Answer
A useful way to run this assessment is to interview the people who touch the system every day, not just the IT team that maintains it. Ask where they lose time working around the interface, where a workflow requires manual re-keying between screens, and where a report or lookup takes minutes instead of seconds. Those friction points translate into measurable numbers: help desk ticket volume, new-hire training time on 5250 green-screen menus, or the number of manual steps between an order landing and inventory being adjusted. Bottlenecks that show up in more than one of those measures are usually the right phase-one target.
Sequencing matters as much as selection. Touching the interface, the workflow engine, and the underlying RPG or COBOL code in the same phase makes it hard to isolate what actually fixed the problem, and it multiplies testing effort on a platform where regression testing against Db2 for i data integrity has to be taken seriously. A tighter approach picks one layer, defines what success looks like in business terms before the work starts, and holds off on the next layer until that phase is proven in production. When evaluating a partner or platform for this first step, ask them to name a comparable phase-one project they scoped narrowly, and what they would have done differently.