How to Prepare for a Decision Engine Implementation

Introduction

In previous articles we examined the topics of best practice decision engine implementation and implementation project considerations. In this article we will review what preparation a lender should undertake prior to commencing a decision engine implementation project.

Implementing a credit decision engine involves navigating through various commercial, credit risk, and technical considerations, in order to ensure its successful deployment and operation. By completing this ‘home work’ before and during the implementation project, credit grantors will ensure the likelihood of a successful project:

Successful Preparation Steps

Implementing a credit decision engine involves multiple detailed business and technical steps to ensure it functions effectively, integrates well with existing systems, complies with regulations, aligns with organisational goals and provides reliable decisions.

This is a comprehensive guide to successfully implementing a credit decision engine:

Business Considerations

Define Objectives and Scope

• Objectives: Clarify the specific goals of the credit decision engine (e.g., automate credit scoring, streamline loan approval processes, optimise credit risk assessment).
• Scope: Determine which types of credit decisions the engine will handle (e.g., consumer loans, mortgages, credit cards) and the level of automation required.

Gather Requirements

• Business Requirements: Collect detailed requirements from stakeholders (e.g., underwriters, risk managers, compliance officers) regarding decision criteria, decision rules, integration needs, and reporting requirements.
• Regulatory Requirements: Ensure compliance with relevant regulations (e.g., Fair Credit Reporting Act, GDPR) and industry standards.

Data Collection and Preparation

• Data Sources: Identify and gather relevant data sources, including applicant information (e.g., demographics, income), credit history (e.g., credit bureau scores, repayment behaviour), and other financial data.
• Data Quality: Cleanse, pre-process, and integrate data to ensure accuracy, completeness, and consistency. Address any missing or outlier data.

Model Selection and Development

• Model Selection: Choose appropriate modelling techniques based on the business objectives and available data. Common models include logistic regression, decision trees, ensemble methods (e.g., Random Forests, Gradient Boosting Machines).

• Feature Engineering: Identify and create relevant features that influence credit decisions (e.g., application score, credit bureau score, debt-to-income ratio).
• Model Development: Train and validate models using historical data. Split data into training, validation, and test sets to evaluate model performance.

Decision Rules and Policies

• Decision Rules: Translate model outputs into actionable decisions (e.g., approve, reject, refer for manual review) based on pre-defined decision rules and thresholds.

• Policies: Incorporate business rules (e.g., maximum allowable risk level, product-specific policies) and regulatory guidelines into the decision-making process.

Technical Considerations

Integration with Existing Systems

• System Integration: Integrate the credit decision engine with existing IT infrastructure, including loan origination systems, CRM systems, and data warehouses.
• API Design: Design APIs or interfaces for seamless data exchange and decision integration with other systems.

Testing and Validation

• Unit Testing: Test individual components of the credit decision engine (e.g., data pre-processing, model training).
• Integration Testing: Validate end-to-end functionality and integration with external systems.

• Performance Testing: Assess the decision engine’s performance under various loads and scenarios to ensure scalability and responsiveness.

• User Acceptance Testing (UAT): Involve stakeholders in testing the system to validate that it meets business requirements and user expectations.

Deployment and Rollout

• Deployment Plan: Develop a deployment plan outlining the rollout strategy, including pilot testing and phased implementation.

• Training and Support: Provide training to users (e.g., underwriters, credit risk team) on how to interpret decision engine decisions and use the system effectively.

• Monitoring and Maintenance: Establish monitoring tools and processes to track system performance, detect anomalies, and ensure ongoing reliability and compliance.

Compliance and Security

• Compliance: Conduct audits and reviews to ensure the credit decision engine complies with regulatory requirements and internal policies.

• Security: Implement security measures (e.g., data encryption, access controls) to protect sensitive information and prevent unauthorised access or breaches.

Continuous Improvement

• Feedback Mechanism: Establish mechanisms to gather feedback from users and stakeholders to identify areas for improvement.

• Model Retraining: Periodically retrain models using updated data to maintain accuracy and relevance.

• Optimisation: Continuously optimise decision-making algorithms, strategies and rules based on performance metrics and business insights. Conduct multiple A/B tests to ‘test and learn.’

Considerations

• Ethical Considerations: Address potential biases in data and models to ensure fair treatment of applicants.

• Scalability: Design the engine to handle increasing volumes of applications without compromising performance and Turn Around Time (TAT).

• User Experience: Ensure the decision engine provides clear and understandable decisions to users and applicants.

Summary

By following these detailed steps and considerations, you can successfully implement a credit decision engine that supports efficient, consistent, and compliant credit decisions within your organisation. Each step plays a crucial role in ensuring the decision engine meets both business objectives and regulatory requirements while also providing a positive user experience.

By addressing both business and technical considerations in the implementation of a credit decision engine, organisations can build a system that not only enhances operational efficiency and risk management but also improves customer satisfaction and regulatory compliance.

Each consideration plays a critical role in ensuring the success and effectiveness of the credit decision engine within the broader organisational context.

About the Author

Stephen John Leonard is the founder of ADEPT Decisions and has held a wide range of roles in the global banking and credit risk industry since 1985.

About ADEPT Decisions

We disrupt the status quo in the lending industry by providing clients with customer decisioning, credit risk consulting, predictive modelling and advanced analytics to level the playing field, promote financial inclusion and support a new generation of financial products.