Industry: Insurance

Using AI-powered solutions to automate cost estimation for insurance companies

The client’s mission is to create a damage cost estimation tool. The tool collects data about an insured event, such as a car accident, and uses algorithms and machine learning to estimate the cost of the damage. It is designed to be user-friendly and efficient, allowing the insurance company to quickly and accurately estimate the cost of the damage and process claims more efficiently.

3 Weeks in scoping            26 Weeks in development

THE PROBLEM

How can we automate the cost prediction process effectively and accurately?

In the past, assessing damages and estimating the cost of repairs for insurance claims involved manual inspections of numerous low-resolution photos captured by mobile cameras, phone calls, and onsite physical inspections conducted by assigned assessors. This process was time-consuming even during normal working days when car accidents were a common occurrence on busy roads. During the holiday season, when road traffic and car accidents peak, the car damage inspection process for auto-insurance claim assessment could take a long time. As a result, customers had to make repeated calls to inquire about the progress status, with no certainty, and often being instructed to call again. The challenge was to build an AI-powered model that not only assessed the damage but also predicted the estimated damage cost. 

THE PROCESS

Brainstorming and crafting the perfect roadmap 

To deploy this solution, the company had to ensure that the AI model was trained effectively to accurately detect defects. Building the AI model for car damage assessment posed several challenges for the insurance company. Firstly, creating an appropriate dataset for training the AI model was a challenge in itself, as there were few publicly available datasets of broken or damaged cars, if any. Consequently, the company had to create its own comprehensive dataset, including various types of cars, body parts within the cars, and different types of damage, including varying degrees of severity. 

The process is carried out in a systematic way. After some mishap, such as a car crash, the claimant captures the pictures of the affected vehicle and uploads them on the mobile application. The pictures make their way to the cloud and it leads us to computer vision. Computer vision recognizes, interprets, and understands visual data uploaded by the claimant. This can include images, videos, and other types of visual information. The tool enables the development of algorithms and is trained to extract relevant information from visual data. These algorithms are designed to mimic the way the human brain processes visual information, using techniques like image processing, pattern recognition, machine learning, and particularly instance segmentation.

Instance segmentation algorithms typically use a combination of machine learning and deep learning techniques, including convolutional neural networks, to detect and segment objects within an image. By identifying each object separately, it provides more detailed information about the objects present in an image, enabling more advanced and accurate prediction of the cost that would cover the damage. By applying all these steps to the visual data provided by the claimant, our AI tool with the help of subject matter experts from the company forecasts the cost for the company.

THE SOLUTION 

An AI-powered tool that would assist the insurance company in predicting the estimated cost, effectively and processing the claims more efficiently.

It was an end-to-end solution that yielded a simple yet powerful software that automates the prediction process. It used data sets from different open sources and customers’ provided data files to predict the estimated damage price. AI model performance monitoring and retraining are well underway using MLOps techniques. We are incorporating insights and feedback from domain experts such as insurance adjusters and accident investigators that would result in enhancing the accuracy and relevance of this AI model.