Demand forecasting of fule sales on an SKU level per gas station
Uno-X operates a chain of unmanned fueling stations in Denmark and Norway. Tank stations have one tank for each fuel product they sell and the pumps connect to the tank, this means they need to operate a set of fuel trucks that deliver fuel to each station.
The challenge is to figure out how often to deliver fuel as it takes hours to do the delivery. If too often and one wastes money on unnecessary deliveries too seldom and one risks running out of fuel to sell, resulting in a loss of revenue.
To better be able to plan deliveries they wanted a set of models that predicts when the tanks will run dry such that they can plan the fuel deliveries to be as efficient as possible, without risking a dry run. So we built a set of models that would predict the sales of each product for each station every hour for the coming week then by combining this with the current amount of fuel in the tank we can predict when the tanks will run dry. Since we built the models with a Bayesian framework, uncertainty is automatically included in the predictions. This allows for even better decision-making in terms of the risk of actually running dry.
Minimizing the risk of lost revenue, customer churn, and negative reputation.