Start with the operation, not the codebase myth
Logistics systems encode years of exceptions: cutoffs, customer rules, warehouse quirks, and handoffs between TMS, WMS, and finance. A clean rewrite that ignores those exceptions ships elegant software operators cannot run.
Map the workflows that lose money when they fail — dispatch, exception handling, document packs, POD, billing — and treat those as the modernization boundary.
AI-assisted characterization before change
Use AI to help document modules, dependencies, and tests. Require humans to accept characterization suites that capture today’s behavior — including the awkward cases planners already know.
Only then refactor or wrap modules behind stable interfaces. Each release should leave the estate working.
Incremental cutovers that operators survive
Move one workflow at a time. Keep rollback paths. Measure exception rates and cycle time after each cut. Expand only when evidence supports it.
If AI capabilities are the destination, modernized contracts and data access should make retrieval, agents, and evaluation easier — not harder.