Production AI work, methodology walkthroughs, and applied experiments. Every case shows the problem, the approach, and the business impact.
Most retail marketing budgets are flat per customer. But customer value isn't flat — it's a power law. Without segmentation, you overpay the cheap customers and underspend on the ones who actually carry the P&L.
Last-mile is the single most expensive line in a delivery P&L. Naive dispatch order — visit stops as parcels arrive — quietly burns 50–70% more kilometres than the optimised version. The math has existed since the 1950s; most ops teams still don't run it.
Most grocers lay out shelves by supplier or alphabetically — not by what shoppers actually buy together. The lift the floor leaves on the table is invisible until someone runs the data.
Most supply-chain networks are sized by gut feel and Excel. The result is networks that are always 'a little too big in the wrong place'. Linear programming says exactly where the flow should go — and what it actually costs to respect every capacity constraint.
Retail DCs running continuous review at scale drown in daily replenishment lines. The instinct is to switch to periodic review and expect a big drop in order volume. That's the wrong number to chase.
Most teams jump from raw sales data straight to a time-series model and lose the question that actually matters: what's driving demand, and how much of it can a model learn at all?