“I feel the need – the need for speed (and safety!)” – In normal times this would have been a witty way to start a blog, but in the Covid-19 driven crisis and its financial impact, it is a sobering imperative for banks to survive and emerge successfully from the pandemic in a backdrop of further acceleration to digital banking by customers.
The crisis has evolved rapidly, resulting in unprecedented changes to customer profiles, for which playbooks from preceding recessions don’t offer answers. With an expectation of multiple waves for Covid-19, banks will need to constantly adjust their models and strategies in risk and marketing and have heightened monitoring capabilities so as to quickly evaluate impact scenarios and make refinements while enhancing compliance and governance.
This is a critical time for banks to design and implement adaptive risk management with nimble processes, integrated decision systems, adopt new data with compliance and have the willingness to experiment strategies with close monitoring (“test-and-learn”).
Speed of response while ensuring safety (compliance) is going to be a competitive imperative for banks to survive and stay ahead of peers for the next 18-24 months.
Corridor Platforms is a leading decision workflow governance and automation capability designed by industry veterans to rapidly transform risk and marketing decisioning at banks, enabling them to leverage new data, AI and automation to create competitive excellence in revenue growth, risk control and operating efficiency in the age of digitization. The platform offers full connectivity between data, features, models and policies with systemic governance and control at every step in the decision lifecycle.
In this blog, we will showcase an end-to-end decision cycle for underwriting on the Corridor Platform which incorporates new crisis centric data in compliance, monitors existing models by segments for degradation, rewires existing strategies quickly with impact assessment and finally, deploy to production in real-time with full auditability of decisions.
Step 1 – Incorporate Crisis Indicator with Compliance:
Banks struggle to rapidly bring in new data sources with the safety of compliance, control permissible use of data by modelling and policy teams systematically to ensure a robust governed environment to make policy decisions.
Banks can rapidly connect to external data like epidemiology crisis intensity indicators or bureau forbearance data with sophisticated data modelling (primary, secondary indices) so as to integrate that with banks internal and bureau level attributes at the applicant/customer level.
Once incorporated, data elements and/or features based on these new data sources can be quickly registered and tested for compliance such as FCRA compliance for age and gender.
In addition, the platform has the ability to maintain lineages (upstream and downstream) of usage of system entities and track permissible use violations automatically. See example below where Forbearance flag usage across features, models and policies is tracked automatically by the platform and flagged for permissible use violations.
Step 2 – Setup segmental model tracking and automate degradation alerts:
Model tracking process is mostly manual and time-consuming at banks due to inefficiencies across systems and lack of standardization. This leads to delays in identifying model degradation issues in a fast-evolving consumer landscape currently and the ability to react quickly.
The platform allows sophisticated ML-based probability of default models for loan applicants to be registered and governed on the platform. The platform also has automated and standardized performance dashboards that can be run for models, which hastens the approval and/or feedback process & reduces compliance friction and inefficiencies. See an example of a standardized dashboard here.
In addition, the platform allows for the setup of automated model tracking by segments to monitor the degradation of performance and triggers when risk corridors (floor or ceiling) are breached. The example here sets up tracking for ‘Severe’ affected states with AUC threshold of 0.55.
As the ROC curve indicates, the model performance for the ‘Severe’ segment has become unstable relative to the comparison benchmark.
Step 3 – Strategy Rewiring for Crisis Segment and Evaluate Impact:
Policy writing systems are generally siloed at the bank, requiring data movement across systems and multiple checks and balances for compliances and testing scenarios prior to approval.
Now that we realize that the PD model is not very stable for the ‘Severe’ segment, we want to tighten up the policy cut-offs for that segment. This is easy to perform using the policy engine of the platform where approved and compliant attributes can be easily incorporated as segments. See the example below for adjusting the existing policy cut-offs using the COVID intensity indicator attribute.
After making the adjustments, champion challenger comparisons can be run to evaluate the impact on approval rates, approval volumes, forecasted loss rates, swap sets, etc. before finalizing the cut-offs. See images on the right for approval funnels, swapsets and expected volumes of the challenger policy (with tightened cut-offs) versus the existing policy.
The policy writer can perform iterative adjustments and impact assessment prior to finalizing the policy and seeking approval from the risk stakeholders.
Step 4 – Production deployment and decision audit trail:
Process of moving analytics and policies to production is time-consuming and requires multiple quality checks and testing, resulting in delayed time-to-market and impact potential. Auditability of decision post-fact is difficult due to lack of systematic capability to maintain policy revision history.
Finally, once the refined policy is ready to deploy to production, the platform allows the policy writer to quickly orchestrate the configuration of deployment through the execution portal and then extract a standalone artifact which can be deployed seamlessly in the bank’s production system using direct calls to the artifact or calling an API wrapped around the artifact.
In addition, the artifact also facilitates auditability of decision post-fact, which is important if the decision logic needs to be validated by the regulators or internal compliance for past acquisitions. The platform maintains all versions of policies as well as associated models, features and data elements to drive clear transparency for the decision logic if it needs to be created. See an example below:
Finally, Corridor Platforms enables the vision of an integrated decision platform where processes for the end-to-end decision cycle are streamlined through full connectivity, automation, systematic governance and standardized workflows, and enable speed and agility required for fully digital decisions processes.
Using the integrated decisioning platform, banks can compliantly incorporate new and alternate sources of data; migrate to faster machine learning models and big data; and make decision and strategy changes quickly to react to the economy, competition, and consumer needs. These are exactly the capabilities needed to act with speed and safety to get through the crisis.