Situation & Context

A Data Governance Finding spurred production model run governance concerns. Prompting a review of process controls and execution workflows with goals of being able to execute models in a controlled, governed and efficient way. It was paramount that the efficiency gains be substantial to be able to execute at the nearly 2-3x higher volumes for the future scenario planning the bank needed.

Diagnosis

Overall production model execution processes were extremely manual with no E2E auditability. Controls and Evidence were generated adhoc in ways that often duplicated process or data - creating additional stumbling blocks to personnel resources and required processes. Data was sourced from dozens of sources independently, with very little automation or consolidation of data/computing power resulting in duplications of code and data. Finally, this was all compounded by the various countries, LOBs & portfolio segments which each had their own code bases, workflows and operated under their own rules / requirements.

Solution & Task

Create a standardized central model orchestration platform that could modularize the scenarios, models and portfolio inputs required for various production model executions for regulatory and non-regulatory purposes (adhoc management risk planning, back-testing, model monitoring, etc…). This needed to use the existing hardware and software, provide a comprehensive governance layer for all production model executions, and have the ability to automate various capabilities so it could keep up with the volume of scenario runs needed in the future.

Action

Designed & created architecture diagrams mapping existing data flows, software service calls, and operating process flows. With slight fixes and adjustments, we took these modernized process flows, and built corresponding standardized folder structures, naming conventions and placed code in modularized components. All of this enabled us to implement a new software model orchestrator layer on top of their existing compute hardware that called the client’s code in a governed way. We opened access to an expanded set of users, and generated reports of execution process for oversight and tracking.

Result

The first pass implementation achieved the equal or faster process execution speed with added governance and transparency. We provided the client with extensive process diagrams and data flow diagrams outlining client code call structure, data dependencies and process dependencies that highlighted risks and potential efficiency gains. Design and PoC work was done towards a fully automated model orchestration platform that could automatically execute controls and evidencing - while updating management reports on progress of execution so that a model execution could be completed with little to no human intervention.

This cover has been designed using images from Flaticon.com