A leading financial service company leverages AI and ML models to automate their loan application process|PreludeSys

About the client

The client is a leading community-based financial services company based in San Francisco that offers a wide range of financial products and services to individuals and businesses. Their offerings include banking, investments, mortgages, and consumer and commercial loans.


Client challenges
  • The client faced inefficiencies in manually verifying loan application eligibility, which consumed valuable time and resources. This tedious process often led to delays in moving applications forward in the approval pipeline.
  • Slow approval workflows were a common hurdle, which caused significant delays in the overall decision-making process for loan applications. This not only impacted customer satisfaction but also slowed down the efficiency of the lending operations.
  • Monthly reporting tasks involved extensive manual data preparation, resulting in a time-consuming process for the client. The lack of automation in this area added to the staff’s workload and increased the chances of errors in reporting.
  • Without a robust data-cleansing framework, the client struggled to maintain the quality and accuracy of their loan-related data. Decisions were not necessarily based on reliable and up-to-date information.
  • The client lacked a system for predicting loan eligibility, making it challenging to assess applications proactively. This absence hindered the efficiency of the loan processing system, leading to a less streamlined, sluggish operation.
Solution
  • In-depth Exploratory Data Analysis (EDA): We conducted an exploratory data analysis to create an eligibility prediction model, delving deep into the client’s loan application data. This step was crucial for understanding patterns, trends, and variables affecting loan approvals.
  • Balanced Dataset with SMOTE: We balanced the dataset using the Synthetic Minority Oversampling Technique (SMOTE), addressing the imbalance in loan approval categories. This ensured that the predictive model would not be biased towards the majority class.
  • Algorithm Fitment Analysis:  An extensive analysis was conducted to fit the best-suited algorithms to the client’s loan application dataset. This involved assessing algorithms such as Gradient Boosting, Logistic Regression, Random Forest, and XGB Classifier to determine the most effective model.
  • Hyperparameter Tuning: Hyperparameter tuning was undertaken to optimize the performance of the chosen models. This process fine-tuned the algorithms’ parameters to enhance the eligibility prediction model’s accuracy and predictive power.
  • Model Evaluation Metrics: We evaluated the models using key metrics such as Accuracy and Area Under the ROC Curve (AUC). These metrics provided insights into the eligibility prediction model’s performance and effectiveness.
  • Interactive Visualizations: Interactive visualizations provide insights into loan approvals and rejections. These visualizations included views of reasons for refusal, eligibility scores, and operational perspectives, which empowered the client with actionable insights from the data.
Results at a glance

 

Increased efficiency as ML models reduces manual efforts Faster decision-making enabled the automation of workflows AI and ML models offer accuracy and predictability
Benefits
  • Reduced manual data preparation tasks, simplified processes, and improved accuracy.
  • Operational costs were optimized through automation and efficiency gains.
  • Business users could easily access and analyze reports using Power BI.
  • The ML model, developed in 3-5 months, enabled precise loan eligibility predictions, enhancing proactive decision-making.
Technology

Python, SQL DB,  and Power BI