Loan Propensity Prediction Model for a Retail Bank

The Client

The client is a retail bank offering a wide range of credit products to its customers. Their marketing effectiveness depends on reaching the right customers with the right offer at the right time. As competition increased, they needed a data- driven approach to identify customers most likely to engage with loan products.
 

The Challenge

Marketing campaigns in retail banking often rely on broad targeting strategies, resulting in low conversion rates and inefficient use of resources. The bank needed a more precise way to identify customers with a high likelihood of applying for loan products.

The challenge involved modeling customer behavior over time, capturing patterns in past loan applications, and predicting near-term intent. This required handling time- series data and identifying subtle behavioral signals that precede a loan application. Additionally, different loan products exhibit distinct customer behavior patterns, necessitating separate modeling approaches while maintaining a unified system. Without predictive targeting, the bank risked missed revenue opportunities, inefficient campaign spending, and suboptimal customer engagement.
 

What We Did

Approach

We developed a machine learning-based propensity modeling framework to predict the likelihood of customers applying for loan products within a one-month horizon. The approach focused on capturing behavioral trends through time-series feature engineering.

During the discovery phase, we identified that different loan products required distinct modeling strategies due to varying customer behavior patterns. This led to the development of separate models per product, ensuring higher predictive accuracy. A key decision was to build an end-to-end automated pipeline, enabling continuous data processing, model training, and prediction generation for operational use.
 

Methodology

• Data assessment:
  Analysed raw customer data, including historical loan applications and behavioural patterns
• Approach selection:
  Evaluated multiple machine learning algorithms and selected the best- performing models through systematic comparison
• Model development:
  Engineered time-series features to capture customer behaviour trends and trained separate models for each loan product
• Validation framework:
  Used historical data and grid search optimization to evaluate model performance across different configurations
• Integration:
  Built an automated pipeline covering data preprocessing, feature engineering, model training, and monthly prediction generation
• Compliance:
  Ensured adherence to data privacy and internal governance standards

The Outcome

Accurate prediction of customer loan application propensity

The solution enabled the bank to assign individual propensity scores to each customer for each loan product, supporting precise and actionable targeting.
 

Secondary outcomes:

• Enhanced targeting of marketing campaigns
• Identification of key behavioural drivers behind loan applications
• Improved conversion rates through personalized outreach
• Scalable pipeline enabling continuous model updates and predictions

Facing a similar challenge?

If you want to improve campaign performance and target the right customers with precision, we can help build predictive models that turn customer data into measurable business impact.

Book a Technical Consultation