I’ve been making a case for why banks need to combine centralized model management with automated, configurable workflow tools. Having worked with clients to implement this approach, here’s one of the key lessons learned: The concept of model lifecycle management is broader than the life of any particular model. As analytics proliferate across organizations and the pace of change in financial services markets accelerates, banks need to start thinking in terms of lineages of the models themselves, as well as the predictive characteristics included in the model.
Regulators may, for instance, ask pointed questions about why a retired model was replaced with the current one and why a new characteristic was added to the latest model. When a model characteristic’s logic is changed by its author/owner, banks need to know where that characteristic is used so they can manage all downstream effects.
State-of-the-art model lifecycle management takes this broader view. Automated workflows help banks capture a complete lifecycle history of all models and their components. For each model, users can quickly track the lineage of any predictive characteristic—generated during development, harvested from a previous model, taken from a shared library, etc. For each characteristic, they can see everywhere it is currently used or was previously used—predictive models, decision strategies, etc. Banks can also gauge the value of individual characteristics over time and reduce the proliferation of characteristics that may offer limited value in comparison to existing characteristics.
Of course, there are a number of other considerations to ensure effective model management. For additional best practices, you may want to check out these Insights white papers: Reducing Regulatory Drag on Analytics Teams, Satisfying Customers and Regulators: Five Imperatives and Comply and Compete—Model Management Best Practices.