Alternative Data Gathering for Insurance Broker

Client

The client is an insurance services provider specializing on small enterprises.

Business Challenge

The client intended to expand their business by offering more optimal insurance premiums to the companies with no insurance history. The challenge was two-fold:

  • Collection of information from diverse sources where data is presented in very different ways;
  • Proper risk calculation in the absence of information about previous insurances.

Solution

Information about the companies was being collected from open sources, government databases and social networks. For aggregation of data from different sources, a set of ML algorithms was used. To classify the business type and risk level of the potential insureds DataArt built an ML model that was trained on similar types of companies with known risk tier. The model was built using Random Forest classifier algorithm.

Business Benefits

  • 80% accuracy for the risk tier classification (4 groups) and 66% for the business type (100+ categories) was achieved, so the actuarial process was also optimized.
  • The client won new customers and covered new market segment by offering premium insurance to the companies that found it difficult to get insured for reasonable prices.

Technologies

Python

SQL

Pandas

Scikit-learn

Seaborn

Linux

PostgreSQL

DigitalOcean

Anaconda

Jupiter Notebook

Welcome
We are glad you found us
Please explore our services and find out how we can support your business goals.
Get in Touch