Predictive analytics

The use of statistical algorithms, machine learning techniques, and data mining to identify the likelihood of future outcomes based on historical data.

Source: AIEOG AI Lexicon (Feb 2026), adapted from NIST AI 100-1 and industry usage

Predictive analytics is the practice of using data and algorithms to forecast what is likely to happen next. It takes historical patterns and uses them to estimate the probability of future events — whether a borrower will default, whether a transaction is fraudulent, whether a customer will churn, or whether a market will move in a particular direction.

Predictive analytics has been used in financial services for decades (credit scoring is a classic example), but AI and machine learning have expanded both the sophistication and the scope of predictive capabilities. Modern predictive analytics can process larger datasets, incorporate more variables, detect nonlinear relationships, and update predictions in near real-time.

The distinction between predictive analytics and AI is blurry. Traditional predictive analytics often uses statistical methods like logistic regression. AI-powered predictive analytics uses machine learning techniques like gradient boosting, neural networks, or ensemble methods. The governance obligations apply regardless of the technique.

Predictive analytics is foundational to financial services operations. Credit risk assessment, fraud detection, AML transaction monitoring, customer lifetime value modeling, insurance underwriting, and market risk estimation all rely on predictive models.

Because predictive analytics directly drives decisions that affect customers and the institution’s risk profile, it falls squarely within model risk management requirements. Predictive models used for lending decisions must also comply with fair lending requirements, adverse action notice obligations, and data privacy regulations.

1. Apply model risk management. Predictive analytics models are models under SR 11-7 and should be governed accordingly.
2. Validate prediction accuracy. Test that predictions are accurate, stable, and performing within expected parameters.
3. Assess for bias. Predictive models can reflect and amplify historical biases. Test for disparate impact across protected classes.
4. Ensure data quality. Predictive models are highly sensitive to data quality. Implement data quality controls and monitoring.
5. Maintain explainability. For customer-facing predictions, maintain the ability to explain why a particular prediction was made.
6. Monitor over time. Predictive accuracy can degrade as the underlying data patterns change. Implement ongoing performance monitoring.

Machine learning, AI model, Algorithm, Supervised learning, Data quality/validity, Benchmarking