The distinction between a demo and a pilot is the part that stuck with me. A demo proves the platform can work under controlled conditions; a pilot has to prove it matters for a real problem, with measures agreed before any work begins. Starting from month-end reporting is a good example — everyone in the business already understands the pain, and the result is easy to judge.

I also liked that stop is treated as a valid outcome, not a failure. Too many pilots drift on because nobody agreed in advance what good looks like. Tracking business outcomes, data quality, governance, adoption, capacity and cost together feels like the right discipline — cost on its own tells you very little, and a tool people don't actually use isn't a success no matter how neat the demo was.