suit with better solutions.

What will be your primary focus for the next 12 months, including any specific initiatives, new technologies, or offerings you plan to introduce?

My primary focus for the next 12 months will be the launch of a voice bot and an AI-based WhatsApp bot. These bots will go beyond traditional question-and-answer formats. For instance, rather than providing single-thread responses like most current bots, my aim is to create a more conversational experience. 

For example, a user might ask, “I’m Sunil Jain. How much do I need to pay? What is my total sum assured? When is my next payment due?” This approach allows for a more general and interactive conversation.

We conducted extensive benchmarking worldwide and have finalised a vendor who will help develop this advanced bot. Unlike typical bots that are focused solely on transactions, this will enable users to ask comprehensive questions, such as, “What is my account balance? When will my next payment hit? Can I take an overdraft of 5 lakh rupees?” This functionality will allow users to interact with the system in a more intuitive way.

Will it be a text based or voice based bot?

The bot will be both text-based and voice-based. It’s conversational, allowing interactions through either medium. However, due to the high costs associated with voice technology on OpenAI, I plan to start with a proof of concept (POC) before investing further, as voice tokens are quite expensive. 

Initially, I think we should focus on English, as it covers most regional languages. My POC indicated that English was manageable, but I’m not entirely certain. I aim to incorporate all languages, but the actual field test will provide clearer insights.