The common misunderstanding of forecasting has two parts. The first is to understand what partial information the forecast should provide. The second is how to make a good decision based on the forecast information.

The common ignorance regarding the first part is focused on using the forecast as a single number. The mathematical/statistical handling of all uncertain functions includes, at the very least, two parameters. The common minimum description of uncertain behavior is the use of the average and the standard deviation. Another option is to describe a spread of possible results by the confidence interval: a range of results that encompasses, according to the forecasting assessment, 95 percent or more of the possible results.

The common use of the forecast as a single number is causing huge confusion because the essential range of results is missing. Thus, it is almost useless and definitely misleading. The vast majority of management reports contain only the column of the forecast, namely the average predicted forecast. The forecasting error, the equivalent of the standard deviation, is not mentioned in those reports. The basic misunderstanding is even more destructive when people, mainly from Sales, are required to give their "forecast" for the next period. What kind of single number does management hope to get? The average? Do they really get a fair assessment of the average from the salespeople? Could it be that a typical salesperson provides his intuition regarding what he hopes to sell, rather than his or her estimation of the average? The salesperson wants to be sure he would have available all the quantity he might sell. On the other hand, the salesperson may want to give his estimation on what he is sure he can sell and not be caught failing to meet his forecast. The point is that for an unclear question, people get answers to whatever interpretation the person answering the question has in mind.

Is the "average" forecast the required information for a decision regarding how much to produce for stock? Let's consider the following example.

The forecast for next month sales is 1000 units. We should have in stock, at the start of next month, 300 units (also "on average" depending on the sales until then). Assuming the policy is to produce the whole monthly requirement in one batch (usually an unwise policy, but that is not the point right now), should we produce 700 units?

Well, if that is the only information we have, then we are led to make a faulty decision. A proper forecast should also contain, at least, an indication of the forecasting error. Suppose the forecasting error is 500 units. The hidden meaning is that it is perfectly possible that the real demand next month would be 1500 units. Even 2000 is still a valid possibility. Of course, it also means you might sell only 500 units. Managers are required to make sound decisions even when living in such an imperfect world. A much better decision than to produce 700 units would be to produce 1700 units to cover for the valid possibility of having demand for 2000 units. Another sound decision could be to produce only 200 units when the concern of being left with unsold products is more severe than being short of products. In other words, any sound decision has to take into account the damage of producing too much versus the damage of producing too little and the larger damage should dictate whether to produce more than the average or less than the average. In most cases, the decision to produce according to the "average" prediction (based on a single average forecast number) is a truly bad decision because the element of risk is not brought into the picture. Suppose that the plan is to produce more than the average. However, without any indication of the possible spread, how should one make up his mind regarding how much more to produce?

Misunderstanding of the forecast has more aspects to it. Forecasting the sales of just one item in the coming month might be too "erratic"; thus, the idea is to forecast the sales of the whole product family. This should yield a much better forecast, shouldn't it?

Well, usually the term "better forecast" means a relatively smaller forecasting error, while the term "erratic forecast" means a very large forecast error. The problem is you cannot use that "better forecast" for a better decision on the level of the individual item. Suppose it is "known" that the sales of a certain item are approximately 10 percent of the sales of the total family. Do we get a "better forecast" for that item when we take the forecast for the sales of the family as a whole and then take 10 percent of it as the forecast for that individual item? No, you do not get a better forecast for the individual item this way. You have a gross estimation of the average, but the possible spread of the results for different units in the product family is pretty high and you cannot reduce the spread of the sales of an individual item by forecasting the whole family. For a decision on the demand level of an item, one needs its forecast including the assessment of the spread of results.

Another aspect of ignorance regarding forecasting relates to the forecasting horizon and the time periods within that horizon. Management likes to look at "the big picture" and thus wants to see not just the forecast for next month, but also for the subsequent months-at least up to one year. Suppose the forecast for next month is 1000 units plus or minus 500 units. Now, if the forecast for the month after that is also "on average" 1000, the "plus or minus" is probably larger. The farther in time we go, the larger the spread of the forecast. There are two reasons for it: 1. Naturally when estimating, the forecasting error gets larger for subsequent periods because any deviation in the trend of the sales would get larger the farther out in time we look (increased uncertainty).

2. The most troubling point in forecasting is based on the assumption that the characteristics of the past are not going to change, and thus we can deduce the future from the past. As we look farther in time, there is a higher chance that an event will change the basic parameters. Just consider the case where today your main competitor has opened his manufacturing facility not far from you and he is going to try to move your clients to him. Suddenly, the rules of the game change and you cannot rely on the past to deduce what your future sales will be.

The direction of the solution to forecasting demand is hidden exactly in the notion that for the very short-term we have a good idea of what is going to happen. Even when we look at the short-term, we must consider not just the average, but also how much we might sell. In other words, we need a confidence interval to give us a reasonable range so we can decide how to prepare ourselves to a valid level of sales. When response time is rapid, there is no real need to forecast beyond the short-term except to look at approximations of capacity, materials, and cash requirements.

The Current Undesirable Effects in MTS

Every time a production order is released without a definite customer order, it might create a surplus of inventory and at the same time delay the production of another product, which might be in high demand on short notice. There is no way to avoid these mistakes, simply because we are not prophets and we cannot really know the future.

The unavoidable result is that at any given time the finished-goods inventory of some of the products is excessively high relative to the actual demand, while there are shortages of other products. All we can hope for is to eliminate the shortages to such a low level that we have almost "no shortages," while the surpluses of inventory are rather limited.

The current state causes many other undesirable effects within the shop. Holding too much stock takes its toll on financials, limits space for other items, and causes pressure to "get rid" of stock. Producing based on misunderstood long-term forecasts leads to producing very large batches ("we should produce this product only twice in the year to gain efficiency"), which causes long production lead times and delays for products that are truly required by the market.

Already mentioned is the confusion between MTO and MTS that leads to inability to state clearly when a current request can be met safely. Another common undesired effect happens at the level of common components that go to many end items. Components are often "stolen" by the overproduction of end items with low demand, while other items are short. What makes the "stealing" effect special is that it creates anger and tension because the cause and effect are visible to employees. One can clearly see how the decision to produce too large a batch has exhausted all the necessary components for a truly urgent order.

What to Do? The Direction of the Solution

The Basic Principle of Flow

The immediate conclusion in understanding the characteristics of forecasting should be: the faster we can respond, the more reliable is the forecast. Embedded in that sentence we have the recognition that we should not look too far into the future when framing production orders. However, there is a minimum time into the future where we need to ask the question: how much might we sell? The default assumption is that we want no shortages and for that, we are ready to pay the price of holding more inventory than we would need in a world with perfect knowledge. Therefore, our aim most of the time is to have full availability of those items on which we choose to maintain excellent availability, while the amount of stock is nicely controlled at a level that is still appropriate for preventing shortages.

The question points to the need for a different type of forecast, not the regular one. The question is not directed at the average sales within the response time, but at what we might actually sell. In other words, forecasting the maximum sales that we could reasonably expect for that period of time. In order to fully support availability while refraining from overproduction, two practical insights emerge: 1. Production still needs to focus on the flow to finished-goods inventory, flowing the required quantity as fast as possible through the shop.

2. Unless we have a good reason to believe that the market demand is going to change, or that the current inventory in the system is either too high or too low, then a simple straightforward reaction to any sale is to replenish that quantity. This means that replenishing the exact quantity of what was sold is a natural default. From the production planning perspective, it means that everyday production should initiate producing the exact quantity of what was sold yesterday.

From the two insights, it is clear that we need to determine the appropriate inventory in the shop that would provide perfect availability, thus maintaining the fastest flow of goods to customers. The other critical point is to improve, and keep improving, the internal flow.

Another understanding emerges. If the objective is to provide perfect availability to customers, then we should state that openly and probably take one more step and commit to maintain that availability openly by letting our customers know that is our commitment.

From MTS to MTA

The verbalization of the ultimate objective is: We commit to our chosen market to hold perfect availability of a group of specific end products at a specific warehouse.

This objective has two critical elements in it. One is the marketing message, defining the target market, the items that it includes, and possibly also some limitation of the one-time demand that such an availability would provide. The other is the operational element. Once a commitment is given, Production must perform to meet that commitment.

Let's now clarify the relationship between MTS and MTA. Certainly any case of MTA requires MTS, unless production can be done in a few seconds. However, many cases of MTS are definitely not MTA. Those cases happen every time there is no concrete commitment to availability.

Let's present two different examples for MTS that is not MTA.

Example 1 Painters, including famous ones, paint regularly for stock, meaning without a specific client commissioning the painting. However, many paintings are done only once, there is just one unique single copy. In some cases, a limited number of copies (authorized prints) are produced. This definitely is not a commitment for availability. Similarly, exclusive fashion items are also promised to be single units; there is no promise of availability.

Example 2 Items that are going to be sold in a specific period of a few days, for instance, souvenirs for a specific sporting event, like hats or T-shirts, with the appropriate logo and colors for one of the teams for the big final game, will need a lot of stock before the event. After the event, sales will be very low. The time those items are sold is so short that there is no practical chance to replenish. In such a case, there is no clear commitment for availability. Actually, the producer hopes to sell all his items (while not losing too many sales) and usually that means leaving some demand unsatisfied.

Determining the Appropriate Inventory

The idea of replenishing exactly what is sold, or more accurately what is consumed,4 has an interesting ramification: the inventory in the shop remains fixed. However, it is fixed through the whole shop floor, both at the finished-goods and work-in-process on the shop floor itself. Therefore, the set of parts and assemblies needed to complete the end product committed for availability may exist in various levels of completion on the shop floor. When fully fabricated, this inventory would equal the quantity needed for committed availability.

Of course, here and there the total inventory might be less than the regular fixed amount because Production is slow to release the next work orders, but the concept means trying to keep the total stock fixed.

There is a lot of sense in determining a fixed quantity per item to protect availability. Ideally, it'd be best to keep fixed stock in finished goods. However, this is quite impossible because once there is demand the stock goes down. Then what do you do? The only way to react to actual demand is to initiate replenishment. Thus, defining the "shipping buffer,"5 the protection mechanism that protects the availability, as the total amount of finished goods plus the work-in-progress (WIP) is the simple and straightforward way to institute the appropriate protection mechanism.

Let's call the buffer, the fixed stock in the system, the target level for this item. How should the target level be determined? From the time one piece is sold until the time the replenishment piece arrives at the finished-goods warehouse, availability has to be maintained. Let's describe the average time it takes to replenish a piece as the replenishment time6. Most certainly, the target level should include the average demand within the replenishment time, but this is definitely not enough, as there is a need to address what might be sold and also to address any case where the replenishment will take more than the average replenishment time.

There are two practical ways to handle the determination of the target levels. One way is to take the above average demand within replenishment time, which is information that is usually easy to get, and multiply it by a "paranoia factor" to include peaks of sales and certain blockages in production. In the area of the production floor, a minimum "paranoia factor" is an additional 50 percent (factor of 1.5 of the average). Using this number is recommended in situations where no sequence-dependent setups exist, thus managing the priorities on the floor (still to be discussed in this chapter) toward rapid work flow. In other cases where the demand fluctuations are especially high and blockages in the flow are frequent, a factor of 2 should be used.

Another approach looks at the recent 6 to 12 months of history for the actual maximum sales that have occurred within that window of time defined as the reliable replenishment time. The reliable replenishment time means that when you really need it you can safely get it within that time.7 Two important points to notice at all times: 1. If currently there is no stock, or very little, in finished goods (do not count stock that is already assigned to clients), then first build the finished-goods stock and only later move to the TOC MTA solution. More on that point later, but please take note!

2. The determination of target (maximum) level based on the appropriate criteria as discussed above is only to set the initial inventory levels. As we'll see, future changes to the target levels (increase or decrease) are done based on a special algorithm that monitors the actual behavior of the finished-goods stock.

Buffer Management in MTA

Once the target level is operational, daily replenishment orders are initiated based on the previous day's consumption. Every production order for replenishment is released without any due date. The immediate question is how should the priorities on the floor be determined? There is no doubt that there is a real need to settle the priorities.

The idea is that the priority of the orders depends on what lies downstream of the production order, in other words between the order and finished goods. The amount of stock not dedicated to any client, and therefore available to fill new customer orders relative to the buffer size (the target level), is the real indication to how urgent the production order is. We do not really expect to have 100 percent of the target level completed and in the finished-goods warehouse. This would be way too much as we expect the replenishment to arrive at the warehouse much faster than the time the whole target level is going to be consumed.

Let's look at the situation shown in Table 10-1. It shows the full picture of a target level inventory buffer for a product P1. Suppose that a production order for 200 units for product P1 lies somewhere on the shop floor. Downstream from that order is the finished-goods stock, which contains 100 units. Suppose that the target level, the amount of inventory we believe would provide excellent availability, is 500 units. We know that the whole target level should be in the production system somewhere, either in finished goods, or at some level of completion within the shop floor. This means that right now only 20 percent of the target level actually resides at the finished-goods inventory. It looks like replenishing the finished-goods stock is urgent. Note also the fact that the size of Order 1 of 200 units is not required for the assessment of how urgent the order is. The question of urgency relates to how much is in finished stock downstream from Order 1. Like BM in MTO, we like to denote the priority of any order by a color code: green, yellow, or red. Color code definitions are defined more fully next in this section. They are shown in Table 10-1 to complete the picture in our example showing the relative priority of the orders. Order 1 is urgent and is in a red buffer status. Order 2 is upstream from Order 1. It has 300 units downstream from it, or 60 percent of the target inventory. It is in a yellow buffer status. Order 3 for another 100 units has 80 percent downstream in front of it, 80 percent of the target, and a buffer status of green.

TABLE 10-1 Availability Targets and Priority Status of Orders for a Buffer Target of 500

Defining Buffer Status

We define the state of the finished-goods buffer when containing two-thirds or more of the target level as green. In other words, one-third or less of the buffer is not in the finished-goods inventory, but somewhere on the way.

In a similar fashion, when the finished-goods inventory contains between one-third and two-thirds of the target level, as shown in Fig. 10-1, we call that state yellow.

When the on-hand stock, the inventory at the finished-goods warehouse, is less than one-third, meaning more than two-thirds are not at the warehouse, then the state is red.

FIGURE 10-1 The structure of the stock buffer.

At any given point in time, the stock buffer is divided into the part that exists as finished goods on-hand and available for immediate sale, and the stock that complements the previous part to the full target level. Assuming we keep the target level intact, then the latter part is in the form of all the product components required for the finished goods to be equal to the target level. The part of the buffer that is not in the finished goods is called "penetration into the buffer" because that stock has not yet completed manufacturing and therefore is not currently available for immediate shipment. The buffer status is defined as the percentage of the penetration into the buffer. Table 10-1 shows Order 1 with only 20 percent of the target inventory ahead of it in a status where 80 percent of the target is still on the shop floor. Therefore, its buffer penetration is 80 percent, which is greater than the 67 percent limit putting it in the red zone. When the penetration into the buffer is less than 33 percent, we are in green-actually too much finished-goods stock at the moment. When the buffer is in yellow-buffer status between 33 and 66 percent-then the buffer state is truly satisfactory. (So we have at least one-third of the target in finished-goods inventory and the rest lined up in the shop for fabrication when needed.) Likewise, when the buffer penetration is equal to or above 67 percent, the buffer is red. The immediate message is expedite the order as you are about to stock out.

The priority rules are now clear: red orders should be expedited and should trigger management attention. Red orders definitely should have priority over all other orders, while yellow orders have priority over green orders.

Within the same color code, the decision of which order to do next is in the hands of the operators on the floor. The author believes that the buffer status, on top of the color code itself, is valuable information for the operator. If two red orders show up, one with buffer status of 70 percent and the other with 96 percent, it seems clear that one needs a very persuasive argument not to process the 96 percent order first. However, if one order is 70 percent and the other is 74 percent, then the real choice probably lies in other factors.

Generating Production Orders and the State of Capacity

The ideal situation is to generate new production for all items that were consumed the day before every day. What is the obvious negative branch of doing exactly that?

What could easily happen is that too much time is devoted to setups. Should we be concerned about doing too many setups? When at least one resource is losing too much of its protective capacity, then we definitely should care. The problem in losing the protective capacity is that the replenishment time grows longer and longer. Then, with longer replenishment times, more and more end products would become red. When the number of red orders exceeds 20 percent, then the whole scheme of maintaining priorities loses its effectiveness and a significant number of shortages will occur.

The lesson we should keep in our mind is that MTA requires a certain level of protective capacity. We deal more with that issue because losing protective capacity might be caused by too much total demand (not just the demand for one item, but also the demand for the whole product mix). Right now, we do not wish to let too many setups be the cause for losing protective capacity.8 There are two ways to deal with the issue: 1. Dictating a minimum production batch. The minimum batch is not part of the target level! It comes on top of it. That means that once the inventory in the pipeline plus on-hand is less than the target level, a production order is generated, but its size equals a quantity at least equal to the minimum batch quantity. We may discover that the total inventory is above the target level, but it should be less than the target level plus the minimum batch.

2. Managing the capacity of the capacity constrained resource (CCR) and releasing new production orders only when it seems reasonable, the CCR would work on them soon. The concept of the "planned load" was defined in Chapter 9; here we need to define the planned load for the specific environment of MTA.

Definition9: The regular planned load for MTA is the summation of the derived load on the CCR of all the production orders already released that have not yet been processed by the CCR.

Releasing orders only up to a certain limit of the regular planned load causes the release of new orders to be under control as this procedure releases new production orders only up to that level where the regular planned load approaches the agreed limit. Production orders that were not released today will be considered again the next day. The regular planned load for the next day should be smaller by the amount of work the CCR has processed during the previous day and this allows more orders to be released.

What should the criteria be for choosing which production orders to release? The relative priority of each new production order competing for being released today is based on how much replenishment is required to reach a full target level of inventory in the shop floor (including the warehouse) relative to the target level. In other words, the production orders in the queue awaiting release are getting a buffer status, similar to the orders that were released. That status serves as the relative priority of the order. The orders with the highest status would be released first. Every order that is released updates the regular planned load. Once the planned load limit has been exceeded, the daily release of orders is stopped. The rest of the queue awaiting release has to wait until the next day, and as consumption continues, their relative priority will go up accordingly.

Let's demonstrate the mechanics of this procedure with the following example: Suppose the replenishment time is 5 days with 16 hours of CCR time every day (5 days 16 hrs/day = 80 hours). A natural limit for the regular planned load is 80 percent of the replenishment time-64 hours. We use only 80 percent of the replenishment time assuming that this way the overall replenishment time, including the operations downstream of the CCR, would be easily maintained (the part downstream of the CCR usually takes much less time than the time to reach the CCR waiting for its turn and then being processed by the CCR). This procedure of release would avoid having too much WIP on the shop floor. Suppose that on a given day the planned load reaches 50 hours. Suppose 11 items are in the production pipeline. The relevant information is shown in Table 10-2.

As the planned load is already 50 hours and the limit is 64 hours, we can release up to 14 hours of work. Right now we need to release 19.3 hours-a little too much. The most straightforward method is to add from the highest priority, P10, then P3, then P1, P7, P5, P8. This would bring us to a total of 13.2 hours. The next, P9, would penetrate the 14 hours limit. Should P9 be released? This decision should be subject to the judgment of the person in charge (usually the master scheduler) and is not too critical. The main point is that P6, P2, and P4 would wait at least one additional day.

In most cases, a certain minimum batch is required. That must be considered in addition to the CCR load. When a minimum batch is used, then the priority is determined just by the quantity required to replenish to the target level. However, the load on the CCR needs to consider the size of the batch. For instance, if the target level is 100 and currently there are 49 on-hand and 50 in the pipeline (it could be that the 50 units are included in two production orders, not yet finished, somewhere in the shop floor, etc.). The replenishment to the target level requires only one unit, but the minimum batch is 25. The priority for releasing the next replenishment order is based on 1 100/100 = 1%, but the time on the CCR needs to consider a batch of 25. When the load on the CCR and the relative priorities of the other orders will permit the release of this order consisting of 25 units to the floor, the time it'll take on the CCR is planned based on the load of processing 25 units. It may thus prevent other orders from being released on that day.

TABLE 10-2 List of the Orders in the Queue Awaiting Release

Peak and Off-Peak Behaviors

At off-peak periods, there should be no blockages to the release of small replenishment orders to the floor. When the demand starts to pick up, then monitoring the replenishment order release starts to be critical because continuing to release small daily batches increases the number of setups, creating more blockages of the flow in the shop. The impact of this algorithm is to set a limit on the actual wait time to the CCR, thus limiting WIP to a degree that will allow orders in the shop to move at a speed that is in line with the assumed replenishment time. However, at the same time there are orders that still wait to go into the shop. This means the actual replenishment time is longer than the formal replenishment time. If that situation extends for a long time, there is real risk of losing control of the availability commitment. The suggested behavior of prioritizing the release leads to a dynamic batching mechanism, where at off-peak the batches are naturally small (equal to daily sales) while at peak periods larger batches are used because of delaying the release of certain replenishment requests. This provides a little relief to capacity. In the short-term, this institutes the right priorities for the system in a way that saves setups on the CCR.10 We still need to examine the longer-term impact of monitoring the level of capacity and taking the right measures on time because enlarging the batches has only limited impact on the capacity and it could be that a true need to increase capacity is emerging.