Long the exclusive domain of AI (artificial intelligence) research, statistical probability analysis got its real start during World War II, when military intelligence wanted a way to predict possible outcomes. Throughout the 1950s, the field was continued by mathematicians such as John Nash, and now this complex discipline is becoming the stuff of ordinary business applications.
From a common-sense point of view, the way statistical analysis works seems illogical. The more complex a data set becomes, for instance, the easier it is to make predictions.
For example, take NLU (natural language understanding) research. If I begin with only the word “the,” a computer program would have far less than a one per cent chance of predicting the remainder of the sentence. However, if I add the word “day” to follow “the,” making the sentence more complex, the likelihood of guessing the third word might be 50 per cent or better. The word “day” is singular and “the day” will require a verb.
NLU program designers input millions of sentences into their software. So, statistically, they know that there might be a 90 per cent chance that the next word in my sentence is going to be “is.”
In similar fashion, supply-chain management companies such as G-Log have begun using a form of statistical analysis called stochastic optimization, first as a guidance tool for managers and later to actually automate the decision-making process.
In supply-chain management, most companies use constraint-based systems, in which every constraint is based on transportation rules. For example, almost all transportation management systems require inputting of date and time windows.
The difference here is that, using stochastic optimization, applications can consider the possibility of change. Managers can think outside the old box, rather than being constrained by artificial rules.
This will have its most dramatic impact on inventory costs. These include the cost of the physical location, property taxes, worker compensation, insurance, as well as transportation costs, the cost of taking goods off the truck or rail car, and of putting them in a facility.
The longer an item sits, the longer it ties up funds that might be used for better things. If goods never stopped moving until they got to their ultimate destination, the carrying costs would be reduced by an order of magnitude.
Stochastic optimization will help create a supply chain that never rests.
Here’s a small example. Say I need 20,000 pounds on my trailer to meet my contract and sitting in the warehouse today are 10,000 pounds. I also need to get this shipment out as soon as possible. Stochastic optimization can determine that there is a statistical 80 per cent chance that another shipment of 10,000 pounds is likely to come in at the end of the week.
The G-Log system, GC3, presents these options to the manager. He or she can make some calls and see if any of the likely shippers has the goods ready and can ship early.
In the second phase, the program will automate these opportunities. It will not wait for the manager to make the calls, but will instead send the alerts to the other shippers directly, asking if they can send their goods early.
As mathematicians always knew, mathematics — a field with so many constraints of its own — surprisingly leads to a great deal of creativity.
