Roopam Asthana, CEO & Wholetime Director, Liberty General Insurance
Roopam Asthana, CEO & Wholetime Director, Liberty General Insurance

With the advent of AI use cases in the BFSI sector, insurance companies are getting bullish over leveraging this technology to meet the changing customer requirements and keeping the relevance of their business model. In this aspect, one such insurance company which came out from its traditional ecosystem and started to invest in technology is Liberty General Insurance.

The company has recently implemented AI to automate its car insurance processes. And here’s how they started this project by understanding its use case and customer’s perspective towards the final automation.

Evaluating AI

“We conducted a study to understand the market penetration, and how AI can make our business more efficient and customers happy. Post completing the research, we noticed that almost 25% to 35% of the old cars need an inspection before getting insured for reasons like a gap in insurance or cover enhancement. The traditional process of manual car inspections & quality checks before policy issuance was not only cumbersome & human-intensive but also wasn’t cost-effective,” Roopam Astana, CEO & Whole Time Director, Liberty General Insurance told ETCIO.

Given these challenges and the insurer’s aggressive business plans, Astana sensed an opportunity to disrupt the traditional process. He wanted to offer value to the company’s potential customers as well as bring efficiencies internally. After evaluating the study, he asked his tech team to work upon a few use cases that can fulfill customer’s changing demands.

With several brainstorming sessions and technology evaluation, Astana observed that AI ruled out every other technology in terms of benefits, cost-effectiveness, and implementation process. Thereby, he started working on the AI project a couple of years back.

According to him, while the AI use case was compelling, it also required commitment, skills, investments & consistent efforts to make it work.

“For this project, we did explore the vendor market but with a few considerations and factors to judge. We looked at IT vendors to get into a long-term partnership, and from the technology side, we were focused on key elements such as consistency, scalability, ease of implementation, and accuracy,”

Groundbreaking steps

The insurer’s research shows that even with strong mobile penetration in India, people are very conscious about using their mobile data and data storage to download a new mobile app. Hence as an unconventional step, we used SMS-based links to the customer’s phone to ensure a seamless experience in line with the existing customer behavior.

Secondly, the web UI has an easy-to-use photo-taking guide that provides step-by-step instructions on how to properly take photos of the car from all sides so that the customers do not get lost. As the AI-enabled inspection does not require manual intervention, customers can perform an inspection at any time they choose. Astana’s team observed that most customers prefer to do inspections on weekends.

The automation of the inspection process helped the company to serve more customers in less time. With the entire process digitisation, customers can get notifications of inspection in minutes instead of days.

“Our underwriting and technology teams have been working with IT partners over the last couple of years starting from conceptualising to implementation. Considering the number of car models in India, nature of damages & quality of photographs/videos, it was not an easy task. It was an intensive and iterative testing process to keep refining the end outcome for the end consumers,” Astana highlighted.

The AI Model

Under the AI-based inspection process, end customers receive a web link over SMS. Upon clicking the link, they are led to a mobile responsive web platform that prompts them to capture photos and video of a car for break-in policy renewals and upload them.

These photos or videos are sent to the cloud and an automated inspection report, covering damage and claim assessment, gets generated within a few seconds. The automated process replaces human intervention in repetitive work at a very high accuracy level.

According to Astana, It not only saves cost but also increases customer satisfaction by reducing the time required to renew a policy. This allows the company to respond to customers within minutes instead of days, and top of it the service remains 24/7 available.

Challenges are normal

Just like any other tech project, Liberty General Insurance faced a few challenges while working on this AI project. key challenges were generalisation of damage detection across car sub-models, lighting conditions, and orientation.

Astana emphasised that false positive, particularly the ones about dents vs reflection, and shatter vs reflection. A high level of accuracy has been achieved with research work with 350+ ML experiments, optimised network architecture, and a large image library that his team has collated over the last 2 years.

Another aspect of AI project challenges is to manage the bias factor.

AI bias is a continuous challenge, and we have been working on this. Car damage detection models certainly tend to overfit. We have to manage over-fitting with a 3-pronged approach that is data collection in which we remove biases in data across damage type, car model, lighting condition, and orientation; data labeling wherein data must get labeled consistently. This is particularly true for damage categories that do not have polygonic shapes e.g., scratches. Last but not least is the model selection in which neural network architecture can also be tweaked to manage biases better,” Astana explained.

Reduced turnaround time and quicker policy issuance are major benefits post implementing this AI project. The developed model learns from the history of inspection outcomes corresponding to a specific set of images and video. Because of this capability, the model can assist in fraud detection including detection of fresh as well as old damages. The AI module thinks logically without emotions, making rational decisions with fewer or no mistakes.

“We initiated this project with the intent of offering a faster turnaround & superior experience to our potential customers. In line with expectations, we are seeing a substantial reduction in the time taken to complete the entire activity from car inspection to subsequent policy issuance. Traditionally, an activity that used to take 4-6 hours on average (or more than a day in some cases) is now reduced to a few minutes. We are still working on further improvements to this solution to make it a near real-time experience for our customers & partners. Further, the accuracy of AI results is satisfactory, & is improving day by day,” Astana concluded.





Source link