In an example embodiment, a series of machine learned models are trained and utilized in conjunction with each other to improve the reliability of predictions of fuel costs. One of these models is specifically trained to learn the “gap” time for a particular retail location, meaning the amount of time between when the futures contract market price on a trading exchange making up the fuel blend has the most correlation with the retail price of that fuel blend (for that particular location). This greatly enhances the reliability of the predictions of fuel costs, and, as described in detail herein, these predictions may be used in a number of different applications in unique ways.