Trace the work as it is performed. Not as it is written down. Every handoff, every re-keying, every wait state, every exception path and who absorbs it. 1 Draw the line. Decide which steps should be handled by software, which by a model, which by a person — and which should simply stop being done. 2 Build the connection. Working software over the systems you already run. No migration, no replacement programme, no new platform to adopt first. 3 Produce the evidence. Measured behaviour against real cases, with failures named and kept, so trust is earned from data rather than demonstration. 4 Hand it over. Transfer the system and the reasoning behind it to your team, and leave.