Mr. Chirag-Bhojani Head-Data-Science-&-Analytics

ICICI-Lombard General Insurance Co Ltd

1:-What is the current state of AI adoption in India? Is there a surge on insurers leveraging AI to streamline processes?

New research commissioned by IBM found that about 59% of enterprise-scale organizations (over 1,000 employees) surveyed in India use AI actively in their businesses. The ‘IBM Global AI Adoption Index 2023’ found early adopters are leading the way, with 74% of Indian enterprises already working with AI, having accelerated their investments in AI in the past 24 – 30 months in areas such as R&D, repeat-task automation, workforce reskilling & recruitment, automation of customer self-service answers and actions(Report). Rapid advances in AI accessibility, cost benefits accrued from automation of key processes and the flexibility of embedding AI into standard off the shelf business applications are leading to this rapid adoption. In a recent news article, Mark Zuckerberg claimed that India has become the largest market for Meta AI usage.

Coming to insurance, the insurance industry in India continues to be predominantly a high touch industry where customers buy insurance through dealers, agents, brokers, bancassurance sales representatives and hardly 5% of customers buy insurance through digital channels. Having said that, the interactions driven through these physical channels are digital, and this digital adoption rapidly accelerated during and post the pandemic. At the same time, the Indian consumer has become increasingly digital as is evident from the increase in digital payments driven by UPI, the increased usage of Swiggy and Blinkit. This has led to the increased generation of digital data, and this digital data has helped in the increased usage of AI models since the backbone of AI is large amounts of data.

Over the last several years, the insurance industry has started seeing a lot more digitized data, which will help build and strengthen AI driven models. We are seeing insurance organizations building AI driven models around claims adjudication for better claims experience, risk-based pricing and segmentation models, models that help in the identification of fraudulent customers / channel partners and AI driven BOTs for better customer experience. With the advent of generative AI, insurance organizations are also looking at newer methods of serving customers as well as in increasing productivity of employees across marketing, technology, customer service and other functions and processes.

2:-Speaking about ELEVATE, what specific aspects of the health insurance plan are powered by AI? 

As an industry, we have seen significant AI-driven innovation happening in the fields of claims adjudication, fraud identification, risk-based segmentation and customer service. AI-led product personalization was an area that we had been experimenting with but there had been no large-scale implementation. The Elevate product was constructed keeping in mind the current insurance product features in the industry, feedback from consumers on current insurance products offered by the Company and feedback from intermediaries who were selling health insurance. Basis feedback collected, a modular product that would serve different customer segments and their needs was to be developed. To complement this modularity, product personalization at a customer level was extremely critical.

This led to the design of a first-of-its-kind health product that was loaded with cutting-edge features & add-ons and offered personalised solutions to meet the needs of dynamic lifestyles, unforeseen medical emergencies and medical inflation. The Elevate product is powered at the backend by a collaborative-filtering based AI engine that interprets customer’s behavioural, financial and risk level data to help deliver an optimal coverage recommendation based on the customer's persona. Each persona’s perceived requirements have been kept in mind and a personalized coverage and recommendation plan is subsequently created for each customer. A study of what coverages similar personas buy over time is then fed back to the AI engine to sharpen the recommendation algorithm.

We are seeing customers buy marquee features such as infinite sum insured, infinite claim amount, power booster and reset benefits basis the recommendations that have been suggested by the AI engine. We continue to remain excited about the growth in health insurance business driven by the Elevate product and look forward to providing extensive protection while catering to personal health and financial considerations, making it more adaptable and responsive to customers’ needs.

3:-What are the main challenges in launching an AI-powered health insurance plan? 

To launch an AI-powered personalized health insurance plan, a significant amount of data related to consumer behavior, financial and risk level data is required. Digitization over the years has helped us collect this kind of data. At the same time, since we have moved all our applications and data on the Cloud, this has helped us store massive datasets and has also helped us in agile development and deployment of compute-intensive AI models.

The biggest challenge while creating personalized health insurance plans is to ensure that there is no unintended and unpredictable bias in profiling. Responsible AI is an upcoming area that helps in conducting rigorous testing to proactively test such drift or bias. Our data governance framework establishes principles for ensuring fairness while developing and deploying such models. At the same time, human oversight has been established to ensure that algorithms are performing as expected and decisions related to underwriting and policy rejections are done by a health underwriter and not the AI algorithm.

4:-In your opinion, does the role of AI in insurance businesses streamline or disrupt processes? 

AI clearly helps in minimizing repetitive tasks and in duplication. It typically helps in the reduction of manual workload and makes the overall process more efficient. In terms of the insurance industry, organizations like ours are seeing significant benefits.

For example, on the customer service side, our Responsive Intelligent Assistant (RIA) helps address customer queries on the website, on our ILTakeCare app and on WhatsApp. RIA is powered by a powerful NLP engine and has recently been augmented by generative AI to help understand customer intent and resolve customer queries. Customers are now able to download policy copies, intimate claims, understand the status of their claims, understand the policy coverages and exclusions and purchase and renew policies. At the same time, the call center has been equipped with a voice BOT that helps answer routine queries such as claim status and policy information.

These AI enabled features has helped streamline the customer service process while bringing superior experience for the ever-changing digital customer. As an organization, we have seen an increased productivity of ~30% for customer support teams driven by these digital and AI initiatives.

Another example is our health cashless authorization initiative. The health insurance cashless claims process used to be a manual process where each case would have been seen by a doctor who would decide on the admissibility and adjudication of such claims. Leveraging the power of AI and ML, we have been able to service more than 58% of our group health cashless claims within 90 seconds. An AI model reads the authorization request and determines admissibility of the claim while an ML model predicts the authorization amount driven by the history of similar claims.

 5:-What measures are being taken to mitigate the impact of AI on employment?

Increased digitization and data are helping us understand consumers a lot more than before. Computing power of the Cloud is helping us to deploy large scale AI models quickly and efficiently. AI is helping in automating repetitive tasks and in the creation of automated responses, images, videos and in deciphering complex problems faced by us.

To build such models, organizations are having to upskill, and in many cases, reskill workers to be able to feed data to such models and in helping create such models. Core skills that are required to perform a job today are different from 10 years ago and will not be the same 10 years hence. Continuous learning and skilling are required for the skillsets of tomorrow. As per a recent OECD report, we are seeing AI help in higher productivity, improved job quality and stronger occupational safety and health. All in all, organizations are investing time and money in upskilling of workers to ensure that they have the skillsets required to harness the power of AI and ML in years to come.

At the same time, there is widespread discussion around how decisions are being made by these AI models and the transparency and explainability of such AI models. Since explainable AI is still nascent and has some time to become mainstream, many organizations continue to follow a hybrid approach to the implementation of AI solutions. Such organizations leverage manual oversight to ensure that fairness is maintained while taking decisions using AI models. At the same time, AI models are helping skilled workers focus more on complex problems and let AI take care of the simple repetitive problems.

6:-Generally, how transparent is the AI decision-making process to policyholders?

Data is helping insurance organizations design new products and services for customers. Organizations will continue to experiment using data and it will not be long before we see an AI underwriter that assesses risk, an AI driven claims adjuster that helps in straight through processing of claims or an AI driven BOT that helps answer queries in real-time. We have already seen organizations, including ours, creating similar solutions even today. While AI is helping make sense of the volume and the variety of big data, explainable or interpretable AI is still a relatively new field but will help explain the rationale behind the complicated AI and ML models that are built. It is expected that complex neural networks that are behind a lot of these AI models will be simplified by explainable AI as it will help in identifying and interpreting the predictions made by such models. We all know that transparency helps build trust and ensures fairness.

While we continue to focus on building AI models that help in superior customer service, we are following a manual oversight approach to the implementation of AI models. As an example, in our break-in AI model where we are requesting customers to capture photographs of their vehicle when they do not renew their motor policies in time, cases that are rejected by the AI model are not rejected directly but are sent to a human to ensure that there is no bias or unintended errors. Cases that are approved by the AI model go for straight-through processing and are not seen by a human adjuster.

7:-Which emerging technologies are expected to drive the next wave of AI innovation in India?

AI and ML will continue to drive innovation across the world. Multimodal AI will be driving this next wave of innovation. The incoming generation of interdisciplinary models, comprising proprietary models like OpenAI’s GPT-o can move freely between natural language processing (NLP) and computer vision tasks and can now bring video into the fold. "The most immediate benefit of multimodal AI is more intuitive, versatile AI applications and virtual assistants." (“The most important AI trends in 2024 - IBM Blog”) Users can, for example, ask about an image and receive a natural language answer, or ask aloud for instructions to repair something and receive visual aids alongside step-by-step text instructions.

The next-in-line are small language models (SLMs). Massive models jumpstarted the ongoing AI golden age, but they are not without drawbacks. "Only the very largest companies have the funds and server space to train and maintain energy-hungry models with hundreds of billions of parameters." (“The most important AI trends in 2024 - IBM Blog”) SLM’s, on the other hand, can be run at lower cost on more attainable hardware, run locally on smaller devices and make AI more explainable due to its fewer parameters. SLM’s should also help organizations pursue differentiation through bespoke model development.

"As AI systems speed up and incorporate new streams and formats of information, they expand the possibilities for not just communication and instruction following, but also task automation." (“The most important AI trends in 2024 - IBM Blog”) While we saw a lot of use cases around chat in the last year, these coming few months should start seeing an amalgamation of this chat with automation using AI.

While we continue to see the above changes in general, on the insurance front, IoT driven telematics that helps in driving behavior-based pricing in motor, sensor data combined with gene data to help in more personalized pricing in health, drone based inspections and underwriting in commercial businesses combined with remote sensing and weather data in crop insurance could become the next generation AI driven models in insurance.

 8:-In your opinion, should there be an ethics board or committee overseeing AI implementations?

Many insurance organizations have been leveraging AI/ML over the years in various areas such as claims adjudication, fraud detection and customer service. All this has been possible due to the increased digitization and collection of data over the years. Collection of this data has potentially come at the price of data privacy. Regulations such as GDPR and the Digital Personal Data Protection Act in India are looking to bring some order to data and its privacy. At the same time, increased data usage has led to innovative models that have helped streamline and create superior customer experiences. Balancing innovation and privacy are going to be critical in the long run.

Large insurance organizations like ours are creating frameworks that help oversee data consumption and the responsible use of AI models at scale. At ICICI Lombard, we have a robust data governance framework that emphasises the responsible usage of data and ensures fairness in models and algorithms across the organization. The EU AI act is already on its way, and it will not be long before we see other countries adopting similar legislations for AI implementation.

 Note – The interview was  published in Asia Insurance Review in Sept 2024