By: Tarun Kapoor, Managing Principal – Architecture, Banking Technology & Architecture Office

Today, consumers expect a better experience and more personalized offerings from their financial service providers. This has led to an increase in unbundling^[i] of banking services, where consumers pick and choose from multiple providers rather than relying on a single bank. Digital banks are capitalizing on this trend and making inroads by offering modern, flexible, and affordable products.

On the other hand, traditional financial institutions are struggling to compete due to their legacy core banking infrastructure. While many have already started modernization, multiple challenges have plagued their journeys, and successful migration remains elusive. According to McKinsey, the financial services sector spends about 70%^[ii] of its IT budget on maintaining legacy systems, which hampers innovation and agility while increasing operational risks and costs.

Banks have tried to remedy this by leveraging generative AI or Gen AI to run pilots to improve their operations and services. Gen AI is also expected to add up to $350 billion^[iii]to the industry’s annual value. However, to fully utilize Gen AI’s capabilities, banks require a set of resilient, scalable, and adaptable core technology components that are missing in their legacy infrastructure.

In this blog, we will discuss the challenges of legacy infrastructure that keep traditional banks from competing with their digital-native peers and how implementing Gen AI can accelerate legacy modernization.

Challenges of legacy banking applications

Legacy systems in financial institutions are often plagued by inefficiencies, high maintenance costs, and significant integration issues. These outdated systems can’t keep pace with rapid technological advancements and increasing regulatory demands. Here’s why:

Stifle competition: Legacy systems hinder traditional banks’ ability to launch new products quickly, making it challenging to meet customer expectations and stay competitive with neobanks and fintechs.

Raise performance issues: The monolithic infrastructure put in place 30-40 years ago can no longer handle the demands of the new banking system. These legacy systems struggle with stability and performance issues, leading to frequent downtimes. Moreover, integrating these old systems with modern applications and platforms is quite complex.

Hamper regulatory compliance: Rising cybersecurity, data privacy, and governance challenges have led to tighter regulations, requiring banks to generate new types of reports and ensure higher levels of transparency and accountability. However, the legacy infrastructure cannot meet these demands, forcing banks to find workarounds manually.

For instance, a leading Asian bank was mandated by local regulatory authorities to ensure a minimum availability of its core banking system, which its legacy infrastructure failed to meet. The bank was forced to modernize its system to avoid hefty fines and ensure regulatory compliance.

Create operational risks and increase cost: Modern cloud-native core banking systems allow banks to utilize optimized cloud capacity, leading to significant cost savings. Legacy systems, however, lack this flexibility, resulting in higher operational costs and inefficient resource utilization.

Most banks have started legacy modernization but face critical challenges, leading to long transformation programs that often take years and fail to achieve set objectives. These challenges underscore financial institutions’ importance in utilizing generative AI for their modernization programs.

Role of generative AI in legacy modernization

Generative AI is already transforming how financial institutions approach legacy modernization, offering comprehensive solutions throughout the lifecycle. Here’s how:

Legacy system document generation

Generative AI can assist in analyzing existing systems/code/configuration to extract detailed insights and generate comprehensive documentation of legacy applications. This assisted analysis helps identify business logic, dependencies, and integration points, often poorly documented in legacy systems.

Generating modern system requirements and user stories

Once the legacy documentation has been generated, generative AI can assist in creating requirements and user stories for the target system by simulating various business scenarios and predicting user interactions with the new system. It can also assist in prioritizing these user stories based on impact and feasibility, ensuring that the most critical functionalities are addressed first. This approach helps align the development process effectively with business goals and consumer needs.

Designing and building application

Gen AI tools can assist and accelerate the creation of design documents and system architectures. AI can recommend best practices and design patterns, ensuring the new system is scalable, secure, and efficient. In the build phase, AI-driven code generation tools can convert legacy code to modern frameworks and languages with code quality aligned with the enterprise’s standards and generate unit and system test automation scripts. According to Forrester, AI-driven development can reduce coding effort by up to 50% and improve application time-to-market by 20%^[iv].  As part of the build phase, Gen AI can also streamline the following:

• Code refactoring and conversion: Refactor and translate legacy code into modern languages such as Java or Python by using Gen AI to understand the existing codebase, identify redundant or obsolete code, and rewrite it in a modern, efficient manner.
• Data migration: Facilitate seamless data migration by mapping old data structures to new ones, improving data integrity, and minimizing data loss or corruption during the transition.
• Automated testing: Generate test cases and test automation scripts to ensure the new system functions correctly and meets all requirements. This can minimize errors and accelerate the testing phase.

Application deployment

Gen AI can generate deployment scripts and Continuous Integration/Continuous Deployment (CI/CD) pipelines. This enables a smooth transition from legacy systems to modern infrastructure with minimal downtime. AI-driven deployment strategies also include automated testing and rollback mechanisms to quickly identify and fix any issues that arise post-deployment.

Continuous monitoring and optimization

Post-deployment, different frameworks can be used to integrate Gen AI models with monitoring and observation tools. These tools derive insights from system usage and errors and, therefore, suggest optimizations to improve efficiency and reliability.

Generative AI can accelerate this transformation by predicting maintenance and through analysis of system logs and performance data, enabling proactive issue identification and reducing downtime. Automated alerts and ticket generation for recurring issues further streamline IT operations, ensuring efficient resource management.

Additionally, Gen AI-driven optimizations can enhance system reliability and performance by analyzing resource usage patterns and suggesting optimal configurations, leading to cost savings and improved performance.

Gen AI for mainframe modernization

Any talk around Gen AI and its utilization and benefits in use cases will be incomplete without proper validation and application.

LTIMindtree’s BFS Finnovation CoLab has carried out extensive proof of concepts (both internally and for our clients) to modernize and convert mainframe banking applications of simple and medium complexity.

The overall lifecycle involves significant pre-processing of legacy application code and data before Large Language Models (LLMs) can be utilized for any document generation and code conversion. We’ve used AI/ML-friendly frameworks and languages like Python, LangChain, and Llama Index for this phase. Post the LLM assist phase, a post-processing phase is initiated to get the generated artifacts in a form usable by stakeholders. This validates two key points:

1. Gen AI can assist and accelerate legacy application modernization across phases, reducing efforts by 25-30% and achieving an accuracy of 70-80%, depending on the application’s complexity. Further efficiencies can be achieved via improved prompt engineering and surrounding frameworks.
2. Human-driven pre-processing, post-processing, validation, and review phases underscore the human-in-the-loop requirement.

In conclusion

The transformative potential of generative AI in legacy modernization is undeniable. Banks and financial institutions are already running Gen AI pilots to optimize and automate their software development lifecycle, customer experience, and data analytics. Moving forward, these use cases will be scaled enterprise-wide, and new ones will be introduced, including fraud detection, financial crime prevention, and the generation of compliance and regulatory reports.

However, leaders must approach Gen AI-driven legacy modernization with a clear understanding of its limitations and implications. So far, most Gen AI applications use a human-in-the-loop model, and given the regulations and data implications, it will most likely remain that way. Additionally, any use case with Gen AI should be thoroughly evaluated for accuracy and business value before it can be industrialized. Lastly, Gen AI is not a one-size-fits-all solution but a tool that, if used strategically, can create a lasting foundation for competitive advantage and innovation.

References

[i] Bain and Company, Customer Behavior and Loyalty in Banking: Global Edition 2023, https://www.bain.com/insights/customer-behavior-and-loyalty-in-banking-global-edition-2023/

[ii] McKinsey, Winning in digital banking, https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/tech-forward/winning-in-digital-banking

[iii] McKinsey, Scaling gen AI in banking: Choosing the best operating model, https://www.mckinsey.com/industries/financial-services/our-insights/scaling-gen-ai-in-banking-choosing-the-best-operating-model

[iv] Forrester, Predictions 2024: Software Development Adapts To TuringBots, Ajar Source, And Backstage, https://www.forrester.com/blogs/predictions-2024-software-development/