Saturday, May 14, 2005

Optimization In Supply Chain Management part 1

One area that has always fascinated me is the pace of adoption of optimization in supply chain management. As I had written in a previous article, compared to other industries like airlines and oil and gas trading, the creation of optimization based solutions for supply chain management has lagged. The adoption of these sophisticated tools by manufacturing companies have been even slower. Even leading edge manufacturing companies run their supply chains with planning tools which are very simple heuristics with obvious shortcomings. Companies seem to be happier with understandable solutions (read simple heuristics) rather than sophisticated algorithms. Why is adoption of advanced optimization to manage supply chain lagging other industries?
Who best to pose this question than to Narayan Venkatasubramanyan. Narayan has been a pioneer and a leader in the application of cutting edge optimization techniques to supply chain problems. In his prior life he had applied optimization to tough problems in the airline industry. Here are his thoughts.
Like any sweeping generalization, you will find numerous examples to the contrary but I'm inclined to think that airlines have embraced optimization much earlier than manufacturers and hence have reached a higher level of sophistication. There are several distinct reasons for this:
  1. Early adoption
  2. Dramatic changes hastening acceptance
  3. Embedding in business processes
  4. Early growth of commercially viable companies offering optimization solutions
  5. Customer Relationships
Early adoption
The fact that airlines industry started sooner is no accident. Here's why: The use of information in decision-making can be thought of as a 3-phased process: It starts with the development of systems that provide visibility, i.e., easy access of information to the decision maker. It moves on to predictive modeling, i.e., providing the decision maker with the ability to see the consequences of those decisions. Finally, once the decision maker has faith in the correctness and the currency of the underlying information infrastructure and in the fidelity of the predictive model, there is a readiness to allow the machine to recommend, and in some cases, even make those decisions. This is where optimization comes in.
1.1 Head start in Systems for Visibility
In the airline industry, the information infrastructure like airline reservations systems and flight operation systems, was built out of necessity. Airlines, by their very nature, are globally dispersed enterprises that require precise coordination across multiple locations. Therefore, they were among the early adopters of information systems and networks.
Airline reservation systems (later following global distribution systems like Sabre) were essential because the product had to be made accessible to a large number of relatively small consumers (unlike in the case of manufacturing enterprises where there are only a small number of customers with whom the enterprise established a long-standing relationship via salespeople). Similarly, flight operations systems were essential because of the operations at various locations were unavoidably and finely intertwined by the movement of planes and crew (unlike in the case of manufacturing enterprises where each location isolated itself by means of inventory).
Although they were probably not meant as such, these early investments into the information infrastructure became phase one of this 3-phased growth to the optimization of airline operations. All the data that the models needed was captured routinely, accurately, and at the finest level of granularity. This helped in the development of tools such as flight and crew scheduling and pricing and yield management which heavily leveraged optimization.
Contrast this with manufacturers. Most manufacturers operated with tight control on customer orders and procurement orders. But there was little data collection and hence control of material that was work in progress. There was little useful data collection on the bills of material needed to produce a product and machine capacity needed to produce an unit of product. enterprise resource planning systems in the manufacturing industry is a relatively recent introduction. In that sense supply chain management companies came into existence at just the right time. A little earlier and it may have been stillborn because of lack of ready data.
1.2 Early acceptance of predictive models in airlines
Airlines tend to place a greater degree of credibility in their predictive models. In contrast, in my experience, planners in manufacturing enterprises seem to place much less credence in models of their supply chains. I can think of a couple of reasons:
Airlines have the advantage of dealing with a large number of customers, each of whom is relatively unimportant in the larger scheme of things. Manufacturing enterprises tend to have far fewer customers; in such cases, it becomes important to weigh something as nebulous as "long-term relationship" against short-term gains.
Airlines tend to move in relatively stable patterns over extended periods. Their operations tend to follow a repeating weekly pattern for months at a time. This allows operations researchers to model long-term behavior by endlessly repeating that weekly unit. In contrast, there are no simple long-term models for supply chains. Predictive models are almost always myopic. (Even corrections to such myopia are often heuristic in nature.) The optimal solution to an inexact predictive model can recommend decisions that are patently wrong.
As a result, airlines show a much greater level of confidence in models that predict the bottom-line consequence of their decisions. For instance, when it comes to fleet assignment decisions (fleet assignment being one of the most important components of flight schedule optimization), the difference between assigning a 737 on a flight from DFW to ORL as opposed to a 757 is quantified in dollar terms.
In contrast, a simple bottom-line based analysis is almost always rejected by planners in manufacturing industries. Because of the finite-horizons of planning models, a myopic predictive model fails to recognize the effect of short-term choices on long-term relationships between the enterprise and its few, large customers. Instead, they prefer to "optimize" against a complex set of conflicting, non-commensurate objectives. The use of such schemes often makes the goodness of a plan more a matter of the predisposition of the beholder rather than any innate quality therein. This makes it hard to measure, and hence justify, the true value of optimization.
1.3 Early breakthroughs in airline optimization models
Operations Researchers sought out areas where there is a wealth of opportunity to make an impact via optimization. As Tom Cook, founder of American Airlines Decision Technologies (later part of the Sabre group) points out in his article on success factors of OR projects, you need to check carefully for the following:
  • Value of the decision
  • Data availability
  • Progressive decision makers
  • Funding and resources
Airlines provided all of this in the early 80s. This drew the attention of many of the pre-eminent OR researchers and practitioners throughout the 80s and the early half of the 90s. Airline OR continues to remain a hot area of academic research. It also serves as the source of several benchmark problems that are used by vendors of optimization components. All this had the effect of reinforcing the move to apply optimization in airline decision-making. Compare that to supply chains. Individual pieces of the supply chain have been optimized for a long time. But optimization of entire supply chains was attempted successfully only in the late 90s.
Do you agree with Narayan?
Look at part 2 of this interview for thoughts on the other 4 reasons from Narayan. In the next few weeks, I am hoping to coax him to talk about where he sees the future of optimization in supply chain management.
Karthik Mani

No comments:

Post a Comment