AI Could Breathe New Life Into Core Systems

The new kid on the block, AI, meets the original gangster, COBOL. Is it magic or is it a miss? COBOL, the Common Business-Oriented Language, is still very much an integral part of how data is stored and used by some of the institutions we trust deeply such as banks, insurers, healthcare providers, and governments. One would think that moving away from COBOL in our hyper connected and technologically advanced world would be easy. It is the opposite. And AI, while it cannot or should not prevent the end of COBOL, might offer a way forward. 

The risks of attempting to upgrade core systems at these institutions are extremely high. Risk departments at financial institutions may mandate, for instance, that core technological changes at multinational financial institutions may occur within only one country at a time per year. For the major multinationals, a lifetime or several will be required for meaningful change. If we randomly choose a large global bank, we have an idea of what that implies. Take Citibank, which has presence in 95 countries according to its website. If it were to espouse the one-core-change per year risk control as part of its risk framework, it would mean a 95 year project.  

Let's expand the example a little bit to gain some more context. Assume that a core database holding protected and highly sensitive personal information about clients and their holdings may have to relate parts or all of its data to five other applications.  These applications help with account management, managing the loan book, compliance, anti-money laundering, fraud detection, customer relationship management, marketing, and more. In each instance, the database at each country branch would need to be mapped to each of those applications, compounding the problem.  

Making the wrong choice on programming languages, on system architecture, or on what technological stack to use, among other things, has astronomical cost implications. And this doesn't account for the considerable risks, such as outages due to a bad migration to the new system or the new system being corrupted upon release, forcing the institution to move to backups.  

COBOL-based systems are still very much the core of the computing backbone in several industries. If you're simply looking at payment messages in the financial services sector, only 15% of financial messages have moved away from COBOL to the ISO 20022 standard. SWIFT aims to have banks in the world on the ISO 20022 standard and the market is now in an interoperability period. This migration, however, only impacts the streams of data, and not the stores. The stores are likely going to continue to be COBOL in many organizations.  

COBOL is highly reliable and efficient, historically making it the top choice for application in finance, government, and large enterprise. Whether it is processing batches of data, processing files, generating reports, validating data, or performing database operations, COBOL was the go-to for every institution that needed to process and manage large amounts of data.  

Times change. Newer companies have had more choices and often chose to use other languages to manage database operations, validate data, or generate reports. As the incumbents expanded and started to either acquire these new and smaller players or built new systems to enhance their value propositions, they were left with the lingering dilemma of what to do about their legacy systems.  

AI is now being tested by IBM to assist with the translation of COBOL into Java, a highly extensible programming language. Among other benefits, Java is platform independent, has a large developer community around it, can be used for enterprise applications, and is fit for both mobile and web applications.  

AI can help translate existing COBOL code to Java but not yet in an optimal way. This is because human programmers cannot rely on a consistent output by the AI programs. This is not, however, a signal of defeat. It just means that the possibilities are vast and that in the coming years, we can anticipate that more computing companies will take on the challenge of accurate and usable code translation.  

The bigger question—and opportunity—is whether these translations can help keep legacy systems intact, keeping big organizations competitive through the transition to new systems. Skills and resources remain an issue for organizations and these form two huge areas for why AI may just be so attractive – to free up resources to work on higher level challenges related to the transitions.  

Thinking beyond the COBOL example, if the multiple hard-coded translation layers between each database and the applications they need to talk to could be replaced with one intelligent layer that could learn to talk to each application, that would be an enterprise panacea. Yesterday, it may have seemed far-fetched. Today’s rapid emergence of AI suggests this is within reach. This new reality signals exponential value creation both by the employability of old systems in new commercial approaches, and by the potential significant reduction in cost and risk of the redesign and rebuilding of old applications.  

---- 

This brief article is not meant to delve deep into the technical aspects of computer programming. Nor is it a report on the prevalence of programming languages used in the world of business or otherwise. It is meant to provoke thought about the longstanding issue of legacy systems used by businesses, the cost, and the risks involved in trying to replace them, and the emergence of what may pose potential solutions.  

Antonio De Lorenzo has spent his career focused in technology and finance, specifically in capital markets and payments. His experience and interests meet at the intersection of fintech, web 3, policy, products, ventures, and people. In addition to his current work, he advises entrepreneurs and serves on the task force for the future of finance on the Bretton Woods Committee. He studied finance and computer science at Virginia Commonwealth University and business administration at INSEAD.