In contrast with Figure 7, now the curve (in this case the Equal Service curve) is out in front of the current position. This means that instead of the current 52 day fixed time supply, we have the option of moving left to the curve (keeping total inventory constant while improving service), moving down to the curve (keeping aggregate service the same while reducing inventory) or moving diagonally left and down (cutting inventory while raising service). The latter option is the preferred one. The reason for this is that no matter what your service and inventory targets, every company must first demonstrate to top management that it can control service and inventory. The best way to do this is to reduce inventory at the same time you improve service. Once you have demonstrated your credibility on this issue then you can implement your ultimate targets.

Let's look at the details by part for our 98% service target.

Now this is 98% service! We have consistent 98% dollar fill service for every part in the service group; and, because we are computing targets based on the forecast error, the inventory required to achieve 98% is only $3.1 million, a drop of 25% from the current $4 million required to achieve only 96.4% aggregate—but which varies wildly from that aggregate at the detailed level. Here the variation is in the days of inventory required to achieve the 98% target.

Now that we have eliminated variation in individual service levels, let's consider whether we might allow a little variation. If we do so, perhaps we could increase service slightly for products with small forecast errors in order that we might decrease service slightly for products with large errors. We will constrain the solution, however, to require that the aggregate service still meet our target. At 98% for example there is less room above the target than below. This means that we will have to increase service on more items than we decrease. The solution to this problem is called optimization. While there are an infinite number of combinations that achieve the aggregate service target, only one solution has the lowest possible inventory. Optimization is the name of the technique which finds this optimum solution. Let's begin by looking at the high end of the service range, shown in Figure 11.

This shows us a range of service from 96% to 100% (actually 99.99%) along with the requisite safety stock in aggregate dollars and as an average (not fixed) time supply. This analysis indicates what is possible to achieve by optimizing dollar fill and is in stark contrast to the current 52-day fixed time supply. For example, if we are willing to drop service from the current 96.4% to 96%, the corresponding safety stock drops almost 40% from about $4 million to about $2.5 million. Conversely if we need 100% service for this group, the cost in inventory is somewhere between $5.5 million and $6.5 million, depending on how precisely you want to hit the target ($6.5 million buys 99.99%, the others a bit less).

Assume you have a safety stock budget of $3.5 million for this product line. Looking up this figure in column one of Figure 11, you find it falls between $3.3 and $3.6 million. The corresponding service falls between 98.6 and 99.2%. You want to know service more precisely, however, so do an analysis on this service range to home in on the correct value.

From the analysis in Figure 13 you see that your budget falls between the $3.471 and $3.5011 million lines. Since the corresponding service for both is 99.0, we can conclude that a safety stock budget of $3.5 million for this product line will buy us 99% customer service. This is a savings of $0.5 million from the current fixed time supply while still increasing service from 96.4% to 99%.

Using the same type of repetitive search (here done for us by the computer) you can start with a particular service target and observe the required safety stock. For example, assume you need to achieve 98% for this product line.

Setting service to 98% (up from 96.4%) requires just over $3 million in safety stock, a savings of over $1 million, or about 25%. Let's look at the details for our 98% target:

As you can see, we have reintroduced some small service variations, although nothing like what we started with. Service on three parts is slightly higher, but none higher than 99%. Service on one item has dropped to 94.1% (but not to 60%). It is no coincidence that this item also has the highest inventory in days. Item B has a large forecast error relative to the others. This serves as further proof that fixed time supply is not the method to use. Safety stock has dropped for parts A, C, and D; by about 75% for part C. It has increased by over 50% for part B. Part B with a 114-day safety stock can support just 94% service, while part C with a 19-day supply can expect 99% customer service!

Let's put all three methods up for comparison. From the service point of view, here's how they compare.

Under the fixed time supply method, service is essentially out of control. The only possibilities to consider are the other two: equal service if you must have 98% on everything, or optimization if you can allow a slight variation. From an inventory point of view, however, the differences are even more striking:

Either method 2 or 3 is a significant improvement over fixed time supply. In this case optimization shows about a three percent improvement over equal service. Sometimes, however, the difference is substantially more. A client using optimization was once asked by an overseas affiliate what he thought of their plan to use a less capable package. His immediate reaction was that they would give up the benefits of using selectable forecast calendars (see Reference 2) and optimization. (Since this was a service parts application, the first was a significant blow to forecast accuracy.) The affiliate expressed doubt that something as nebulous sounding as "optimization" could have any substantial impact. The client, intrigued, said "Let's find out." Stepping over to his computer, he noted the service level being achieved by their current optimization and then ran an analysis of what it would cost to achieve that same service level without optimization (using the equal service rule). The inventory went up fifty percent!

There are many forms of optimization. Each achieves the target service level for the chosen measure of service while minimizing the inventory required to do so. Or, from the other point of view, for a given inventory investment each maximizes the service obtained for the chosen measure. The above dollar fill optimization provides the highest dollar fill service (i.e., minimizes the dollars short) for any given inventory level. Other optimizations minimize the number of customer line items short, the number of units short, cases short, pounds short, etc. One can even optimize on profit fill, achieving the highest profit dollars possible from any given inventory level. Of course, your customers are not likely to want to hear that the reason you are out of a needed part is that you make little profit on it.

There is one other optimization which—while it is not directly related to customer service—can make your life easier. It is called Optimized Lot Fill. This method minimizes the number of production lots (manufactured or purchased) which will arrive after you have run out of stock. That, in turn, minimizes expediting. If you do expedite, this means you could set an inventory which achieves, say, 95% optimized lot fill. Then if you can successfully expedite the remaining 5% of your replenishment lots, this means you can achieve 100% service, while only paying for 95% with inventory, plus expediting 5%. For many companies that is a real bargain. You can literally trade the cost of inventory against the cost of expediting to achieve your objectives.

Going back to the 98% service level using optimized safety stock, what should you do if the $3 million is too high, yet 98% is still the target? First, consider that there are several things which affect safety stock. These are (1) forecast error, (2) service level, (3) lead time, and (4) working stock (or replenishment frequency). Except for forecast error, there are penalties associated with tweaking any of these in order to reduce the safety stock. The penalties may or may not exceed the benefits in reduced safety stock. Let's consider each of these areas.

Of the four areas, this is the one that is (almost) free. A reduction in the forecast error yields a linear reduction in safety stock to achieve any given level of service. This means a 25% reduction in forecast error yields a 25% reduction in safety stock. The only penalty associated with attacking forecast error is the time you invest in doing so.

In our example, you could look over the four parts and choose a likely candidate to work on. Part A seems a good choice since it has a safety stock (in dollars) more than twice as great as any of the others. Part B may be a better choice, however, since its safety stock is almost 3.5 times greater than part A when expressed as a time supply. This is the same as thinking of the safety stock as a percentage of the forecast, and in B's case it is for service substantially lower than A's.

You might want to check with the sales people or product managers to find out whether unusual events are affecting the ability to forecast this part. It may be that there are people with information which the computer does not have. This is called gathering marketing intelligence.

Another approach to reducing forecast error is to consider whether you are trying to forecast part B too frequently. You may find that by forecasting the part less frequently than monthly—quarterly or semiannually, for example—you can substantially reduce the forecast error.

Recently a consultant, in a column in a major trade publication, expressed the opinion that all companies should forecast all of their parts monthly. His reasoning was that weekly forecasting was too often for some companies and quarterly forecasting was too infrequent for others. This is rather like saying that since size 6 shoes are too small for some people, and size 10 shoes are too large for others, everybody ought to wear a size 8!

A more enlightened approach is to forecast each item on the calendar most appropriate for that item. In the fast food business this can mean forecasting weekly, biweekly, and in some cases, daily. For the service parts environment it means forecasting bimonthly, quarterly, semiannually, and even annually. One company doing this achieved a 43% reduction in their workload at the same time they achieved 30% lower aggregate forecast errors (and inventory)! This method is discussed further in References 2 and 4.

While we said we wanted to maintain a 98% service level, you might take this opportunity to re-examine your service categories. This might indeed be a category which requires 98%, but you ought to consider whether part B belongs in this service category. If, after gathering information and talking with the right people, you decide that part B can be assigned to a different (i.e., lower) service group, you will find the total safety stock requirement to be less. Even using optimization, the service for the remaining parts can drop closer to the 98% target now that they no longer have to compensate for the huge error in part B, thus reducing their safety stock. If B is set to 94.1% or anything lower, the safety stock for it will also be reduced.

Service category considerations depend greatly on the type of business. Some of the things to consider are how the customer uses the product (critical vs. non-critical) and the consequences of running out, competition, ingredients, manufacturing processes, suppliers, etc. You may find that the typical ABC classification scheme bears little relevance to the establishment of service classes.

The longer lead time for a part, the greater its safety stock, because it must protect the forecast over a longer period of time. This is not a linear, but rather a square root relationship. This means that reducing the lead time by a factor of four (75%) reduces the safety stock by a factor of two (50%). This is a good area to consider if you have unrealistically inflated lead times on file. You want the lead times in the computations to reflect reality as closely as possible.

If your lead times are realistic and accurate, then reducing them is likely to be associated with increasing the cost of the product. For example you might require your suppliers to keep larger stocks so you (and they) will be able to respond more quickly. Most likely they will want to be paid for this privilege. In fact, you might consider doing just the opposite. You could find that freezing the schedule for a longer period results in such additional efficiencies that your costs drop more than your safety stocks increase. (Perhaps we should add cost review to the above list.)

An alternative to having your vendors maintain your inventory is to keep an inventory yourself in raw materials or components (also known as semi-finished). Recent developments in technology have made it possible—just as you can compute finished goods safety stocks—to compute component safety stocks. Where components or intermediates have long lead or processing times, and where there is some amount of commonality of components, this technique can lead to significant (i.e., 30% or more) reductions in total inventory (finished plus component plus raw material). For more information on this, contact the author.

If you do address lead time, use Pareto analysis to decide where to spend your time. One service parts company did such an analysis on what their inventory savings would be if they could cut their lead times. They were startled to discover that by cutting lead times on only 13% of their parts, they could achieve 80% of the total inventory savings they would achieve if they cut lead times on all their parts. Any time you can get 80% of the pay back while doing only 13% of the work, you know you're on the right track! The trick, of course, is to pick the right 13%.

The level of working stock also affects the safety stock. This is perhaps easier seen when thinking of working stock as replenishment frequency. If a product is replenished only once per year, it is less likely to be out of stock than one which is replenished weekly. The one has, on the average, only one opportunity per year to run out, while the other has 52. Normally the best approach is to use the appropriate Economic Order Quantity (EOQ) formula to compute the working stock before computing the safety stock. If you have a situation with unusually large forecast errors, you may want to set a replenishment quantity higher than the EOQ. In this case the increased working stock is more than offset by the reduction in safety stock. Methods for both situations are discussed in Reference 1.

There is a well-defined relationship between customer service and finished goods inventory, but it is dependent on the forecast accuracy, service level, and replenishment frequency to a much greater extent than it is on the forecast size. Using techniques well-documented in the literature, it is possible to develop a price list for service, but without having to guess. Doing so puts the inventory where it is required, matching service to management's objectives. By examining forecast error, service categories, lead times, and replenishment frequencies you can manage the inventory to the point that the minimum possible inventory is used to meet your service objectives.

The benefits from doing so are (1) significantly lower inventory (e.g., 30% or more), (2) higher customer service, (3) reduced expediting, and (4) no sleepless nights worrying about whether a fixed time supply guess is sufficient.