Our client – largest retail Bank in Middle East – wanted to develop and deploy a solution based on a business rule management system (BRMS) for hosting multiple behavioural scorecards:
Our client started out in the banking industry in the late 1960s. Now, it provides banking services in Asia-Pacific (APAC), Europe, the Americas, Middle East, and Northern Africa (MENA) regions. The bank offers deposits, personal loans, e-banking, trade finance, foreign exchange, and other banking services worldwide.
The Bank purchased a business rule management system (BRMS) from one of the top analytic companies in the US. With the BRMS, business and risk team users are capable of:
Configuring predictive models of various types, including machine learning.
Importing and exporting models in PMML format.
Configuring business rules and complex calculation logic.
The Bank was keen to set up the scorecard management by risk managers themselves, with no IT team involvement.
The vendor recommended our team as an experienced product integrator and the experts in automating decision-making strategies.
Though the Bank was geographically situated thousands of miles away from us, this didn’t slow our team down one bit. The project was 100% remote and delivered joint cooperation between teams in Great Britain, India, Russia, Belarus, and UAE.
The Bank expected ABLE Platform’s team to deliver a turnkey solution with the following features:
Scorecard hosting and data preparation. We had to configure the mechanism of importing scorecards, including importing and adjusting models, pre-calculation of derived variables for models and data quality checks.
Extraction of data from Bank systems. This part of the job was completed by the Bank’s IT department. Our task was to convert the data given to us by the Bank into scorecard input variables. We were also asked by them to configure the logic of exception processing.
Performance requirements. Our task was to meet the Bank’s performance requirements and make it possible to process a certain amount of data within a particular time limit using the equipment the client has available at the moment. And during the batch job, the server CPU load must not exceed 80%.
Business user-friendly interface. Our specialists had to configure the web interface layer of BRMS in such a way so that risk managers could work with scorecards by themselves.
The main task of ABLE Platform’s team was to create a mechanism for scorecard import for non-IT users and create a batch engine for the BRMS.
After a brainstorming session, our team found the optimal solution: to create a standalone multi-threaded application—Batch Web Application. This would coordinate data processing on the BRMS side. It also allowed us to avoid complicated BRMS customization and benefit from more advanced data processing methods.
Within the framework of the project, ABLE Platform’s team completed the following tasks:
Configured import for 12 behavioral scorecards, which covered all Bank’s retail products (auto loans, credit cards, consumer loans, mortgage). All models are imported in PMML format and stored in a single repository, which allows for the re-use of derived variables calculation logic.
Configured the web interface for model adjustment, allowing the Bank’s employees to change the scorecards, pre-process data logic, segment clients effectively, and complete data quality rules without IT.
Implemented the logic of the exclusion processing.
Developed a general solution architecture in the form of a multi-threaded web application—Batch Web Application.
Met the client’s performance requirements: we made it possible to process a certain amount of data within the Bank’s process window frame using the client’s hardware.
Enabled the regular batch of the models in automatic mode. For credit cards, 15 times per month, and for other models, once per month. The models are used in batch mode and cover the whole loan portfolio of the Bank.
Ensued technical support during user acceptance test (UAT) and production environment implementation.
Batch Web Application is a multi-threaded batch wrap-up for the BRMS. Its repository hosts scorecards, derived variables calculation, and exclusion logic.
We designed the application to be the only trigger for initialization or stopping BRMS rules. Our specialists created around a dozen URL commands to control the Batch Web Application lifecycle (start/end of execution, pause, display the result, etc.). These commands can be executed either automatically or manually.
URL-commands can be run via HTTP requests. For this purpose, we designed a web interface for business administrators to control the data processing execution.
ABLE Platform’s technical specialists created a logging mechanism that holds details for data pre-processing and a description of errors occurring during calculation.
Geographically, the Bank is situated far from our team’s location. That’s why, in our office, we build hardware configurations identical to that of the client to undertake the stress tests.
Additional requirements for server load. For the purpose of centralized monitoring and the stability of processes, server load should be manageable. For the purpose of remote monitoring components used by IT, the production server CPU load should not exceed 80%.
Our team made the mechanism of multi-threaded calculations in Batch Web Application more advanced. For example, the number of parallel threads could dynamically change during the calculation process. Thus it’s possible to control the server load. In doing so, the performance of the solution remained high enough and we could meet the given process window.
Extended scope of testing. During UAT, the Bank also asked us to include a part developed by the Bank’s IT department within the scope. In the final result, we tested Batch Web Application and BRMS (our part of work), and ETL processes of data extraction from source systems (Bank’s part).
The Bank received the completed project after 4 months. It took us 2 months to finish the UAT, including testing the ETL developed by the Bank’s IT department. Another 2 months were spent on piloting the solution to ensure it was running correctly.
As a result, our client got the following opportunities:
1. Full coverage using behavior scorecards of all Bank’s retail products.
2. Mechanism for derived variables calculation and data quality management.
3. Configurable mechanism for scorecard hosting, including model adjustment, versioning, and audit of changes.
4. Business user-friendly web interface, which allows the risk team’s employees to manage the solution by themselves. The IT department is responsible only for server administration and the ETL.
5. Optimal software performance within the boundaries of the specific timeframe, data volume, and Bank’s existing hardware.
