Large transformation programs promise everything and deliver late. AI makes the temptation worse: every workflow looks automatable, so the roadmap grows until nothing ships.
Start where the value is
Pick the one or two workflows where AI can save the most time or win the most revenue, and where you can measure the result. An audit or opportunity assessment usually finds them in a week or two.
Make every milestone stand alone
Each phase should produce something useful on its own: a working agent, a measurement baseline, a trained team. If the program stopped after phase one, you would still be better off. That is what makes it safe to start.
Measure before you scale
Agree on metrics before the build starts and track them from the first release. When phase one shows a result, the case for phase two writes itself, and you have the numbers to price it.
Compare both paths honestly
Sometimes a full overhaul is the right call: legacy systems that block everything, or a hard deadline. Price both paths side by side, phase by phase and all together, and decide with the trade-offs in view.