Digital case study

Situation:

Can the insurance claims estimate generation process be optimized and scaled to support exponential growth?

The insurance industry has experienced its share of technological disruption over the past ten years. As widespread smartphone adoption made it easier for consumers to submit images of collisions in support of auto insurance claims, insurance carriers were deluged, further bogging down an already frustrating and time consuming process. Where others saw chaos, our client saw an opportunity. They hired a team of experienced estimators and auto body experts and started to evaluate photos of auto damage for carriers as a service.

Drawing on their collective experience, the team would look at the photos and provide insurance companies with a preliminary assessment of the damage to the vehicle. Our client was filling a void in the industry, and their business exploded. They now service over 70 insurance companies and are a significant player in the insurtech space. Their challenge was that there are only so many experienced estimators and auto body experts available for hire, and at the rate the company was growing, they knew their growth would soon be constrained by limited resources.

Our client knew that using artificial intelligence (“AI”) and image recognition technology was a possible solution. It could address their scaling challenge and help them continue to improve customer experience. They built a data science team as a start. But then quickly found that building trust in the AI-assisted system, both for their estimators and customers, was key to making this approach successful.

Solution:

Analyze the estimation process, leverage insertion points for PwC’s Analytics and AI solutions and focus on building trust with the estimators through “explainable AI”

The process

The client is laser-focused on enhancing customer experience. PwC’s AI and Analytics team realized that improving the efficiency, quality, accuracy and consistency of the estimation process would have a major positive impact on their insurance company clients and policyholders through a decrease in claim resolution time. The initial project scope was to work with the estimation team to look at the estimation process involving images and determine whether an AI model could detect which parts of individual vehicles had been damaged based on submitted photos. Automating this previously manual process was an important first step in our collaboration.

PwC’s Analytics & AI Transformation Solution

Working together with the client’s data science team and estimators, we helped create three AI models. The first was to detect and classify car damages from the images, isolating where in the image there was damage and what type of damage was represented. The second model translated damage into individual parts affected by adding additional information to the analysis, such as parts lists for damaged subassemblies, etc. The third model retrieved images of similar vehicles, both damaged and undamaged, to help estimators evaluate whether or not the part was actually damaged. The estimators were pleasantly surprised with the accuracy and depth of the results. The AI model was even catching details that estimators had missed through the manual process. So far, so good.

Building trust in the AI model

Now for the hard part. The ability of an AI model to accurately interpret images was only half of the equation. The AI model needed to be trusted by the estimators as an important augmentation of their analysis. To address this, we partnered with our client’s data scientists to implement an “explainable AI” (XAI) technique. We then used it in a novel way, showing visually why the model arrived at a particular prediction by making analytical determinations at every stage in the process. By doing this, we greatly increased trust in the model as well and demonstrated its value to the data scientists, the estimators, the insurance companies and the insurance company’s clients

Results:

PwC partnered with our client to use explainable AI to improve customer experience and establish trust in the system for its estimators.

The company’s explainable AI model was a game changer as It enabled the following:

• empowered auto claim estimators to identify where to focus attention during an assessment
• provided approaches for sharing knowledge among the estimator team to accurately determine which group should handle specific estimates
• identified 29% efficiency savings possible with full implementation of proof of concept models across the estimator team
• reduced rework and improved customer experience through reduced cycle times

Through this unique approach to explainable AI, the company has offered a powerful example to the industry of the potential power of AI-assisted processes. They now have concrete validation that an AI-assisted approach works and has the potential to improve other aspects of their overall process. The company has a clear model for growth and a foundation for scalability.