The oil industry spends billions of dollars annually on the distribution of petroleum products. It is estimated that these costs alone add 4 cents to the cost of a gallon of gasoline in the U. S. With an annual volume in 1995 of 74 billion gallons, that segment alone accounts for close to $3 billion annually. Dispatching "oil" products includes determining sources, destinations, timing, composition, and size of shipments, assigning transportation modes, and finally assigning individual transportation units. Dispatching "petroleum" products involves considering transportation and product-source costs, operating rules, inventory considerations, customer service policies, and other factors.

Set partitioning-based optimization models have emerged as the most viable optimization approach that can adequately address the complexities of these transportation dispatching problems. However, set partitioning models have been hard to solve optimally for the problem sizes typically encountered in daily operations. The rapid development of computing power, combined with the development of ever more efficient algorithms, has facilitated the recent emergence of set-partitioning-based optimization models in transportation dispatching systems.

A dispatching system consists of models and algorithms, together with user interfaces for the input and output sides, as well as databases and data management routines. Dispatching is often the act of balancing multiple objectives. The primary objective is usually minimizing delivery cost. Additional objectives may include balancing the workload among transportation units, minimizing the number of violations of service guidelines, and determining the proper quantity to ship at any given time. However, even the minimization of total cost is not always a simple objective to meet.

An effective dispatching system should account for differences among available transportation units (size, cost, type of equipment, etc.), because a company rarely uses a uniform fleet. Due to the numerous considerations involved in dispatching, some of which are unforeseen and therefore cannot be programmed into the dispatching system, the system should be designed to support, not replace, the human dispatcher. Manual overrides must be built into an effective dispatching system, and the dispatching should be able to see the cost impact of any overrides that are made. This capability is probably the greatest contribution of computerized dispatching systems.

Significant progress was made in the 1990s in dispatching products in the oil industry. In some cases, mathematically optimal dispatches are now a reality, while in others only good solutions, with known bounds on just how good, are possible. The next step in development will be integrated solutions to dispatching problems that involve both inventory and transportation. The rapid development of computing power, together with the continued development of efficient algorithms, is expected to lead to the solution of such problems in the not too distant future.