The Opportunity
- A railroad transports crude oil from refineries to oil customers. They need to understand demand on their network so they can provision the right number of tank cars for each origin-destination pair.
- They wanted to see if crude oil prices could help them better predict demand.
Our Approach
- We joined data on the number of the company’s tank car loadings with pricing data for four types of crude oil (e.g. Brent, West Texas Intermediate).
- We used time series models to uncover the lag structure between the variation in prices and the impact on demand.
- We used cluster analysis (unsupervised learning) to segment customers in terms of their response to changes in demand.
The Impact
- Our models found that demand pattern change customer and shipping destination more as a result of the spread between the four crude oil types, vs changes in the absolute level of oil generally.
- Our segmentation found four distinct time lag patterns in the way car loadings relate to changes in the spread of prices. The patterns can be classified by their immediacy (sooner rather than later) and their magnitude.
- These models helped the railroad anticipate demand at the customer-destination level, based on evolving prices.