InsuranceData Engineering

Predicting insurance policy claims

An insurance company in the US is facing a 5% of claims and is struggling to understand their distribution and factors. Lets develop a predictive model.

Industry: Insurance
Solution: Data Engineering

Book a meeting

01 / 03

The challenge

An insurance company in the US is looking to:

Detect and predict patterns in policy claims.
Reduce the % of claims in its business portfolio.
Improve profits by avoiding insuring high-risk cars.

Problems faced: the currently percentage of claims is 5%.

Difficulty in understanding the distribution of claims and their factors.
Complexity in identifying probability patterns in claims.
Knowledge based mostly on unverified hypotheses.

02 / 03

What Tekne did

Objective

Therefore, it was necessary to achieve greater use of the data from their systems, and to do so they aimed to:

Organize data to draw conclusions, detect patterns and predict behaviours.
Verify existing hypotheses about claims.
Generate and contrast new hypotheses and insights.
Develop a predictive model to estimate the probability of claims over the next 6 months.

Proposed Solution

We work together with the company proposing:

Implementation of predictive models to predict policy claims: random forest and logistic regression using 36 months of data.
Deployment of the models via API and integrated into its origination platform.

Work continuity:

Advice on good data management practices.
Implementation of tools for policy management.
Analysis to obtain valuable insights.

Architecture and Processes

Implemented architecture

Model diagram

03 / 03

Results

After working with us, the company now has:

A standard application model for claim detection.
Insights on which aspects influence the making of claims.
Predictive scoring for analyzing the behavior of customer policies and establishing different prices according to behavior.
Predictive scoring for carrying out campaigns in sector where the probability of claims is lower.

More cases