Analytics and AI leadership
I build AI agents that make operating decisions and carry them out, on data infrastructure solid enough to trust them with the call. Today they run a six-country retail operation. The same problem waits in every industry with a digital transformation to deliver.
The first ones grouped hundreds of locations by the shape of their demand and pushed the answer straight into execution. Work that used to be a monthly spreadsheet now runs on its own, and every decision it makes is logged.
Most of the gain came from moving assets that were already paid for to where they earn. Any business with inventory, capacity or people sitting in the wrong place has this same problem, whatever it sells.
Agents are only as good as what they read. A cloud warehouse, governed master data and automated planning came first. That order is not optional, and it is where most AI programmes quietly fail.
These agents cleared the pilot stage and now run inside the weekly operating cycle, with named owners on each decision. Their output lands in the same numbers the business already reports every month. Most AI work I have watched die, died before this point.
Saying an AI project worked is easy. I compare the sites that got the change against similar ones deliberately left alone, so the gap between the two lines is the part that holds up when a board asks how we know.
Difference-in-differences, reconciled against the warehouse before anything leaves the team.
Three notes on how this work is actually done. No case studies and no client names, because the method is the part that travels.
I run analytics across six countries and sit on the executive committee, reporting to the board each quarter. Planning, allocation, S&OP, data governance and the agent work above all report here.
I arrived as a BI specialist and left the role as manager. In between: the cloud warehouse the company now runs on, reporting that stopped being manual, and delivery times cut from two and a half months to one.
Nine years on flight performance and fuel. I automated analysis that took twelve hours down to one, and found $700K a year in savings validating how fuel was actually used. Different industry, same method.