For years, banks have cited regulations as the reason for their inability to operate faster, even as the industry sinks trillions of dollars into modernization investment. That certainly is one factor, but it is not the full picture. The more pressing issue is the sheer amount of legacy data and tech that most banks sit on.
Many banks are still running decades-old platforms and systems that very few people, if any, are able to understand and operate. The core logic of these systems is often undocumented, and that knowledge gap is a serious impediment to modernization efforts across banks.
Without recovering that lost knowledge, banks cannot move forward with modernization strategies that are scalable and reliable. Ripping and replacing tools and systems in the bid to digitize and accelerate processes is not a feasible approach when key institutional knowledge has left the organization as engineers retire.
Rather, banks must decipher their own legacy systems. Understanding the hidden institutional knowledge that ties the infrastructure together must be the first undertaking, one that AI can help with. Here’s how.
Why Lost Knowledge is the Real Bottleneck
The record-keeping of legacy code is at the root of the bottleneck. Systems designed years, if not decades, ago were often implemented by specialized expertise, but their knowledge was, more often than not, undocumented. And where documented knowledge does exist, it is out of sync with many banks’ current infrastructures. That makes it at best unreliable.
A significant chunk of banks still use antiquated code like COBOL. To put that into perspective, a good portion of the engineers who are able to understand COBOL are no longer active in the workforce or are very close to retiring. Banks have a slim window of opportunity to capture that institutional knowledge before it walks out the door—if it hasn’t already.
Unfortunately, translating the code to more up-to-date systems does not solve the problem. Pursuing this course of action means that banks end up transferring the lost knowledge without gaining the clarity behind the hidden logic of the code. Essentially, teams do not gain real familiarity or comprehension of the system and code they must still work with.
The Case for AI that Reads Instead of Writes
Here’s where AI comes in. Most narratives around AI algorithms have honed in on how quickly these tools can build things. In this case, AI is best positioned to help banks decipher their own systems.
The depth and scale of banks’ data ecosystems render a manual approach impossible. It’s like chipping away at Mount Everest with a pickaxe: decades-old platforms come with countless layers of historical detail. This is on top of the fact that the documented processes and workflow logic are packaged in code that most people no longer understand.
AI can move through that record at a scale no human team can match. The technology is able to distinguish layers of business rules and infrastructure to expose what code actually does. Just as importantly, it does this in such a way that teams can actually understand and articulate.
Additionally, AI flags undiscovered dependencies in old code that otherwise create modernization failures. AI can analyze call graphs and dependency graphs across an entire codebase faster and more completely than any manual audit. This means that what’s hidden is now visible, empowering banks and teams to operate and modernize more proactively.
Prioritize Orchestration
Banks do not want to shoulder the maintenance of these massive, time-consuming, and expensive legacy systems. The end goal should not be to own code, but to orchestrate old and new systems and architectures. A well-orchestrated architecture facilitates incremental updates and replacements that do not carry the same risks tied to a full system migration.
Because AI gives a clear view of what should be replaced, kept, retired, and updated, this significantly helps with orchestrating entire systems. It helps banks slash the maintenance headaches and pinpoint what can be outsourced to external vendors—without compromising on security and privacy.
AI-powered system interpretations yield the full picture on what each legacy component actually does. That knowledge in itself is power, providing the strong, informed foundation to move forward with orchestration decisions confidently. Cumbersome, complex legacy systems are no longer a barrier to integrating modern external platforms and systems because they are now clearly laid out through an AI-assisted knowledge recovery model.
Be Aware of Governance and Compliance
In fact, banks stand to win on the regulatory front if they are able to explain their core processes. When teams migrate code without understanding it, there is no reliable way to verify that the regulatory intent has been preserved in the new implementation. The opaqueness of an uninterrogated system leaves compliance teams to potentially sign off on risk they cannot actually anticipate or quantify.
This is a crucial consideration from a governance perspective because business rules embedded in code, no matter how old, often include regulatory obligations such as reporting logic and surrounding payment infrastructure. Even a small change that seems isolated and insignificant can carry serious cascading consequences in a rushed migration with no clarity.
And modernizing fragile, highly complex, dense systems without understanding their possible failures and risks only exacerbates that fragility. AI brings to the surface those failures before a new architecture inherits them.
When leveraged correctly, AI tools can create auditable trails that become a governance asset. There are demonstrable, documented rationales attached to every downstream modernization decision that can be traced back to what the AI systems brought to light. That auditability and transparency potential strengthen banks’ ability to modernize safely and at scale.
Capturing lost knowledge and logic is becoming increasingly urgent. Banks that treat AI as a code generator will miss out on the opportunity to modernize successfully because of the hidden knowledge that will soon be lost to them. Treating AI as a reader, not only a code writer, is key to resolving this issue and keeping pace with increasingly demanding regulations.
