How agentic AI is changing fashion merchandising

Inventory is where merchandising decisions meet reality, which makes it a practical place to start. Here is how we approached it for a US apparel brand: 

The problem – As noted in the introduction, the retailer, like most apparel brands, decides how much stock to allocate to each store at buying time, months in advance. Once the season starts, actual demand diverges from the plan. A size sells out in one store while the same SKU sits on the shelf in another. Because these imbalances weren’t identified during the season, they surfaced later as markdowns or returns to vendors, putting margins under pressure. 

What we built – Our inventory leveling agent has three parts. The first is a two-stage machine learning forecast, trained on about 30 months of sales history alongside weather, public holidays and the promotions calendar, which estimates demand for each SKU at each store. The second is a transparent planner that identifies stores with stock shortages, stores holding surplus inventory, and potential donor stores that can help meet demand. The third is a conversational agent that lets planners ask questions in plain English, such as which stores are likely to run out of stock this week. 

Each weekly run produces three outputs for three teams. Transfers go to store operations to ship. Exceptions, where nobody holds the stock, go to buying to reorder. Excess stock that nobody wants goes to merchandising for a markdown decision. 

Where the AI stops, by design – The model forecasts demand, but it does not make the final inventory allocation decisions. Which stock moves, how much moves, and where it moves from are determined by transparent rules that a planner can check manually. Each threshold can be adjusted if the planner disagrees with the recommendation. 

Here’s an instance in which we saw this agent in action: A store had run out of a popular shirt in size L and had been selling it steadily. A store 20 miles away held six units and had sold two in 12 weeks, so the agent recommends moving four across. It passed over a closer store, because that store sells the same shirt five times faster and needs its own stock. 

Checking the recommendations – We tested the recommendations against 12 weeks of the brand’s real sales history. In 91% of recommended transfers, stock was moved to a store selling that SKU faster. The figure is not 100% by design, because a slower store sometimes needs topping up when it has run out entirely. In 40% of cases, the donor store had sold none of that SKU in 12 weeks, so dormant stock was being moved to where it sells.