- Hartford
- Insurance
- 20 weeks
- B2C
- Python, R, Apache Spark, Tableau, Databricks, AWS
Project Brief
A popular insurance firm wanted to use data analytics, to prevent false claims, increase underwriting accuracy, and optimise pricing models. The objective was to create a comprehensive data-driven strategy for risk assessment and detecting fraud.
Client Background
The company, which is a large supplier of casualty and property insurance, had substantial issues due to false claims and erroneous risk assessments. Traditional methods of identifying fraud depended mainly on manual inspections and rule-based systems, both of which were time-consuming and error-prone.
Key Challenges:
Solution:
1. Discovery and Planning
We worked with the insurance company's data science and fraud detection teams to obtain a thorough grasp of their current procedures and difficulties. We carried out a complete data audit, selected key aspects for investigation, and created an organised data analytics strategy.
2. Development
Our skilled data analytics team used Python's Pandas and Scikit-learn tools to create predictive models. We used Apache Spark on Databricks for distributed computing and scalability, which allowed us to analyse large datasets effectively. We created algorithms for anomaly detection to find unexpected trends in claim data that may suggest fraudulent behaviour. The models were installed on AWS, including services such as S3 for storage of data and Redshift for data warehouses.
3. Implementation
We incorporated the data analytics platform into their current claims processing system. This included creating data pipelines to import claims data, training models on historical data, and deploying the models in a production environment. We also gave fraud investigators extensive training on how to utilise the platform to detect and probe suspected fraud cases.
Tools & Technology Used
Python
Programming Language
Tableau
Data Visualization
AWS
Cloud Services
Apache Spark
Data Processing
Features:
Fraud Detection
Fraud detection involves identifying possibly fraudulent claims through abnormalities in claim facts, policyholder behaviour, and external data sources.
Risk Assessment
Risk assessment predicts future claims for individual policyholders, leading to correct pricing and underwriting choices.
Real-Time Alerts
Real-Time notifications: Automated notifications for questionable claims activity allow for early investigation and response.
Explainable AI
Explainable AI offers straightforward explanations for model predictions, increasing confidence and openness in decision-making.
Customizable Dashboards
Customisable dashboards enable users to analyse claims data, track model performance, and spot patterns.