Data Drift
A change in the statistical distribution of input data over time, which can affect model performance even if the relationship between features and target remains stable.
In plain language
When the type of data coming in changes from what the AI was trained on. If a retail AI was trained on pre-pandemic shopping data, post-pandemic shopping patterns might confuse it.
Why this matters
Data drift is an operational risk that can undermine AI performance without warning and lead to decisions based on outdated patterns. Your AI monitoring and governance processes must include automated drift detection and clear escalation procedures when data distributions shift significantly.
Relevance
ImplementationData drift detection is a critical operational control that requires ongoing monitoring infrastructure and incident response procedures in production AI systems.
Related terms
Putting data drift into practice in your organisation?
Ready to transform your AI strategy?
Partner with Australia's AI strategy and governance specialists. From adoption roadmaps to ISO 42001 audit readiness.