Replacement parts management: The value of information

Journal of Business Logistics, 2001 by Tibben-Lembke, Ronald S, Amato, Henry N

Inventory managers are charged with keeping inventory costs as low as possible, while still maintaining an acceptable level of service. More accurate demand forecasts should always allow the inventory manager to better fulfill this goal.

Billions of dollars are spent every year on replacement parts and warranty claims inventory. Exponential smoothing (ES) and weighted moving average (WMA) are two of the traditional tools used to forecast demand for these parts. These methods are simple to implement and require relatively little data.

If a company tested a part to destruction or kept a database of consumer use failures, this information could be used to estimate the failure function, the probability of a part lasting a given amount of time before failing. As this paper will explain, the failure function could be coupled with the sales history of the product using the part itself to create a more accurate forecast of demand for it. However, the cost of testing and data collection information could offset any reduction in inventory cost due to improved forecasting. A parts inventory manager must be able to answer the question: when will the benefits of the better forecast outweigh the additional costs?

Generally, apart fails because of normal wear, accidents, or abuse. In the case of normal wear, replacement demand depends upon the age of the products in use, and their level of use. In the case of accidents or abuse, a wide variety of joint, random factors make it difficult to determine the demand distribution. Although the proposed method can be applied to any type of failure cause, this paper will limit its discussion to demand caused by normal wear, in which the usage rate of the product is constant across customers. For example, a refrigerator is plugged in and used at a similar rate by all consumers, whereas the wear on an automobile component varies from consumer to consumer, depending on the hours of use, number of uses, and conditions of use, including driving style.

If the firm performs laboratory testing, an approximation of the product's failure function can be constructed. If accurate records regarding past failures of products in use have been kept, this information could also be used to estimate a failure function. In either case, there is a cost associated with collecting this information.

The method to be described requires the manager to subjectively choose an initial mean time between failure (MTBF) and a failure distribution. The probability of a new part failing at a given future time then can be computed. This information can be combined with estimated sales data to predict the number of parts that will fail at some future time and thus the number of replacement units that will be demanded. To illustrate the method, a simulation model is used to generate sales and part failure data for a range of MTBFs. These data are used to compare the cost differences between the direct forecast method, and ES and WMA.

LITERATURE REVIEW

The importance of forecasting in inventory management has long been understood. In order to have sufficient supplies, inventory managers must have information about expected demands. Forecast accuracy directly impacts inventory holding and stockout costs. In most inventory models, the amount of safety stock that must be carried is directly proportional to the degree of uncertainty in demand (Silver, Pyke, and Peterson 1998, pp. 241-274), which is commonly expressed as a standard deviation. When demand is forecasted, the uncertainty is due to forecast errors. These are typically normally distributed (Silver, Pyke, and Peterson 1998, pp. 253-254). When this holds, the standard deviation of forecast error can be estimated from the forecast mean absolute deviation (MAD), as shown in equation 1 (Silver, Pyke, and Peterson 1998, p. 112):

A number of surveys have shown that the most commonly used objective forecasting methods are WMA, straight-line projection, and ES. Makridakis, Wheelwright, and Hyndman (1989) give an overview of many of these surveys. The 1998 APICS forecasting software survey indicates that the method most likely to be offered in a software package is simple ES (Melnyk 1999).

Inventory management for replacement parts is somewhat different than managing other products. As in the case of new product sales, sales of replacement parts will depend on many general economic and marketing factors, but they also will be affected by past new product sales, and this is the difference that makes possible the method described in this paper.

Historically, the majority of research related to replacement parts has focused either on estimating the total costs a company will incur over the life of a warranty (Amato and Anderson 1976; Amato, Anderson, and Harvey 1976; Blischke and Murthy 1994), or on predicting demand for replacement parts, given past failure information (Wasserman 1992; Wasserman and Sudjianto 1996). Such methods offer improvements over simpler methods, but their increased complexity puts them beyond the reach of the typical practitioner.