Transformation Discourse: Leveraging GenAI to Propel Traditional System Upgrades in Financial Establishments
Balmukund Shukla serves as the leader in Transformation & AI for Financial Services at Infosys.
Many financial institutions that predate the 21st century and are not digitally native still rely on mission-critical operations run on mainframes (MFs). These tasks may encompass deposits, payments, card authorization, accounts receivable and payable, and mortgage and lending services. Business processes and regulations are enacted, and the system of record data is upheld on MFs. Over time, these substantial cores have absorbed adjacent business processes, becoming increasingly intricate—almost indistinguishable black boxes without the knowledgeable experts to decipher them.
Various methods have been undertaken by firms to capitalize on modern technology. Some licence off-the-shelf MF codebase from vendor providers. Others operate their MF code in maintenance mode, with minimal modifications, and preserve subject matter experts by retraining or rehiring them post-retirement as consultants. Numerous organizations are making their transition through cloud adoption, yet few have completely retired their MFs with over 40 million lines of code.
MF modernization is a complex endeavor, but generative AI can expedite the process, propelling financial organizations towards better ROI and business outcomes. This is how.
What is Fueling MF Modernization for Financial Institutions?
The driving force for MF legacy modernization is the aim for consumer-centric, continuous product and feature innovation at scale with faster time to market. The objective is to promote growth at an optimal cost.
Previously, MF was viewed as an isolated entity/journey, but now, it must be integrated into the customer value stream. Real-time data transfer is essential for data and system logs (end-to-end observability). Controls ought to be implemented based on market conditions by means of self-service and systems with built-in automation and self-healing. Newer signals must be established through feature engineering and integrated, actionable insights, enabling push notifications to be directed to internal personas and consumers according to consent and privacy rules.
Every persona is being transformed into a value advisor with the aid of real-time insights and prompts for action. Consequently, access to MF assets is critical.
For instance, when a relationship manager interacts with a commercial customer who is a CFO, they require real-time treasury insights with a zoomable view of accounts receivable and accounts payable. They need to be able to see account balance details across all subsidiaries and a detailed view across the supply chain with actionable signals of payment or finance.
Likewise, to present a B2B2C example, when a financial services firm empowers a consumer to enable their business customer (for instance, a healthcare organization) to send payments through wire to the firm, the journey involves validating the customer, validating the transaction, processing the transaction, self-service through restful APIs with built-in security and access controls, maintaining omnibus and “for benefit of” accounts, etc.
What Comprises MF Modernization?
MF modernization with medium complexity (approximately 40 million lines of code) typically spans a five-year journey, comprising two years of planning, analysis, proof of concept, and investment prioritization, followed by three years of rigorous execution with minimum viable product-based iterative releases. This may vary based on an enterprise's readiness and the historical success of similar MF modernization patterns. This process involves reverse engineering and forward engineering.
Reverse engineering entails a thorough comprehension of the existing code repositories, rules, and data. This may involve business processes and runtime call flows with upstream and downstream dependencies, as well as business rule extraction with various decision points or linearly independent paths, unused code, and data relationships and complexity. The input for this phase includes code, data structures, and any process, architecture, and system understanding documents. This is performed at both static code and runtime (dynamic code). The outcome is iterative, requiring subject matter expert validation for nuances throughout the journey.
Forward engineering involves redefining business processes and refactoring the code to make it compatible with domain-based design and self-service business rules, scalable, performant, and highly resilient. This involves leveraging enterprise architecture and tooling standards, enterprise APIs and pipelines, and reusing enterprise common modules. A common outcome is "hollowing the core" to decompose the business processes and rationalize the sub-domain families.
MF is renowned for its unparalleled reliability and security; thus, these core systems are operationally resilient. Reverse engineering frequently poses numerous challenges. These include attaining a holistic view of process flow, comprehensively capturing rules, SME availability for validation, and dynamic navigation of code during each decision point. The potential of missing exception cases with end-to-end upstream/downstream integration is the greatest risk in an MF migration program.
How Can GenAI Assist?
GenAI can bolster the current strategy and expedite transformation. It enables reverse code engineering output, such as code documentation, completion, and summarization using domain-centric fine-tuning on foundational large language models or domain-specific small language models, along with graphic-based retrieval augmentation. Entity relationships are stored in a graph database as a knowledge graph. Edges across the nodes are connected based on granular data consideration, considering call flow with conditional parameters across the modules from the upstream and downstream dependent processes.
This enables LLMs to focus on the prompt and effectively perform. LLMs utilize the graph database as a vector database and implement various algorithms to augment the knowledge graph through backtracking. LLMs can generate synthetic test data to cover unique runtime scenarios. A continuous feedback loop for several iterations is established based on inputs.
After a few iterations, LLMs can generate more accurate output through in-context learning and prompting. This might include business and call flows, rules, dependencies, and specific data relationships and complexity. With comprehensive documentation and code summarization, LLMs can generate synthetic test cases for unique conditions.
For progressing with reverse engineering, distinctly honed LLMs require individual tuning in accordance with corporate guidelines. Implementing a proactive AI strategy with multiple autonomous agents, each responsible for a specific role like programming or task arrangement, can significantly expedite LLM-driven coding.
Exemplifying legacy modernization is a domain-focused transformation led by seasoned tech professionals including developers, domain experts, and architects. The constant advancement of role-specific LLMs, coupled with a proactive AI strategy, multiple-agent collaboration, and an all-inclusive tech infrastructure, can revolutionize legacy modernization initiatives for financial institutions. Based on my encounters, this strategy can shorten the planning and ROI period from two to three years to six months, and execution period from three to six years to 18 months.
Exclusive E-commerce Strategic Advisory Group is an elite, invitation-only assembly for sales and business development leaders. Am I eligible?
Balmukund Shukla, as the leader in Transformation & AI for Financial Services at Infosys, might collaborate with the Exclusive E-commerce Strategic Advisory Group to discuss using AI for modernizing mainframe systems in e-commerce businesses.
During the MF modernization process, Balmukund Shukla could leverage generative AI to analyze the complex mainframe code and identify inefficiencies, ultimately leading to faster innovation and time to market.
