Smart Risk Assessment: Bitech’s AI-Driven Solution for Property Insurance

Bitech faced challenges in assessing and managing risk associated with particular geolocation. The goal is to determine the risk of a particular location based on multiple factors such as crime, unemployment, flood, climate, and fire to prevent loss in the property insurance segment. In addition to this, they faced issues with extracting  key information from clients data , summarize their data, reducing the time and effort needed to process and understand complex data.

In the solution for the insurance industry, AWS Bedrock's Titan model was leveraged to enhance the user experience by summarizing risk scores for specific locations. This solution is particularly useful for property insurance companies looking to assess risks associated with geolocation factors like crime, unemployment, flood, climate, and fire.

Architecture:

Data Management

The Raw Nexiga data and EEA data are stored in an S3 bucket.

There are two separate folders for storing 

1. Nexiga data :- The Nexiga datasets contain different social economical features and geolocation features such as postal code and city name. The datasets contain multiple crime columns such as home breaks, motor theft, damage to property, and total crime. 
2. EEA data:- This folder contains all the climate datasets that we selected from the EEA  agency for climate analysis.

All Jupyter Notebooks files with the extension .ipynb are present in the ec2-user folder.

The Nexiga datasets contain socio-economical features and EEA contains climate-related features.The two datasets contain city columns as a primary key. Created master datasets by combining these datasets based on city columns.

Data Imputation 

The combined datasets contain missing values for the climate and flood factors. The missing value is imputed based on the nearest city from the climate data(EEA) with minimum distance (haversine).

Analytical Hierarchy Process

AHP is a structured group decision-making technique for organizing and analyzing complex decisions. It represents a semi-objective approach to quantifying the weights and preferences of decision criteria. Using a specially designed format, each team member uses the forced choice paired comparison process to rate the relative importance of each pair of items.

Once the appropriate criteria or elements of the decision have been selected, the group members systematically evaluate the various elements by comparing them to each other two at a time. In making the paired comparisons, the group can use actual data about the elements, but they typically use their subjective judgments about the elements’ relative importance.

Application Flow:

The preprocessed data has been stored in s3 bucket to perform summarization  User actions trigger requests to the server, which communicates with data. The server fetches data from the s3 bucket and further with the LLM model endpoint summarizes the risk score of the specific location.

Here user can enter specific details such as city name and postal code, the app processes this input to fetch risk data. LLM analyses this data and generates a summary of individual risk scores and overall risk when the user clicks on summarize button.

Deployment:

We have used  AWS Codepipeline to deploy the chatbot into the AWS  Elastic Container Service. The source code is pushed to Codecommit based on the commits to the protected branches; the code pipeline triggers the code build job which is responsible for the deployment of new tasks to the service.

• Improved User Experience: The system, through the integration of Titan, reduces the raw data complexity and presents easily understandable risk summaries to the user. It saves time and effort that was required for underwriters and risk analysts, who can now focus on decision-making rather than investing time in digesting huge amounts of complex information.
• Automation and Efficiency: The integration of Titan into the dataset reduces manual work. Since it is held within AWS S3 buckets, automated processing and summarization of data for real-time insights without human time being taken up to a great extent can be done.This helped increasing productivity of the employees .
• The creation of a chatbot  for real-time risk summarization in property insurance holds great commercial potential. Insurance providers may make well-informed judgments, reduce losses and increase customer satisfaction by providing support by facilitating rapid risk assessment.