Final analysis:

Without mitigations like continuous monitoring, adaptive retraining, or diverse data sourcing, the likelihood approaches 90%+ for most models within months to years. However, well-maintained systems (e.g., via MLOps practices) can reduce this to near zero, as degradation is largely a solvable engineering problem rather than an inescapable fate. In critical applications like healthcare or finance, unchecked degradation could lead to biased decisions or systemic failures, underscoring the need for robust oversight.