This use case article gives an overview on the data-driven and automated solution that we built for the insurance SparkassenVersicherung to reduce manual processing time and improve the speed of the verification process.
- Data Science
- Computer Vision
- Deep Learning
- Classification
- Image Analytics
- Insurance
Key Challenges
Our client is a modern regional insurer with a comprehensive range of products and services. They receive many applications in which applicants must attach several documents including a valid copy of their identity card (or ID card). In the initial process, an administrative team is in charge of checking each file manually to verify if the ID card is included and if it is valid or not.
Therefore, our client was looking to:
- Reduce manual processing time
- Improve the speed of the process
- Be supported from the ideation to the implementation in the running process
Our approach
To meet our client’s expectations, we took the time to understand the current manual process of application management and ID card verification in order to propose the best data-driven and automated solution to implement.
Together with the administrative team we divided the process into two key steps. First, screening the presence of the ID card in the application. Second, if so, approving or rejecting the validity of the ID card provided (considering that applicants can send many different format of their ID card).
To develop this computer vision project, we assisted our client in the “make vs. buy decision” for the screening solution. We used publicly available models and performed additional training to solve this binary classification problem (via pre-trained open-source neuronal networks recognizing the ID card pages).
We took care of operationalizing and deploying the models in production. Once an ID card is missing or is not validated by the algorithms, automated emails are sent to the administrative team for human intervention.
Benefits
- Correct recognition rate of ID documents in insurance applications and reduction of manual efforts by more than 85%.
- Successful integration into the productive process of the application system (dark policing).
- Time and cost savings in the development of the screening solution.
- Fulfillment of the requirements of the Money Laundering Act for random checks.
Technologies & Partners
Team involved
Data Scientist, Deep Learning Engineer and ML Engineer collaborated with the insurance expert for 12 months to build this data-driven and automated solution.
Discover our Data & Analytics services
Thanks to our
Authors
