Thousands of modifications, zero documentation — how AI-assisted analysis accelerates ERP migration
ERP systems used for years accumulate thousands of modifications without documentation, making migration a manual review of each one. We developed a methodology for extracting business logic using AI tools, aiming to cut migration from over a year to a few months.
A system nobody can describe
The client's ERP system had been modified for over a dozen years. Every implementer, every maintenance person, every support engineer added something. Not one modification was documented — the system itself was the documentation. When the decision was made to move to a new version, the scale of the problem became visible: each modification had to be analyzed manually by a person who understood both the code and the client's business context, then a decision had to be made whether to transfer it, simplify it, or retire it. This is not a one-off edge case — it's the standard state of long-running production systems whose documentation debt is invisible until the moment of change.
What we didBusiness logic extraction and phased migration
- Conducted a technical analysis of environment constraints — how code is stored, the version control model, and access restrictions — to determine what scope of work can be done with AI tool involvement and what requires a classic approach.
- Designed a method for extracting business logic from legacy code — the AI tool identifies modification areas and generates a description of what a given code fragment does and whether it's still in use.
- The output includes an inventory divided into used, duplicated, and inactive items, along with a specification of what requires client verification.
- Split the migration into four stages — inventory, specification, implementation, and stabilization — with each stage producing a deliverable that has standalone value regardless of whether the client continues.
- Introduced a layer of compliance requirements: enterprise-only models, EU-based data processing, and a strict rule that no client data is used for model training.
- Proposed a free preliminary analysis in exchange for access to code and data structures — this enables estimation based on actual scale rather than client declarations.
What changed
The client receives a modification inventory divided into used, duplicated, and inactive — instead of moving everything to the new environment. System knowledge is recorded in durable form, independent of individual consultant availability. The preliminary analysis provides a basis for estimation grounded in actual scale, not guesswork. The methodology's goal: shorten migration from over a year to a few months, with a significant cost reduction.
“How the organization works was recorded — but in code, by people who no longer work there, and without explaining why. The cost of missing documentation doesn't spread evenly over time. It stays invisible for years, then emerges in full the moment the system needs replacing.”
Let's write down how your company actually works.
- 45 minutes, and no preparation needed
- An honest read on where you sit today
- The first process worth automating — and why that one
