Valor Holdings ran a controlled supermarket trial on prepared foods and other short-life items, where kitchens must commit to production before the day’s demand is clear.
Japan Weather Association and SoftBank worked with Valor and Chubu Foods from January 23 to March 12, 2026 across 20 Valor stores, with results compared with similar-sized stores in the same region.
SoftBank’s Sakimiru demand-forecasting service fused weather data, people-flow statistics and store performance. Those signals fed production and ordering recommendations, dynamic markdown timing, and a generative-AI store agent that explains why a given action is recommended.
Food waste is rarely a single lever: overproduction in the morning, slow reaction to weak afternoon traffic, or overly deep discounts can each damage a different metric. The trial was designed to move waste, labour, availability and margin together rather than optimise one in isolation.
According to IoTNEWS, prepared-food waste in participating stores was 12.6% lower than in the control group. Weekly time on daily production planning and ordering fell from about 127 minutes to 81 minutes (36.5%). Stockouts were 3.2% less frequent. Sales were 2.9% higher and gross profit 4.1% higher in trial stores, the same report said.
Those paired outcomes matter because waste can be cut simply by making less food and accepting empty shelves. Weather and local footfall are especially relevant here: rain, temperature and crowd movement can shift prepared-food demand within hours in ways weekday history misses—if the kitchen still has time to change what it produces.
The generative-AI agent does not issue the forecast itself. It is meant to put recommendations into natural language so staff can see why production or markdown guidance changed, which is often what determines whether a recommendation is followed.
The figures remain a company-partner account, and modelling detail is limited, but duration, store count and comparison locations are specified. Chubu Foods intends to roll the approach across Valor prepared-food departments after clarifying how work is split between people and the system—the test of whether the loop holds beyond the trial.
Grocers have used weather forecasts for years. The operational claim is a closed store cycle: how much to make, when to markdown, why the instruction changed, and waste, availability and profit measured together.