Case Study

Model-Assisted Claims Assessment

Challenge

For property insurance adjusters, assessing roof damage can be very time-consuming and dangerous. This applies to both commercial and residential roofs , and covers all types of roof damage, including hail, high winds, leaks, etc. In cases where direct access to a damaged roof is not possible or is extremely unsafe, high-resolution drone photos can be used to assess the damage. A system was needed to better capture and classify roof damage in order to reduce loss rates and lower the cost associated with claims processing.

Industry

Financial Services / Insurance

Data Type

Image (jpg, png)

Project Duration

3 Months

Ongoing?

No

Solution

To increase adjusters’ efficiency, we developed a model that would automatically review images and find defects. Building such a model requires large amounts of time and specialized talent to assess and annotate thousands of images.

Because their adjusters weren’t able to provide the time needed to train the model, our solution was to use model-assisted labeling, which combines the efficiency of an automated process with a review by an insurance professional.

This human-in-the-loop (HitL) step leverages the efficiency of an automated process to reduce the time demand on the adjusters, and the model is being built and trained while claims continue to be processed.

In addition, this process ensures that all outputs benefit from the adjuster’s expertise and judgment. Corrections and improvements, which are made based on the insurance professional’s feedback, are entered into the automated process to continually enhance the accuracy of identifying the defects.

Outcome

Automatically identifying defects in the rooftop images, resulted in reduced loss rates, while the lowered time demands on the adjusters reduced claims processing costs.

Download Case Study

Download