AI-Driven Software Engineering and Legacy Modernization in Banking

In banking, some of the earliest measurable value from generative AI is not appearing in the most visible customer experiences. It is emerging inside the technology function, where engineering teams are under pressure to modernize legacy systems, reduce delivery risk and move faster without compromising control.

That matters because many banks still run on decades of embedded business logic spread across core platforms, payments systems, lending engines, servicing applications, risk tools and regulatory workflows. These environments are often tightly interconnected, heavily customized and only partially documented. The result is familiar: modernization becomes slow, expensive and difficult to scale because teams must first understand what the legacy estate actually does before they can safely change it.

Rather than treating AI only as a front-end innovation layer, banks can apply it directly to software engineering and modernization work. In Deutsche Bank’s enterprise AI transformation, one of the first value-generating use cases was augmenting software code development with AI, including improving documentation and helping teams understand old code. That is a strong signal for the market. In large, legacy-heavy banking environments, AI can unlock progress where transformation programs often stall first: discovery, documentation, testing and delivery.

Why technology is a high-value starting point for AI in banking

Software delivery sits at the center of nearly every banking priority: faster product launches, better operational resilience, stronger compliance, improved servicing, lower cost-to-income pressure and readiness for future AI use cases. But legacy complexity slows all of it down.

Many critical banking systems contain years of undocumented rules covering interest calculations, fee logic, exceptions, posting sequences, reconciliations, controls and downstream dependencies. Some of that knowledge lives in code. Some lives in manual workarounds. Some lives only in the heads of experienced teams.

AI helps banks surface and structure that knowledge faster. It can analyze legacy codebases at scale, identify dependencies, expose hidden business logic and generate clearer technical and business-facing documentation. That reduces reliance on scarce institutional knowledge and gives engineering, product, risk and compliance teams a stronger shared view of how systems operate today.

This is why AI-driven software engineering can become one of the first places where GenAI moves beyond experimentation into measurable operational value. It improves the work required to modernize the bank itself.

Where banks can apply AI across the engineering lifecycle

Legacy code understanding and documentation. AI can help teams interpret older languages and complex monoliths, convert opaque logic into structured specifications and create reviewable artifacts that support modernization planning. In one example of AI-powered modernization, nearly three million lines of COBOL were converted into verified specifications in eight weeks, while more than 200 implementation-ready backlog items were generated to accelerate delivery planning.

Backlog and specification creation. Modernization programs often lose momentum between discovery and execution. AI can help bridge that gap by turning extracted business rules, field mappings and process flows into clearer requirements, backlog items and modernization designs. This creates stronger traceability from legacy behavior to future-state implementation.

Testing and validation. In banking, modernization fails if teams cannot prove that the new system preserves the right behavior across standard transactions, exceptions, historical variants and downstream integrations. AI can generate and execute broader test coverage, helping reduce manual effort while increasing confidence in releases.

Code transformation and modernization. AI-assisted delivery can help translate legacy functionality into modern architectures, including API-enabled services, modular components and cloud-native patterns. The goal is not change for its own sake. It is preserving critical business behavior while making systems easier to evolve.

Continuous engineering productivity. AI can also improve day-to-day software delivery by accelerating prototyping, coding, testing, deployment and maintenance. Publicis Sapient positions this as part of a broader engineering transformation, using AI not only to build new experiences faster but to modernize the technology foundation that supports them.

Modernization with lower risk, not just higher speed

In regulated banking environments, speed only matters if it comes with control. A bank cannot modernize like a digital native startup. It must maintain uptime, auditability, data integrity, operational resilience and compliance while continuing to run the business.

That is why AI-assisted modernization needs to be governed, traceable and aligned to enterprise architecture. It is not enough to generate code quickly. Teams must be able to explain what changed, validate what was preserved and provide documentation and testing evidence that risk and compliance stakeholders can trust.

This is the logic behind Publicis Sapient’s modernization approach. The focus is on applying AI upstream in the engineering process: understanding systems, generating specifications, accelerating transformation and strengthening validation. That makes modernization more consistent, more repeatable and more governable.

How enterprise AI strategy connects to core modernization

Banks do not scale AI successfully by layering it on top of fragmented systems. The foundations matter. Across Publicis Sapient’s financial services work, the recurring message is clear: scalable AI depends on modern data, strong governance and more modular technology estates.

In Deutsche Bank’s case, the AI agenda was supported by foundational investments in hybrid cloud, data architecture and the breakdown of silos across front, middle and back office functions. Publicis Sapient helped build and prove an AI and machine learning platform, together with use cases, operating models, adoption plans and a broader AI/ML capability catalog. That catalog included infrastructure, automated data quality work, model development and monitoring, governance and core solutions built with data science and natural language processing.

This foundation-first model is especially important for engineering transformation. Cloud-native platforms, containerized environments and modular architectures make it easier to modernize incrementally rather than through a high-risk big-bang replacement. They also create the conditions for AI-assisted delivery to scale across teams and business domains.

From legacy estate to AI-ready bank

The most important shift is strategic. AI in software engineering is not just a productivity tool. It is a modernization accelerator.

When applied well, it helps banks reduce dependency on scarce legacy expertise, improve engineering productivity, generate better documentation, strengthen testing and create a clearer path from old systems to modern platforms. It also supports a more practical transformation model: domain-by-domain modernization with measurable outcomes, rather than multi-year replacement programs that delay value until the end.

Publicis Sapient brings this together through an approach that connects strategy, product, experience, engineering, and data and AI. For banking leaders, that means AI is not separated from modernization. It is part of the way modernization gets delivered.

The result is a bank that becomes more responsive, more scalable and better prepared for what comes next. Modern core platforms, modular services, governed AI workflows and AI-assisted delivery do more than improve the software development lifecycle. They create the technical and operational foundation for the next generation of banking transformation.

In an industry where legacy complexity often blocks innovation, that may be where some of the most important early AI value appears first.