AI-Assisted System Discovery: Opening the Black Box
The operations of many large companies (banks, insurance companies, telecommunications providers) are still built on legacy systems that have been reliably doing their job for 10 to 20 years. These systems often support critical business processes, and over the years, a great deal of business logic has been built into them. This is precisely why it is difficult to change them.
At the same time, certain familiar issues emerge over time:
- documentation is incomplete or outdated
- the details of how the system works reside in the minds of a few key individuals
- further development of the system requires a technological rethink
- the system becomes a “black box” that is difficult to understand
- every change is risky and costly
This is when the idea of replacing the system arises. But whether the company is considering an off-the-shelf solution or a greenfield development, modernization is no longer just a technological issue but also a task of business risk management. The problem, however, is often that the organization lacks consistent, detailed documentation of the system’s functionality.
The Classic Approach: Slow and Difficult to Scale
The traditional reverse engineering toolkit is well-known:
- manual code analysis
- log analysis
- debugging
- disassembler tools
In addition to code-based analysis, it is common practice to collect historical documentation—and users of the system attempt to document the knowledge about the system that currently exists only in their own heads.
This method works—but it is typically time-consuming, heavily reliant on senior experts (whose time is expensive), and, moreover, difficult to scale. Compiling consistent, comprehensive documentation from multiple sources is a lengthy task.
Where Can AI Help?
It is important to clarify that AI does not replace reverse engineering—it “merely” automates and structures it. Our AI-based approach, AI-assisted system discovery, aids in understanding the system on multiple levels:
Structural analysis
- dependency graph
- call graph
- data flow analysis
Semantic analysis
- functional interpretation of code
- recognition of business logic
- identification of function purposes
Higher-level abstraction
- generation of architectural models
- breaking the system down into typical modules
- generating UML and sequence diagrams
- reverse engineering domain models
The result: the operation of a complex system can be understood more quickly and in a more structured way.
A Real-Life Example: Mapping a Corporate Lending System with AI-Assisted System Discovery
A monolithic system supporting the corporate lending processes of one of our banking clients has been operating stably for a long time. However, three strategic issues have emerged:
- significant vendor lock-in
- due to its increasingly apparent limitations, the technology needs to be replaced in the long term
- and uncertainty regarding whether the system’s individual functions are in the right place.
The idea of replacing the system has therefore arisen. However, the goal is not to replace the system immediately, but rather to conduct structured, modular mapping as a first step.
The essence of the task is to “deconstruct” the monolithic system and understand the business processes that actually drive the technology. These include, for example:
- transaction processing
- risk decision support
- collateral management
- product catalog and product selection
- loan administration
- document management
- workflow
- reporting
These modules are often not formally separated in the code, but they are easily identifiable functionally.
Our approach consists of four steps:
- AI-assisted analysis of documentation and source code
- Identification of implicit module boundaries
- Consolidation and structuring of functions
- Validation by banking domain experts
The result will be a business-interpretable, modular function and model map that provides the bank with a basis for comparison with solutions available on the market.
This results in
- transparency regarding the system’s actual functionality
- reduction of the “black box” nature
- and a stable foundation for subsequent GAP analysis and system selection.
During the GAP analysis, it will be possible to clearly determine
- which modules are covered in a standard manner by an existing or commercially available system,
- in which areas customization is necessary,
- and for which functions it is worth considering from a business perspective whether a change is necessary at all.
Two Key Takeaways
AI Alone Is Not Enough
Our experience shows that AI works best when three factors come together:
- appropriate input data (source code, logs, documentation—if available),
- well-chosen technology (e.g., to minimize common LLM limitations such as hallucinations and overgeneralization), and
- (human) expert validation.
Without a business context, AI only sees patterns—experts provide the real meaning.
The First Step in Modernization Is Often Not Development
Many organizations immediately look for a new system. Yet the first and most important question is often this: “What exactly does our current system do?”
If there is an answer to this, then:
- a realistic modernization strategy can be developed,
- solutions available on the market can be compared,
- and an objective decision can be made about what truly needs to be retained.
AI-assisted system discovery helps with this: making transparent what was previously a black box.
If your organization has also considered replacing your tried-and-true legacy system but would like to understand exactly what it’s capable of, let’s talk!
