Inventory optimisation
Your group is the delivery lead for this funded initiative.
The mandate
The board has approved $1.1M (the largest of the four) to build AI inventory optimisation: forecasting demand and automating replenishment across 50 stores plus the online channel, to cut both stockouts and overstock.
What the board thinks it bought: “major efficiency gains across the whole network.”
Your job: deliver it without the predictable failures, and define a “good enough” forecast that’s safe to act on automatically.
Why this one is hard to deliver
- It is the broadest in scope (every store, every category), so the temptation to boil the ocean is strong, and the blast radius of a bad forecast is large.
- It runs straight into legacy systems: the POS and inventory systems are 10–15 years old, and online/store data is siloed. Integration is where this project lives or dies.
- A forecast that’s “about right” in aggregate (the demo) can be badly wrong for a specific store/category, and that’s where the cost and the staff frustration show up.
The data reality (ask Priya)
Priya’s honest read: “significant value potential, but data quality issues to resolve first; integration complexity is underestimated. 9–12 months. Start with ONE category pilot, prove value, then scale.” Her advice basically pre-writes your roadmap, if you ask her.
Who to interview, and the tension they bring
- Priya Sharma (Data): pushes the single-category pilot; data-quality realism.
- Marcus Kim (CIO): knows the legacy-systems pain better than anyone; also impatient for transformation.
- Sarah Thompson (COO): operational reality across stores; the people who’ll act on (or ignore) the forecasts. She slows things down, sometimes rightly.
Your starting question for “good enough”
What forecast accuracy, in which category, justifies automating replenishment versus keeping it as a recommendation a store manager approves? How do you scope a pilot narrow enough to prove value but real enough to matter?