Insurance Payout Estimation
Vehicle insurance claims involve a manual process with expertise needed from domain specialists to evaluate the validity of the claims and their adjudication. Further, there are fraudulent claims that need to be identified, which results in inefficient usage of time for adjudication, and in some cases in fraudulent claims being approved. This manual process is not cost-effective, and fraudulent claims adds to the expense. It is also time consuming, hence there is a need to expedite the claims settlement process for such claims.
Our solution uses a blend of machine learning expertise and database solutions to create an application which involves a site visit where the agent can upload images of the damaged vehicles, with the claims details. These images are then fed through a machine learning pipeline which identifies the damaged parts of the vehicle, and estimates the cost of repair or replacement for the damaged parts.
Customer Success Snapshot
One of the world’s largest providers of claims management solutions.
- Claim settlement process had longer turn-around times as it involved a site visit and manual inspection of each body part by an insurance executive.
- Estimations for each claim were less accurate and had human errors which led to additional processing times to bring in the desired accuracy.
- Identifying fine-grained damages in a full car image was not possible and often missed by executives.
- For building an accurate model, training data such as images tagged with meaningful labels was unavailable.
- A database of parts and labor to give accurate and updated costs was to be built and integrated.
- Developed a computer vision deep learning model which involved artificial neural networks.
- A corpus of images were meticulously tagged with correct labels of damaged parts, building a set of high-quality training data for the deep learning model.
- The model can be trained using either transfer learning or computer vision services from cloud service providers such as IBM Watson, MS Azure or Clarifai.
- Once the damaged parts were identified the solution queried an updated database of parts to get the cost of replacement of the parts and labor charges.
- The solution scored 94% accuracy in identifying damages to headlights, windscreens, and full frontal damages.
- Reduced the time to settle claims since the solution estimated the damage in real time without any latency or manual intervention.
- The solution also helped in avoiding fraudulent claims being approved.
You May Be Interested In
Whitepaper: Facial Analysis: Face Recognition, Expression Recognition, and Gender Identification