Model Drift
Definition
The degradation in a machine learning model's predictive accuracy over time as the statistical properties of the input data diverge from the training data distribution. Model drift requires ongoing monitoring and periodic retraining to maintain performance, and is a key operational risk in production AI systems. The cost of managing model drift is an important consideration in AI asset valuation.
Complementary Terms
Concepts that frequently appear alongside Model Drift in practice.
A type of neural network trained on vast corpora of text data, capable of generating human-like text, answering questions, summarising documents, and performing reasoning tasks. Large language models such as GPT and Claude represent significant R&D investment and are reshaping knowledge work, customer service, and content production across industries.
A dividend discount model that values a perpetual stream of cash flows growing at a constant rate, calculated as the next period's cash flow divided by the difference between the discount rate and the growth rate. The Gordon growth model is widely used to estimate terminal value in discounted cash flow analyses and requires that the assumed growth rate remains below the discount rate.
A set of practices combining machine learning, DevOps, and data engineering to standardise and streamline the end-to-end lifecycle of machine learning models, from development through deployment to monitoring. MLOps encompasses version control for models and data, automated testing, continuous integration and deployment, and model performance monitoring in production.
A virtual representation of a physical asset, process, or entire business operation that uses real-time data and simulation to mirror its real-world counterpart. Digital twins enable predictive maintenance, scenario modelling, and operational optimisation.
A centralised platform for storing, managing, and serving the engineered features (input variables) used by machine learning models in both training and real-time inference. Feature stores ensure consistency between training and production environments, enable feature reuse across multiple ML models, reduce duplication of feature engineering effort, and provide a governance layer for tracking feature lineage and ownership.
A category of artificial intelligence systems capable of creating new content — including text, images, code, music, and video — based on patterns learned from training data. Generative AI is transforming content production, product design, and software development, raising novel questions about intellectual property ownership and the valuation of AI-generated outputs.
A subset of artificial intelligence in which algorithms improve their performance on a specific task through exposure to data, without being explicitly programmed for each scenario. Machine learning encompasses supervised learning, unsupervised learning, reinforcement learning, and deep learning techniques.
An AI architecture that combines a large language model with an external knowledge retrieval system, enabling the model to ground its responses in verified, up-to-date information rather than relying solely on its training data. RAG reduces hallucination risk, improves factual accuracy, and allows organisations to deploy AI systems that reference proprietary knowledge bases without retraining the underlying model.
Further Reading
AI Integration Costs: The Hidden Expenses of AI Adoption
AI licensing fees represent only 20-30% of total AI adoption costs. The remaining 70-80% — data preparation, integration engineering, change management, and ongoing maintenance — are the hidden expenses that derail AI budgets and timelines.
Read more →Put this knowledge to work
Use Opagio's free tools to measure and grow the intangible assets that drive your business value.