The Re•Work AI in Insurance Summit in New York City was held September 5-6 and saw 60 speakers from AVIVA, Travelers, GoCompare, Prudential and other insurance-related companies cover a wide range of topics — from detecting claims fraud to applying machine learning to underwriting and maximizing revenue.
Today’s specialty and commercial insurance underwriters face an overwhelming number of challenges. AXIS Capital Senior Data Scientist Min Yu believes artificial intelligence (AI) will transform the specialty and commercial insurance underwriting from a “detect and repair” mode to “predict and prevent” mode. In her talk on Machine Learning to Specialty Insurance Underwriting, Yu outlined the AI process as follows: receive a submission, retrieve data, analyze risk, automate quote and quick binding. Manual underwriting would be mainly used for review, or on complicated or emerging risks.
Yu stressed one of the most pressing challenges facing commercial insurers is a lack of in-house data for risk evaluation. There are a lot of things AI could do to solve this problem. For example, 80 percent of internal data is unstructured, in the format of PDF and emails. Text mining and natural language processing (NLP) could reveal the core hidden information; and using AI to scrape information from the Internet could help companies extract real-time related information to better understand evolving risks.
The annual cost of fraud is US$1 trillion across all industries; and banks worldwide have paid US$372 billion in noncompliance penalties since 2009. Traditional systems for combating financial crimes such as fraud and money laundering are usually rule-based static systems with limited capabilities for identifying sophisticated patterns.
Manulife Advanced Analytics Director Amir Sepasi recommended leveraging graph databases and link analysis to detect and prevent financial crimes by uncovering hidden patterns in data using real-time graph algorithms. In his speech Uncover Hidden Patterns in Financial Crime Activities Using Graph Analytics, Sepasi shared anti-fraud and anti-money laundering (AML) use cases that enabled greater customer connections than traditional fuzzy matching.
The cases illustrated how to leverage graph databases to link customers based on names, date of birth, etc.; how to create a weighted relationship indicating users’ similarities based on IP usage; and how community detection algorithms for AML can help companies identify potential money laundering activities based on transaction patterns.
Identifying the latest condition of a particular disease type is also a critical challenge in areas such as risk adjustment, which require more supportive evidence than claims data. Anthem Data Science & AI Director Lakshmi Manohar Akella shared his company’s work on attention-based deep learning approaches for classifying health status charts in the context of a current condition.
Akella’s team built an electronic medical record (EMR) chart classification system with deep learning using hierarchical attention networks. The EMR charts are composed of text reports with an inherent hierarchical nature. Hierarchical attention networks (HAN) outperform many conventional methods such as SVMs or traditional linear methods with BoW (Bag-of-Words) features etc. Akella said hierarchical attention-based deep learning models can be effectively used to determine a current condition from patients’ broader electronic health records (EHR), and can also be applied in risk adjustment, planning interventions, and personalization.
Featured presentations from the Re•Work AI in Insurance Summit NYC 2019 are available here.
Author: Yuqing Li | Editor: Michael Sarazen