What factors can predict early payoffs in South Korea? How loan amount, age and reluctance to reveal marital status are key considerations when managing a loan’s life cycle.

The Challenge

• Too many early payoffs (EPOs) can disrupt the credit cycle and hinder a company’s ability to forecast long-term cashflow.
• In recent years, customer churn has increased due to a competitive and lowering interest rate market.

The Purpose

• How do we predict EPO and customer churn? Can we identify customers that will leave, before they leave?
• What are key factors correlated with EPO? What insights do they reveal?
• How do we create strategies and product offerings to reduce churn and extend credit cycle?

The Method

• A combination of internal data, Credit Bureau data and open-source data allowed us to analyse 97,000 accounts and 32,000 EPO events over an 18-month period. 
• Internal data profiled customers by age, income, loan details and marital status.
• External data analysed credit history as well as environmental and employment factors.
• In 4 weeks, a series of models were prototyped and fine-tuned to calculate probabilities of EPO.

The Findings

• Optional data fields– Customers willing to provide additional, non-compulsory information like ‘marital status’ were 3 times less likely to pay out loans early than those who left these fields blank.
• Youth– Those below 30 are more likely to EPO in the first 4 months than older customers. The young are generally more price sensitive and responsive to competitor marketing.
• Interest– Loans with a high interest rate are 3 times more likely to pre-pay than those with a low interest rate. Unsurprisingly, the higher the cost of debt, the stronger the incentive to settle early. 
• Loan size– Smaller loans have higher churn, as the market is more competitive at lower price points.
• Home movers– Those who move home are more likely to switch banks.

The Outcome

• Utilising our data science model, the success could translate to an estimate of $1.2 million improvement in annual net interest margin for the business.
• Open-source data, such as information on traffic flow and air quality, improved the model’s ability to find leavers and avoid non-leavers, raising the potential incremental loan value by $10.5 million.
• Our model continues to be refined and upgraded to deliver greater accuracy and targeting precision when used with compelling offers to retain customers.