Industry
Banking
Banking Services Transformation: Unlocking Exponential Value
Highlights
- Mainframe modernisation in the BFSI industry is obstructed by significant challenges, including fragmented documentation, a limited pool of experts, and IT teams often have knowledge of only 20-30% of the application codebase.
- Firms must adopt a knowledge-first strategy, using generative and agentic AI to build a comprehensive knowledge hub through a structured process of discovery, automated documentation, and human-in-the-loop validation.
- This AI-powered mainframe knowledge hub will serve as a living repository, providing conversational access to reduce dependence on experts, cutting migration costs, and preserving critical system knowledge to de-risk future transformations.
On this page
Overview
Our long experience of engaging with global clients in the banking, financial services, and insurance (BFSI) industry reveals that mainframes continue to power nearly 90% of global financial transactions and around 40% of core banking systems. The percentage is still higher in the insurance sector—approximately 90% of insurers depend on mainframes for mission-critical core processing. Vital business logic and rules reside on mainframe systems, and many of the original developers are no longer available. Legacy skill shortages and accumulated technical debt impact agility, increase operational risk, and delay new technology adoption. Even routine enhancements to existing applications can become time-consuming, costly, and high-risk undertakings. As a result, BFSI institutions face growing challenges in maintaining, enhancing, and evolving their legacy systems to meet changing business demands.
For decades, BFSI institutions have undertaken modernisation initiatives aimed at either extending the value of their mainframe investments through in-place modernisation or moving select capabilities to modern platforms. However, these efforts have often been risk-prone, with extended execution timelines, high costs, and limited business outcomes. The approach needs to shift: BFSI firms must enable a progressive transition to modern architectures aligned with business objectives.
In our view, BFSI firms should adopt a knowledge-first modernisation strategy, using AI to build a comprehensive knowledge hub as the foundation. There are three important steps in the mainframe modernisation journey:
- Step 1: Build a mainframe knowledge hub.
- Step 2: Introduce modern practices into mainframe software engineering.
- Step 3: Modernise mainframes by progressively moving workloads to a modern architecture.
BFSI firms can either choose to retain and modernise their existing mainframe estate or move to a hybrid model where some workloads stay on mainframes while select capabilities move to a modern architecture. A third option is to exit mainframes over the long-term by progressively moving workloads to modern platforms. The path forward will depend heavily on organisation-specific priorities and the size of the mainframe estate. However, it must be noted that these steps are not necessarily sequential. Depending on organisation-specific priorities, firms may choose to modernise mainframes while concurrently moving specific capabilities to modern architectures, minimising the quantum of workloads on mainframes—preparatory to either a partial or complete exit from mainframes.
Regardless of the strategy adopted, creating a comprehensive knowledge hub is the critical first step. A comprehensive knowledge hub can significantly mitigate mainframe-related challenges by providing visibility into legacy applications, business rules, data structures, dependencies, and operational processes. This enables informed decision-making and reduces risks across both modernisation and transformation initiatives.
The first of a three-part thought paper series on mainframe modernisation in BFSI, this paper will dive deep into how firms can go about building a mainframe knowledge hub (step 1). The second paper will focus on modernising mainframe engineering (step 2), while the third will dwell on exiting mainframes (step 3).
Why is mainframe modernisation difficult?
In BFSI institutions, absence of sufficient documentation for critical mainframe systems makes mainframe modernisation highly challenging. Often, firms rely on documentation that is fragmented, outdated, and incomplete. The limited pool of professionals with mainframe expertise is heavily occupied in maintaining and enhancing systems that run on mainframes. Furthermore, IT teams typically have working knowledge of only 20–30% of the application codebase—primarily those related to ongoing maintenance and enhancement. Understanding of the remaining 70–80% is often limited or minimal. As a result, the effort and time required to undertake a comprehensive documentation initiative are often seen as prohibitively high. Compounded by the shrinking availability of subject matter experts (SME), this causes modernisation efforts to be repeatedly deferred in favour of immediate operational priorities.
Addressing this challenge has become a business imperative and will necessitate building a knowledge hub—a unified, context-rich knowledge layer that captures, organises, and surfaces the core business knowledge embedded within mainframe systems. Such a hub must provide conversational, intuitive access to critical system knowledge, enable faster understanding, improve insight-driven decision-making, and reduce dependence on a shrinking SME base, while building a strong foundation for engineering and modernisation.
How can BFSI firms build a knowledge hub?
The advent of generative AI (GenAI) and agentic AI offers intelligent and scalable approaches to building an enterprise knowledge hub quickly, cost-effectively, and sustainably. A structured multi-step process that combines GenAI-driven automation with human-in-the-loop oversight at every stage will ensure that the outputs are continuously verified, validated, and enhanced for enterprise use (see Figure 1).
Figure 1: Detailed step-by-step process to build a knowledge hub leveraging agentic AI and human-in-the-loop oversight
Discovery and blueprinting: This is the first step involving a thorough analysis of the mainframe inventory. We recommend a hybrid approach that combines deterministic automation with GenAI to produce a comprehensive application map, including all the components, dependencies, and relations required to understand the system landscape. The generated output must subsequently be validated by SMEs through an iterative process. Human oversight and feedback will result in a high quality, more relevant, and accurate blueprint. This foundational blueprint then serves as the basis for all subsequent stages of the knowledge hub creation.
Documentation: The next step is program-level document generation. Agentic AI workflows can be used to extract data definitions, business rules, and control flows in a structured, step-by-step manner. Multiple agents can be deployed to collaboratively analyse, interpret, and consolidate these findings into a unified document, which can be further enriched using the application map created earlier. Enterprise context and existing legacy documentation can further enrich the document generated. Consolidated documentation at the process or subprocess level must be created, where required, to provide a broader business and operational view across multiple related programs.
Evaluation mechanisms should be built into the agentic workflows to continuously assess the accuracy, completeness, and quality of the generated outputs. Human validation and approval are a must for every document to ensure that the final output meets enterprise standards for reliability and usability. One way of achieving this is for SMEs to manually document a representative sample of programs or a subprocess and compare the output with the AI-generated document to identify gaps and differences. The workflow can then be finetuned to improve accuracy, completeness, and consistency.
Document upload to ECM: The next step is to upload the generated documents into the enterprise content management (ECM) system. Users can directly access the documentation through this platform.
Conversational access
Conversational access must be enabled to the legacy knowledge hub through enterprise AI platforms. BFSI firms can also build custom chatbots using agentic AI frameworks to provide conversational interfaces to retrieval-augmented generation (RAG) pipelines, enabling intuitive and context-aware access to mainframe system knowledge.
BFSI firms must begin with a focused pilot covering a small set of critical applications or subprocesses, combine GenAI-driven documentation with validation by SMEs, and use the learnings to refine the workflow. This creates a practical, scalable path to preserving core business knowledge, accelerating impact analysis and modernisation, and ultimately establishing a knowledge hub that makes mainframe knowledge accessible, reusable, and future-ready.
Case-in-point
A leading Nordic banking and financial services group embarked on a mainframe modernisation journey, incrementally moving legacy applications and workloads to a modern cloud-based tech stack. The firm wanted to build a comprehensive mainframe knowledge base to enable an in-depth understanding of legacy applications. Our approach combined AI tools and GenAI-powered analysis, followed by validation by SMEs, resulting in a reliable knowledge base that served as the basis for a cloud-native mainframe modernisation strategy. With this implementation, the firm realised some key benefits:
- Improved understanding of legacy system components and dependencies to accelerate modernisation.
- Uncovered business rules and application knowledge from legacy assets to create comprehensive documentation, reducing reliance on SMEs.
- Reduced migration cost and cycle time by 20-30% through improved conversion, testing, and validation efficiency.
- Reduced modernisation risk by preserving critical business knowledge and establishing a reusable documentation framework.
Making legacy knowledge outlive the mainframe
In the BFSI industry, the future will belong to firms that transform legacy knowledge into a lasting enterprise asset. Given that mainframe systems house core records and transactional processes, the knowledge created from the codebase will serve BFSI institutions throughout the lifetime of their mainframe systems and beyond.
In the next five to ten years, we expect BFSI firms to make significant progress in their legacy modernisation journey: initiatives will involve building a knowledge hub, revamping mainframe engineering, or exiting mainframes by shifting workloads to modern, cloud-native platforms. Through this path, the knowledge hub will serve as a living repository of legacy applications’ business functions, rules, and operational knowledge by staying aligned with the application landscape and systematically capturing changes as the system evolves.
Even after the legacy modernisation exercise is completed, the knowledge hub will offer valuable insights to support strategic decisions and preserve critical business knowledge for future technology change initiatives. Setting up the mainframe knowledge hub is therefore imperative and underscores the need for BFSI firms to act quickly to retain a competitive lead. However, this is only the first step; translating modernisation initiatives into business agility will need more action from BFSI firms.
Part 2 of this thought paper series will dwell on the use of AI in modernising mainframe engineering, while part 3 will deal with exiting mainframes.