The Key Characteristics of a Successful Model Inventory
Once a company realizes it's time to organize its predictive risk model portfolio, one of the first steps will be introducing a model inventory.
The first act should be reaching an internal consensus about what is regarded as a model and therefore what should be kept in the inventory (no easy task itself). The next move is to get all models and their metadata stored in one place. From this perspective, a model inventory can seem... well, just like a plain inventory, something that could be easily coded by that skillful intern from the IT department in a few weeks, or even created in Excel in half an hour.
If it's so easy, why am I still writing about it?
Model inventories start to become more complex when one starts contemplating and later implementing them. They very often become an elementary component for all later model governance and model risk management related efforts. Having a solid foundation is key to smooth sailing in the future. To help you through the process of building a solid model inventory, I've created a list of qualities that you should require your model inventory to possess so it will become a long-lasting asset instead of a technological burden.
Your model inventory should be...
Flexible
Flexible enough to store any type of model used in an organization, regardless of its type (credit, market, valuation, operational risk, or even marketing). Adding or altering a new model metadata field should also not result in months of endless waiting, email ping-pong, new releases, and several downtimes of the whole application.
Integrable
Integrable to your organization's infrastructure. Yes, a model inventory can be standalone, however, the benefits of an inventory that can be connected to various corporate systems (either in terms of data import or export) are clear. An easily integrable inventory can become a central point where all the relevant data, documents, source codes, monitoring and stress testing results are stored, a fact which helps it earn its single source of truth (SSOT) status.
Accessible
Imagine if anyone from any branch of the organization who has some value to add to the model lifecycle had user-friendly access to the inventory. Access can be given to C-level executives all the way down to a receptionist (if there's an internal need for that, of course). This accessibility results in a more robust and efficient model governance framework. Of course, access to data and functions of the inventory should be granted accordingly to guarantee data security and confidentiality.
Try FreeSmart
Additional business logic reflecting the company's own methodology should be applicable to the inventory. This starts with basic checks like "model owner cannot be the same person as model validator" or "if the risk score of a model is above a certain threshold, the model status should not be allowed to be changed to production-ready." More advanced, complex business logic could later be visualized using workflows, combining both automatic and human tasks. This enforces the model governance methodology across the entire organization, makes daily operations of model developers, validators, and analysts much easier, and can also contribute to regulatory compliance.
Ready for future magic
The inventory is usually a starting point, so it should be designed in a way that makes further technological enhancements in model governance and model risk management possible. With an inventory designed in such a smart manner, a whole new world of opportunities opens: real-time reporting, data analysis, and my personal favourite, automation (e.g. automated model risk tiering or monitoring). It is crucial to keep in mind that these advanced features, which provide incredible business value in mid and long term perspectives, are impossible without a robust and open inventory.
Scalable
The number of models and connected regulations is rising constantly, which should be reflected by the inventory's architecture. Storing information about 4 or 4,000 models should not make a difference. The increasing regulatory pressure should be kept in mind as well, so the inventory is ready for future adjustments reflecting these changes. The use of a model inventory in organizations is becoming indisputable, a condicio sine qua non.
Build on rock, not on sand
As the saying goes, "it is wise to build a house on a rock rather than on sand." Accordingly, the model inventory, despite its apparent basic nature, should be well-rounded so as to become a solid rock for further model governance and model risk management efforts.
Of course, these are not the only characteristics that distinguish sub-standard and excellent model inventories. What other outstanding qualities does your ideal inventory have?
Ready to build your AI model inventory? Try Model Inventory for Jira — free 30-day trial, no credit card required. Your models organized in the tool your team already uses.