Preventing Customer Churn for a Multinational Bank
A leading multinational bank implemented C3.ai Cash Management to use cash balance and behavior to predict customer churn and detect customer rate sensitivity.
Client cash balances are highly volatile, making it difficult to distinguish between balance behavior indicative of true customer churn and cash fluctuations from normal business operating activity. As a result of the daily balance volatility, the bank would sometimes take unnecessary preventive action in the form of better interest rate offers or lower fees for a satisfied customer experiencing temporary cash fluctuations. At the same time, the bank was missing subtle signals of impending churn by clients in need of better rates or different products and services.
From the bank’s perspective, offering higher interest rates had been an effective but also expensive lever to address specific client issues, respond to macro-level market factors, and retain balances. However, rate sensitivity varies widely by client. As a result, the bank sometimes made sweeping interest rate changes to clients in a specific industry or segment due to the behavior of a few select clients or macro-level factors. This resulted in decreased profitability due to unnecessary rate increases for clients looking for a better product, different service, or minor change to fee structure.
About the Multinational Bank
- $110 billion Annual Revenue
- 46 million customers
- Over 200,000 employees
- $2.3 trillion Assets Under Management
- $400 billion in deposits
- 200,000 corporate customers
- 16 weeks from project start to production-ready application
- 3 years of historical data from 7 data sources
- Provide up to 90 days early warning for 50% of customer churn
- 4,500+ time-based expressions constructed for machine learning
- Workflow-enabled user interface to facilitate client reviews and targeted action
Preventing Customer Churn with AI
C3.ai and the bank collaborated to deploy C3 .ai Cash Management on the C3 AI Suite™, integrating three years of historical data (deposit balances, transactions, product usage, revenue, credit information, and external market data) into a unified, federated data image. The software applies scalable, production-ready machine learning algorithms to produce human-interpretable predictions.
To predict customer churn, C3.ai first worked with bank experts to develop an algorithm that retrospectively identified over 20,000 instances of permanent balance attrition over a 31-month historical period. In parallel, the team built over 4,500 time-based features to model the complexity of clients and their activity.
Using the identified instances of permanent balance attrition as positive labels of customer churn and the time-based features, the team developed an interpretable machine learning algorithm that produces two key outputs: (1) a risk score that indicates the likelihood of an impending customer churn, and (2) warning signals derived from the top features contributing to the customer churn risk score.
A workflow-enabled user interface provides cash management and sales teams with interpretable AI predictions enabling them to review and prioritize action.
Correctly predicted balance attrition with up to 90-days of warning
Annual economic benefit from saved balances and targeted rate change offers