The first difficulty is that quantitative analysts must often consider conflicting viewpoints in defining the problem. For example, there are at least two views that managers take when dealing with inventory problems. Financial managers usually feel that inventory is too high, as inventory represents cash not available for other investments. Sales managers, on the other hand, often feel that inventory is too low, as high levels of inventory may be needed to fill an unexpected order. If analysts assume either one of these statements as the problem definition, they have essentially accepted one manager’s perception and can expect resistance from the other manager when the “solution” emerges. So it’s important to consider both points of view before stating the problem. Good mathematical models should include all pertinent information. As we shall see in Chapter 6, both of these factors are included in inventory models.

The next difficulty is that problems do not exist in isolation and are not owned by just one department of a firm. Inventory is closely tied with cash flows and various production problems. A change in ordering policy can seriously hurt cash flows and upset production schedules to the point that savings on inventory are more than offset by increased costs for finance and production. The problem statement should thus be as broad as possible and include input from all departments that have a stake in the solution. When a solution is found, the benefits and risks to all areas of the organization should be identified and communicated to the people involved.

The third difficulty is that people have a tendency to state problems in terms of solutions. The statement that inventory is too low implies a solution that inventory levels should be raised. The quantitative analyst who starts off with this assumption will probably indeed find that inventory should be raised. From an implementation standpoint, a “good” solution to the right problem is much better than an “optimal” solution to the wrong problem. If a problem has been defined in terms of a desired solution, the quantitative analyst should ask questions about why this solution is desired. By probing further, the true problem will surface and can be defined properly.

Even with the best of problem statements, however, there is a fourth danger. The problem can change as the model is being developed. In our rapidly changing business environment, it is not unusual for problems to appear or disappear virtually overnight. The analyst who presents a solution to a problem that no longer exists can’t expect credit for providing timely help. However, one of the benefits of mathematical models is that once the original model has been developed, it can be used over and over again whenever similar problems arise. This allows a solution to be found very easily in a timely manner.

One problem in developing quantitative models is that a manager’s perception of a problem won’t always match the textbook approach. Most inventory models involve minimizing the total of holding and ordering costs. Some managers view these costs as unimportant; instead, they see the problem in terms of cash flow, turnover, and level of customer satisfaction. Results of a model based on holding and ordering costs are probably not acceptable to such managers. This is why the analyst must completely understand the model and not simply use the computer as a “black box” where data are input and results are given with no understanding of the process. The analyst who understands the process can explain to the manager how the model does consider these other factors when estimating the different types of inventory costs. If other factors are important as well, the analyst can consider these and use sensitivity analysis and good judgment to modify the computer solution before it is implemented.

A second major concern involves the trade-off between the complexity of the model and ease of understanding. Managers simply will not use the results of a model they do not understand. Complex problems, though, require complex models. One trade-off is to simplify assumptions in order to make the model easier to understand. The model loses some of its reality but gains some acceptance by management.

One simplifying assumption in inventory modeling is that demand is known and constant. This means that probability distributions are not needed and it allows us to build simple, easy-to-understand models. Demand, however, is rarely known and constant, so the model we build lacks some reality. Introducing probability distributions provides more realism but may put comprehension beyond all but the most mathematically sophisticated managers. One approach is for the quantitative analyst to start with the simple model and make sure that it is completely understood. Later, more complex models can be introduced slowly as managers gain more confidence in using the new approach. Explaining the impact of the more sophisticated models (e.g., carrying extra inventory called safety stock) without going into complete mathematical details is sometimes helpful. Managers can understand and identify with this concept, even if the specific mathematics used to find the appropriate quantity of safety stock is not totally understood.

One problem is that most data generated in a firm come from basic accounting reports. The accounting department collects its inventory data, for example, in terms of cash flows and turnover. But quantitative analysts tackling an inventory problem need to collect data on holding costs and ordering costs. If they ask for such data, they may be shocked to find that the data were simply never collected for those specified costs.

Professor Gene Woolsey tells a story of a young quantitative analyst sent down to accounting to get “the inventory holding cost per item per day for part 23456/AZ.” The accountant asked the young man if he wanted the first-in, first-out figure, the last-in, first-out figure, the lower of cost or market figure, or the “how-we-do-it” figure. The young man replied that the inventory model required only one number. The accountant at the next desk said, “Hell, Joe, give the kid a number.” The kid was given a number and departed.

A lack of “good, clean data” means that whatever data are available must often be distilled and manipulated (we call it “fudging”) before being used in a model. Unfortunately, the validity of the results of a model is no better than the validity of the data that go into the model. You cannot blame a manager for resisting a model’s “scientific” results when he or she knows that questionable data were used as input. This highlights the importance of the analyst understanding other business functions so that good data can be found and evaluated by the analyst. It also emphasizes the importance of sensitivity analysis, which is used to determine the impact of minor changes in input data. Some solutions are very robust and do not change at all following certain changes in the input data.

The first concern in developing solutions is that although the mathematical models we use may be complex and powerful, they may not be completely understood. Fancy solutions to problems may have faulty logic or data. The aura of mathematics often causes managers to remain silent when they should be critical. The well-known operations researcher C. W. Churchman cautions that “because mathematics has been so revered a discipline in recent years, it tends to lull the unsuspecting into believing that he who thinks elaborately thinks well.”¹

The second problem is that quantitative models usually give just one answer to a problem. Most managers would like to have a range of options and not be put in a take-it-or-leave-it position. A more appropriate strategy is for an analyst to present a range of options, indicating the effect that each solution has on the objective function. This gives managers a choice, as well as information on how much it will cost to deviate from the optimal solution. It also allows problems to be viewed from a broader perspective, since nonquantitative factors can be considered.