Leveraging Facial Analytics in Credit Evaluation
Do you know that every year, the Insurance Sector in India loses about Rs.30,000 crores of Rupees in fraudulent claims!
According to a report published by India Forensic Research, the Insurance companies loses about 8.5% of its top line due to the frauds committed during the claim underwriting process. The Insurers have long been looking to make their underwriting process more robust and it seems that they might have found their answer, thanks to Facial Analytics.
Facial Analytics uses Artificial Intelligence to detect emotions and suspicious behavior. The human face is a complex multidimensional structure that may convey a great deal about an individual’s expressions, facial characteristics, and sentiments. Claimants can be interviewed and their micro-expressions, pupil dilation, eye movements, speech patterns, and tone of voice analyzed. The system would examine their responses and micro-expressions/reactions to a set of real-time queries using advanced deep learning models to determine whether the claim is legitimate or if more inquiry is required.
To construct an effective estimator, facial recognition technology involves two main components: facial feature extraction and estimator learning. Features first gets extracted from an input image and then compared with database images of various age groups with the set weight. A neural network model may be trained on age-based face photos with class labels as predetermined age ranges, and it can also predict whether or not the test face is of an offender.
In fact, combining Facial AI technology with wearable data can provide good insights for underwriting. Because both face analysers and wearables such as Fitbit can estimate body fat and BMI, they can supplement one another by offering supporting evidence, paving the way for a more accurate UW. This establishes the groundwork for ongoing underwriting. If a customer who also has high blood pressure exercises often (as recorded by a face analyser and wearables), she or he may not need to pay the extra premium, or their premium may be decreased based on the healthy lifestyle chosen.
Use Cases in Credit Evaluation Space
1. Face Detection & Feature Extraction — Facial Analytics can be used in detecting and locating all faces from the image or a video. Once the faces have been identified, facial features could be extracted based on face landmarks using CNN Model
2. Face Authentication — Post the extraction of facial features, a comparison with the database would help in identifying and authenticating the correct user, thereby restricting access.
3. Demography Detection — Estimation of the age and gender based on extracted facial features leveraging the use of Support Vector Machine classifiers. Features like BMI can be calculated basis the age and gender identified through the detection process.
4. Emotion Detection — Recognition of emotions of the interviewee through scanning and comparing with similar faces with multiple expressions.
5. Spoof Detection — The suite of solutions also includes liveness detection, which detects attempts to fool facial recognition software. Anti-spoofing by unsolicited participants can be detected through intra-face variations such as eye blinking, head rotation and movements and changes in facial expressions.
Let us look at how Facial Analytics can significantly simplify the Insurance Process.
User Stories
Globally, some of the fintech firms have already started to experiment the power of leveraging Facial Analytics. Ping An Puhui, China’s second-largest life insurer’s micro lending unit, has devised a digitalized loan process that can “analyse facial expressions of applicants to determine their willingness to repay the loans.” The company claims that by implementing new technology such as facial recognition and big data, it has seen its customer base “more than double to 5.5 million from 2 million a year ago,” and its loan default rate decrease, all without increasing its workforce.
Lemonade, America’s top-rated Insurer, stated that it gathers more than 1,600 “data points” about its customers, which is about 100 times more than the traditional insurance carriers. These data points are then analyzed by an artificial intelligence bot that then crafts and quotes insurance. The AI bot analyses the videos for fraud and supposedly can detect non-verbal cues for better underwriting.
With all these advancements, the day is not far when you might hear this from your insurer company — “Sorry, Mr. X, we can see from your face you’re not likely to repay a loan.”
