Deloitte Luxembourg has launched a trained deep learning model that can accurately recognize car damage.
Our solution benefits insurers by transforming the claim value chain. It can accelerate time-consuming claim settlements and create a better customer experience, while also increasing efficiency. Fraud detection can be added to the model for additional checks and verification of the image uploaded with the claim declaration and the repair quotes collected. Furthermore, our AI engine evaluates different severities of damage across car parts and estimates the associated repair costs in seconds when connected to a repair cost database.
The solution benefits insurers by transforming the claim value chain. It can accelerate time-consuming claim settlements and enhance the customer experience, while also increasing efficiency. Fraud detection can be added to the model for additional checks and verification of images uploaded with the claim declaration and the repair quotes collected. Furthermore, our AI engine evaluates car damage and estimates the associated repair costs within a few seconds when connected to a repair cost database.
This could be of value for repair shops, car resellers or other intermediaries, where the model output could facilitate informed decision-making, accurate pricing and more efficient staff planning. As a flexible solution that is not vehicle brand specific, our solution is attractive for car manufacturers and other large automotive companies.
Our AI detection is set up to recognize car parts based on the standard vehicle inspection template and can therefore accelerate mandatory vehicle inspection. The model also detects damage on a more detailed level. Using close-up images, it is able to detect minor damage, such as dents, scratches and cracks. Furthermore, our damage detection model can add value to car rental agencies who can track damage before and after a lease by sending images of the vehicle through the model to compare output before and after the rental. This can lead to higher client satisfaction and improved fleet management. The damage detection algorithm increases overall damage assessment accuracy as it relies on constant evaluation by our algorithm, which eliminates human bias.
Our model is compatible and can integrate with your company’s existing applications. We have developed an app to display the functionalities and connectivity of the model. The app is fully cloud-hosted, yielding scalability and on-demand service, without the requirements of internal infrastructure or maintenance cost. The solution is easy to integrate into another ecosystem due to the app’s modular structure. As the app has an Amazon Cognito user authentication module, it can easily embed into a client relationship management (CRM) servicing module, linking detailed information to the assessment. Even without a CRM system, a connection to Amazon Cognito can be set up to gain the benefits of our solution.
We disclose all information on the training data at different stages together with generated statistics throughout the training phase. This information openness of both model inputs and outputs makes the model completely auditable. The model is not a black box solution. Furthermore, our solution does not use or process personal data. The standard vehicle inspection template forms the basis of the solution and it is aligned with the EU framework for claim declaration ensuring best practice incorporation.
As a cloud-hosted solution complimented by transfer learning, this solution works without massive computational resources and a large training dataset. It also creates the possibility to recognize other types of vehicles, such as trucks or motorbikes, and reduces the amount of images and labeling effort typically associated with image recognition. It is equipped with continuous learning which enables the model to be retrained as new pictures are processed. This allows the model to evolve with the business needs as new data becomes available. This model convergence allows flexibility to offer a solution fit for your specific needs, without additional time and investment costs.
Our AI damage recognition model will optimize processes across the insurance value chain. The declaration process, the damage analysis and the compensation process can be made more efficient through AI automation. The automotive player has the power over how much control they give to our model, by manually adjusting their level of confidence during the review to trigger human intervention. Color-annotated images and the official accident declaration form, in the case on an insurance claim, is also included in the output.
Additional components such as multi-factor fraud detection and premium-driven repair cost and time are available in combination with the current solution. Multi-factor fraud detection can save insurers the time and cost associated with investigating fraudulent claims, while the premium-driven repair cost and time solutions allow for flexible servicing to a range of customers depending on their preferences. The model can easily be linked to your cost database to include this functionality.
Opens in new window
These key differentiators will set you apart from your competitors if you integrate our image recognition solution into your daily value chain.
Opens in new window
Following an accident, a person can upload photos of the damaged vehicle on the app.
Our AI model detects the damaged car parts and the damage degree within seconds. The app can be linked to a repair database to report accident related information seamlessly to a list of qualified vendors.
From this information, a variety of garages or repair shops report estimated repair costs and repair time directly to the app.
The model is automatically retrained as it becomes exposed to new data.
Opens in new window
This enables both individuals and insurers quick and direct access to quotes. Insurers do not only benefit from garage selection, fraud detection, and shorter claim processing time but also from increased customer satisfaction by instant information and an accelerated process.