model decay
What It Means
Model decay happens when an AI system that once performed well gradually becomes less accurate and reliable over time. This occurs because the real-world data the model encounters changes from the data it was originally trained on, making its predictions increasingly outdated and wrong.
Why Chief AI Officers Care
Model decay can silently erode business value and create significant risks, as decisions based on deteriorating AI outputs become less reliable without obvious warning signs. It requires ongoing monitoring and maintenance investments to detect and address, making it a critical operational cost and risk management concern that can impact customer satisfaction and regulatory compliance.
Real-World Example
A credit scoring model trained on pre-pandemic data might start approving more risky loans because it hasn't learned about changed employment patterns, remote work impacts on income stability, and new spending behaviors that emerged after 2020, leading to higher default rates despite the model showing the same confidence scores.
Common Confusion
People often think model decay only happens when something breaks or changes dramatically, but it's usually a gradual, invisible process where small shifts in data patterns accumulate over time. It's also confused with model bias, though decay is about changing accuracy over time while bias is about unfairness baked into the original training.
Industry-Specific Applications
See how this term applies to healthcare, finance, manufacturing, government, tech, and insurance.
Healthcare: In healthcare, model decay poses critical patient safety risks as AI diagnostic and treatment models become less accurat...
Finance: In finance, model decay occurs when credit scoring models, fraud detection systems, or algorithmic trading models degrad...
Premium content locked
Includes:
- 6 industry-specific applications
- Relevant regulations by sector
- Real compliance scenarios
- Implementation guidance
Technical Definitions
NISTNational Institute of Standards and Technology
"Model decay depicts that the performance of the model is degrading over time"Source: Nayak,_Pragati
Discuss This Term with Your AI Assistant
Ask how "model decay" applies to your specific use case and regulatory context.
Start Free Trial