HDFC Life Insurance Company Limited
Award Category
Finalist Innovation
Question 1: Give a high level overview of your digital insurance innovation.
HDFC Life is utilizing the capabilities of AI and ML across the customer journey by building tools in each sub vertical – vision AI, text AI, voice AI, core Machine learning and cognitive bots. Vision AI has been used to build FaceSense, AgeTymer and BodMeter to assist in the underwriting process and getting extra data points using just a customer’s selfie. PicReader was created to help in digitizing records and reducing manual errors which lead to a delayed claim settlement in the future. Sentilyzer made use of text AI to understand the sentiment in the customers written communication across mail, SMS as well as social media. To understand the customers intent and emotion in voice, an AI backed Emolyzer was built. This solution assists in call center training and real time responses. TrueCue, the in-house built voice authenticator, was also deployed as an alternate authorization tool for the customer. Core machine learning was extensively used to build propensity models for loans, recycling of payouts, cross-sell opportunities, fraud detection and revival propensities. HDFC Life extensively utilized the NLP capabilities by building multiple cognitive bots and chat engines which are capable of handling 300+ queries in real time.
Question 2: How have you demonstrated a strong customer focus through your project?
HDFC Life develops all the projects by abiding to our 5 values – collaboration, integrity, people engagement, excellence and mainly customer centricity. Keeping the customer at the center of our projects and trying to address customer’s woes by simplifying the service journey of the customer through AI based projects, we built various projects like AI powered chatbots to solve customer queries and reduce the response time. This has also allowed us to be present in the customers everyday apps such as whatsapp and twitter. For better identification and reduced fraud, we build FaceSense which is capable of factoring in aging related changes such as wrinkles, skin sagging and hair loss to authenticate better. To simplify the customer identification, TrueCue was built so as to shorten the authentication process. Solutions like AgeTymer enable the customer to get a look into their future appearance and make them realize the importance of insurance. To further simplify the customer buying journey, we have partnered with mobile payment banks and telecom companies leveraging their KYC documents to provide an instant policy purchase. For our existing customers, we have launched a pre approved insurance basis predictive need analysis by leveraging their existing information and documents.
Question 3: How did you use technology in your project – please expand on things like your use of analytics, AI, etc, as applicable.
HDFC Life utilized the latest developments in fields of AI/ML and cloud capabilities to build these solutions. FaceSense used Computer vision and Deep Learning Algorithms that involves Image processing based pre-processing, Video-Analytics based liveliness detection and Convolutional Neural Network based Face Recognition. Haar cascades are used to detect faces in an image and ResNet-50 model is used to extract features from the face. BodMeter and AgeTymer utilize the Active Shape Model and Conditional Adversarial Autoencoder to extract relevant features and provide a output. Natural language processing and deep learning was used while making the chatbots which are now capable of answering 300+ insurance related queries. Sentilyzer made use of standard NLP techniques like tokenization, PoS tagging, lemmatization, abusive word detection and spell checker to categorize text and give sentiment of the customer as output. TrueCue utilized ML to separate the voice from background noise and subsequent authentication against the saved audio clip. A Gaussian mixture model is used to store the voice sample which is later used for authentication. Emolyzer utilized a emotions tagged dataset to build a model which extracts 384 features from the audio. The final output helped in routing of calls to agents and better training programs.
Question 4: How have you measured the results and growth of this project (if available please provide data on traction, partnerships, cost-savings, revenue, etc). All information shared in this section is kept confidential.
The use of AI at HDFC Life has yielded some measurable benefits and innumerable intangible benefits. The most obvious and measurable benefit to the organization was by the fraud detection model which helped avoid early and fraudulent claims worth Rs.2100 million. The propensity model for recycling payouts has helped improve the lead/appointment rate by 100%. The revival propensity model has helped design the persistency campaigns by accurately predicting 70% of the customers who will pay the premium on time. The chatbots, ELLE and ETTY, are now handling 150,000+ queries a month. SPOK, the email bot, handles 28% of emails end to end and has handled 90,000+ queries in FY19 thus freeing up valuable bandwidth. TrueCue has achieved a 98% accuracy thus successfully replacing the lengthy process of authentication which involved the need of password details like Policy ID, DoB, OTP etc. This has shortened the average call duration and allowed the call center agent to handle more calls in the same duration. FaceSense has achieved a 92%+ accuracy in validating a person’s live image with his photo ID. This has saved the customer valuable time by reducing the time required for identification from 3 minutes to near real time.
Question 5: Please provide internal or external testimonials relating to the project if available.
HDFC Life held “TechEdge” day on 11thDecember’2019 to showcase the digital and technological initiatives undertaken by the organization to analysts. The initiatives were received well and can be seen in the comments below :- Citi Research Equity:- “As can be seen from some specific examples highlighted below, this is already leading to tangible benefits. We believe that HDFC Life has been an innovator with respect to products, technology, and distribution partnerships which clearly gives it an edge over its peers.” BNP Paribas:- “In the analyst meet, management showcased HDFCLIFE’s digital capabilities across the value chain – customer identification and acquisition, integration with traditional and new age partners, simplification of processes, data mining, cost efficiency, risk assessment and creation of an internal ecosystem. Management is focusing on creating long term moats through digital channels, expanding the life insurance market by getting into the mutual fund space, encouraging a culture which is pro-change and agile while lowering costs through digitization.” The stock prices saw a subsequent 2% rise after the event and a jist of the event is covered in the news article by LiveMint:- https://www.livemint.com/market/mark-to-market/hdfc-life-drums-up-digital-beat-but-investors-should-wait-for-proof-11576171288736.html Our senior management speaking about the digital initiatives undertaken and the impact they have:- https://www.youtube.com/watch?v=IWN6AmW0zts https://www.youtube.com/watch?v=WHKkIIsEqPg
Question 6: Is there any other relevant written information you would like to share as part of your entry?
HDFC Life’s CEO, Mrs. Vibha Padalkar, want the organization to be recognized as a technology first insurer/InsurTech in the next three years and is aggressively undertaking tech driven projects. We have started migrating our processes to the cloud for a quicker TAT and ease of running algorithms. Projects such as AI based hiring and automated Underwriting are being carried out to reduce the dependency on manual efforts and increase employee productivity. To simplify the customer journey, we are working on conversational bots for form filling. Speech to text in various vernacular languages is being explored for migrating to a voice based form filling method. We are also looking at providing pre-populated forms by utilizing facial or vocal identification. Medical tests have been a deterrent for many as it involves a travelling to a physical medical center for a prick. To solve this, we are exploring the field of non invasive medical tests using AI. To take this initiative a step further, we are trying to waive off medicals altogether for majority of our customers. To achieve this we are looking at utilizing data from alternate sources like fitness trackers, previous medical records and categorizing customers into pincode based cohorts.