Four notes on how this work is actually done. No case studies and no client names, because the method is the part that travels between industries.
Most corporate AI dies between the demo and the operating cycle. The cause is almost never the model. Four conditions decide which side of that line a project lands on.
A control group turns a claim into evidence. What to compare, how to read the gap between the two lines, and the four ways the comparison quietly breaks before anyone notices.
Warehouse, master data, automated planning. The unglamorous sequence that decides whether the interesting part is even possible, and the one test that tells you where you stand.
Four results from a fourteen-course AI program. Accuracy that hid 92% of the events that mattered, an average that hid a whole class, and a simple baseline that almost won.