Building a Serverless Centralized Mortgage Renewal Solution with Amazon SageMaker Canvas
Mortgage renewals can be a time-consuming and stressful process for customers, who must request renewals from each of their financial institutions separately. This can negatively impact a customer's credit score with multiple credit checks. To simplify mortgage renewals, I propose a centralized serverless solution using Amazon SageMaker to predict loan repayment risk.
Tiyam Robati
Amazon Employee
Published Jun 25, 2024
Today, renewing a mortgage requires customers to manually contact each of their financial institutions. This has several downsides:
- Time-Consuming: Reaching out to multiple institutions is tedious and slow.
- Hurts Credit: Each credit check can negatively impact a customer's credit score.
- Manual Processes: Relying on manual renewal processes is inefficient.
- Limited Data: Institutions may not leverage comprehensive customer data in assessments.
- Difficult to Scale: Growing loan volume makes manual processes increasingly unscalable.
By addressing these challenges, we have an opportunity to improve the accuracy and efficiency of loan repayment prediction, ultimately reducing financial risks and enhancing customer satisfaction.
To address these challenges, I propose a streamlined serverless architecture:
This solution uses various AWS services:
- Customer authentication
- Amazon Cognito user pool for centralized authentication, with a hosted UI sign-in page
- In App integration section of the user pool, we create Cognito domain that will be the signing page of application
- Amazon CloudFront as content delivery network and S3 to host a static website
- Creating distribution in CloudFront with the origins pointing to S3
- Leverage AWS API Gateway with RESTful APIs for communication between frontend and backend
- API Gateway Get method used to receive the customer http calls from the UI and invoke the lambda function from the backend.
- Customer Validation
- Store customer data in Amazon DynamoDB tables
- Partition data by CustomerID for efficient queries
- Cross-reference with customer profiles stored separately
- Prediction
- Build, train, and deploy ML models in SageMaker without coding using the Lending Club dataset
- Details is provided in Mortgage Prediction Model
- Create a real-time endpoint in SageMaker for low-latency predictions
- In deployment section, we select Endpoints:
- Then we select the model from deployable model list:
o Invoke predictions from Lambda functions for serverless integration
o Persist results to S3 and return to customers via API Gateway
The last piece of the solution is the legal process to officially opening the mortgage.
The key component is an Amazon SageMaker model to predict the risk of loan default. The model is trained on a dataset of past loan data to identify patterns that indicate default risk.
Benefits of this approach:
· More accurate by using large-scale historical data
· Reduces bias with data-based modeling
· Interpretable models provide transparency
· Scalable to growing loan volumes
With this solution, customers can submit renewal applications via a centralized website. The application data flows to the ML model to assess repayment risk. Low-risk renewals are automatically approved while high-risk loans trigger manual review. This improves efficiency, reduces credit impacts, and provides a smoother renewal process for borrowers. The centralized data also enables more holistic risk assessment.
This serverless, machine learning solution has the potential to greatly improve mortgage renewals. I'm excited by the possibilities and welcome your feedback!
Any opinions in this post are those of the individual author and may not reflect the opinions of AWS.