Hershey deployed AI tools to optimize shelf placement and inventory for s'mores ingredients — its own chocolate bars, plus partner marshmallows and graham crackers — during peak outdoor season, according to Modern Retail. The company is using machine learning models that analyze weather forecasts, search trends, and point-of-sale data to predict when and where consumers will buy s'mores kits, then coordinating cross-brand merchandising in those stores before demand spikes.

The mechanic: Hershey's AI stack ingests regional temperature data, Google search volume for "s'mores" and "camping," and prior-year sales velocity by store. The model outputs a demand forecast by location and week, then Hershey's retail team uses that forecast to negotiate co-merchandising with marshmallow and cracker brands. When the model flags a 10-day warm stretch in the Pacific Northwest in May, Hershey pushes retailers to build s'mores endcaps in Seattle-area stores two weeks ahead, bundling Hershey bars with Jet-Puffed marshmallows and Honey Maid crackers. The result: the full kit is on-shelf when search intent and weather align, reducing stock-outs and capturing incremental basket lift from consumers who buy all three at once instead of one item per trip.

This works because s'mores purchases are event-triggered, not habitual. Consumers do not buy graham crackers weekly; they buy them when they plan a fire. Weather and search are leading indicators that a consumer is moving from intent to store visit. By placing the full kit in a single fixture during that narrow window, Hershey converts a single-item purchase into a three-SKU basket and increases the likelihood the retailer restocks all three categories together. The AI layer compresses the decision cycle: instead of waiting for sales data to show a trend, Hershey predicts it and merchandises proactively.

The steal for a small physical-product brand: identify a complementary product your customer buys *with* yours in a specific use case, then approach the retailer with a co-merchandising proposal anchored in external trigger data. If you sell hot sauce, partner with a tortilla chip brand and pitch grocers on a "taco night" endcap during weeks when "taco Tuesday" search volume spikes in that metro (track via Google Trends, free). If you sell candles, bundle with matchbooks and propose a "power outage kit" shelf during hurricane season in Gulf states, using NOAA storm forecasts as the trigger. The cost is your time: pull the trigger data yourself, build a one-page PDF showing the search or weather pattern, and email the category buyer with the proposed fixture layout and both SKUs. Most small brands can negotiate a four-week endcap test in a regional chain by doing the buyer's merchandising homework and reducing their planning friction.

The broader pattern: the winning brands treat the *occasion* as the product, not the SKU. Hershey is not selling chocolate; it is selling "s'mores night." The AI predicts when that night will happen, and the merchandising makes it easy to buy everything at once. A small brand without machine learning can still win by manually tracking the same signals — weather, search, local events — and pitching the retailer on the bundled occasion before the spike. The brand that arrives with the plan gets the endcap.