Loading...

(Car) Insurance

...

The Problem

Car insurance companies follow a specific process for handling small accidents.One the most fundamental part of these processes which is the accident inspection is time consuming and costly due to the requirement of experts.

By using Machine Learning these processes can be altered to be more efficient and accurate.

The classic process of claim handling follows:

The call center must be informed for the accident.

Surveyor inspects the vehicle

Client provides the necessary documents.

Inspection and approval.

Payment.

The above process as it requires inspectors due to their limited availability results to higher costs and bigger waiting times. By applying machine learning for predicting the damage dealt, the involvement of surveyors is not mandatory.

Solution

The claims handling process with the use of machine learning:

A model assesses the damage inflicted. No need for expert.

Client provides the necessary documents.

Automatic document inspection and approval.

Payment.

Process Followed

Using historical data a model is trained for predicting the cost of the damage..

Using images from car crashes a model is trained to estimate the damage severity.

Combining the two models we get an accurate damage estimation.

A separate model is created for classifying if a case is total loss or not.

By retraining the two models we get more accurate results.

Programming languages: Python
Packages: Pytorch, OpenCv, Pandas, Numpy, Scikit-learn
Other: AWS

All the above steps consist of an automated process covering from data collection to model deployment.

Real time cost evaluation becomes feasible either via an application or with a call to the call center. The client by uploading 2-3 images and answering 2-3 questions possess an accurate cost estimation.There is no need for waiting for an expert to come to the accident’s site . This results in happier clients with a far more optimized claims handling process for the insurance company.

Benefits

Provide smart decision support tools to non experts FNOL handlers (First Notification of Loss) that will help them predict cost and the outcome of a
claim from a minimum amount of information.

Automatic triage cases between repair and total loss without any human intervention.