The situation
A global apparel brand made its in-season trading decisions the way most retailers do: per domain. Demand forecasting in one team, inventory management in another, close-out and markdown in a third, digital personalisation in a fourth. Each domain had its own models, its own data pipelines and its own view of the season. The decisions were connected in reality — what you forecast determines what you buy, what you fail to sell determines what you mark down, what you recommend determines what sells — but nothing in the organisation or the architecture connected them.
What was built
A regional AI function of a handful of cross-functional squads — data scientists, machine learning engineers, product — with ownership of in-season decision making as one connected whole: demand forecasting, inventory optimisation, close-out optimisation and personalisation, tied directly to the brand’s marketplace and digital growth strategy rather than to any single function’s roadmap.
The operating model made the connection explicit. Value was estimated per decision, upfront, and the portfolio was steered on where the P&L moves in season: buying, allocation, markdown. In parallel, the legacy regional AI products were migrated onto a standardised global cloud ML platform, so that models built for one market could be reused rather than rebuilt.
The result
A significant EBIT uplift, read on the lines where retail earns its margin: reduced markdowns, fewer stockouts from better in-season forecasts, improved close-out decisions, and personalisation that feeds demand signal back into the same loop.
Why it is here
This case is the origin of a conviction that runs through the Ardinois approach: inventory, pricing and personalisation are one decision, and AI only reaches the P&L when the operating model — ownership, funding, governance — is designed around the connected decision rather than around the individual model.