Toyota Leveling Paradox

The Toyota Way is full of paradoxes, and one of the most counterintuitive is the leveling paradox: that slow and steady can beat fast and jerky, like the parable of the tortoise and the hare [which the older Toyota Production System (TPS) masters often cite]. The tortoise lumbers along slow and determined while the hare sprints, runs out of breath, and takes a nap. We see a similar trend all the time in the way people work. Work, work, work to meet a deadline, and then coast for a while. Toyota would always prefer a slow and consistent pace of work.

The other side of leveling, besides a steady quantity of work, is a steady mix of work. In some ways this is even more difficult to rationalize. In manufacturing, if you’re making more than one type of part, say 50-50 production between Part A and Part B, it is natural to try to get the most production possible by building large batches of A followed by large batches of B. This is particularly attractive if it takes time to set up the process to switch between A and B. Yet Toyota would prefer to make A, B, A, B . . . This leveled mix is closer to a true one-piece flow.

These days, “build to order” is all the rage. Companies like Dell Computer have led the way building just what the customer orders over the Internet and virtually eliminating finished goods inventory. Unfortunately, what is good for the assembler is not always good for the supplier. Dell expects suppliers to keep a considerable amount of inventory that the supplier is paying for in warehouses near Dell’s assembly plant. From the Toyota Way viewpoint, Dell has not solved the root cause of the problem, but merely pushed the problem backward onto other companies. This will show up in a non-lean value stream and ultimately in higher costs and lower profits for someone—in this case the suppliers.

One might ask: “If Toyota is in fact lean wouldn’t they build exactly what the customer orders in the sequence in which they order, like Dell?” The answer is decidedly no! Customers do not order in a stable, predictable way. Yet the foundation of TPS is a stable, leveled schedule. Another Toyota paradox is that in order to have a lean value stream, you sometimes want to hold the most expensive inventory—finished goods inventory. This allows you to ship to order but build to a leveled schedule. In this chapter we will discuss the whys and hows of leveling the schedule.

Heijunka Provides a Standardized Core for Resource Planning

The term “heijunka,” as we noted earlier, means to level, or to make smooth. As with many translated words, there is some conceptual meaning lost in the translation. In most lean references, the meaning is to level the product mix over a specific time period, with the objective of producing every part every day (or even every few hours). Customers do not typically order products in specific batch sizes, but they’re often produced in batches. The concept is to produce in smaller quantities more aligned with actual customer consumption.

But this is only part of the concept. Pushing a process toward an ideal smoothness in production also pushes the process to the highest degree of flexibility and responsiveness to changing customer demand.

We have never seen a situation where customers conveniently order the same mix and quantity of parts every day. If life were only that simple! Constantly changing demand creates many issues within the value stream; namely, the alignment of resources to the constantly changing need. If the demand swings are large, there will be a need to have higher levels of inventory to adjust to the swings. Equipment capacity is limited when demand swings to the high side, and is in excess when demand is on the decline. The amount of resources needed will be higher overall—generally, set at levels necessary to meet the higher demand, and excessive when the demand falls.

The swings in customer demand create a “bull whip” effect. A slight flick of the wrist by someone skilled with a bullwhip creates a tremendous destructive force at the other end of the whip. Similarly, even small variations in customer demand at the final process ripple through the entire value stream, increasing in amplitude with each successive operation. This whip effect is particularly large for suppliers or subprocesses, at the end of the whip. This magnifying effect creates the need for higher levels of resources (and cost) to be able to accommodate the wide swings.

Equipment Needed

Methods (Lean Tools and Procedures)

Materials Needed

People Needed

Figure 7-1. Basic leveling is the core for all resource planning

This creates a condition that makes standardized work difficult, if not impossible, to implement. Remember, in standardized work we’re trying to create a precise balance of work across operations, based on the takt time, which is based on the rate of customer demand. If the takt goes up and down with the bullwhip, the work balancing and standardized work swings wildly every day. How is it possible to standardize when the takt is continually changing? This is the basis for the second form of heijunka: a self-imposed leveling for the internal benefit of the value stream (and cascading outward to suppliers as well). This leveling of demand creates a standard core onto which all resource needs are attached and aligned, as depicted in Figure 7-1.

Why Do This to Yourself?

Leveling your production is a self-inflicted choice. We say self-inflicted because it is a conscious choice, and there is a consequence. Some negative effect comes with the choice. Leveling means precise timing and being very flexible to cycle through products in small batches. This flexibility taxes the process. Any problem that causes delays will reveal itself immediately and result in a missed schedule.

For example, to level by product type means making small quantities of each item throughout the day, which means changing over from product to product. There is often some time associated with changing materials, changing a fixture, and so on. Changing over is lost production time. If the changeover process is not standardized and precise, then the large number of changeovers will lead to lost production, and the schedule will be missed. From a traditional mass production perspective, any lost production time is bad. From an overall lean system perspective, making smaller batches is good. The choice to level will leave no option but to reduce the time it takes to change over, which means having a controlled and standardized changeover process.

Some people do not like the fact that when you put this level of requirement on the process there is pressure to perform. And there’s some risk of missing production numbers. Our minds are designed to naturally protect us from risks, and the purposeful creation of risk is not a natural act. This is the rub of the Toyota Way. We must put ourselves in harm’s way, but not haphazardly. It requires a carefully crafted system, and diligent effort and management of the process, to minimize the risk. You must realize that when you sign up for the creation of a lean process, you sign up for life. If you want it to work, it’s a permanent commitment.

So, why would you do this to yourself? If we look at any typical operation, we hear terms like “bubble” and “wave,” which refer to the change in demand and the amount of work that flows through the value stream. Many managers spend time managing the waves—attempting to adjust the balance of resources and constantly fighting the fires that erupt as a result of the crashing waves. These managers are always looking for the day when they catch the wave and get things back to “normal.” Unfortunately, like in the ocean, the next wave is not far behind. This continuous riding of the waves diverts efforts from the process of improvement. Management is devoting much of their time to the containment effort rather than the strengthening activity.

Smoothing Demand for Upstream Processes

What if your demand were consistent? How would that affect your process? The introduction into the value stream of consistent “customer” demand signals (the quotes signify that heijunka is not the “true” customer demand) will provide a smoothing effect for all of the processes. This smoothing allows for the standardization of resources, which greatly simplifies planning and control.

Let’s revisit our value stream model introduced in Chapter 3 and depicted in Figure 7-2, below. We see that the future state value stream has a heijunka “board” or “box.” This is a common approach to visually displaying the leveled schedule. Each slot in the box represents a specific time period (such as 8:00 _A.M. to 8:15 _A.M.) in which the material handler might pick up a production kanban, deliver it to the pacesetter as the next order, and pick up what was produced based on the previous order. In reality there are many ways to do this; for example, sometimes the orders are posted on a white board by the hour. There are several variations on the theme, but all serve the same purpose—to show the “pitch” time increment between when orders are delivered and picked up, and the quantity to produce during the pitch(see “Learning to See” for a description of pitch time). This is a mechanism that supports the leveling process. The pacesetter has a clear understanding throughout the day whether he or she is ahead or behind.

Figure 7-2. Future state value stream map with elements identified

If the value stream pacesetter follows this schedule, what happens? The pacesetter will consume the components necessary to complete the task and “withdraw” them from the supermarket upstream. Since the pacesetter is leveled, this withdrawal will also be leveled. For example, say there are three different components used for assembly at the pacesetter—call them A, B, and C—and each is used for a different end product. If the assembly of end products is leveled, the consumption of A, B, and C will be leveled. That is, there will be a smooth rotation among the consumption of A, B, and C. This allows for keeping the minimum amount of inventory of A, B, and C in the supermarket. In contrast, if the assemblers suddenly spend an entire day just using part A and the supplier had put just a part of the day’s worth of part A in the supermarket, the assembly would run out of A and shut down. So once the system is set up to be leveled, it’s critical that the leveling process is actually followed, or you will run out of parts. When production is initiated to replenish the component supermarket, the process withdraws raw materials from the supermarket, which signals the supplier of the need for replenishment. Again, if the pacesetter is leveled, then the signals to the supplier will also be leveled, mitigating the infamous bullwhip effect in which the customer plant makes changes in schedules for its convenience only to jerk around suppliers in violent waves. With leveling, suppliers will have a good idea of what is expected of them and be able to plan with confidence. They can now balance resources to a known takt and get lean by improving quality and operating at lower cost.

We often hear companies say we cannot be level because our customers are not level. The leveled “schedule” for the first flow loop is created by production control even when the customer is not level. Note that production control has two sources of information to create the leveled schedule. There is a direct arrow from the customer—the build-to-order signal—and a second arrow from the finished goods supermarket—the build-to-stock signal. In lean systems this is a common way to handle high-variety product mixes. The relatively high-volume products that you know customers will buy are built to stock—kept in the supermarket and replenished as they are shipped to the customer using a kanbantype system. The lower-variety, less predictable products are built to customer order. Production control sees the stream of real customer orders coming in and the kanban orders from the supermarket. Typically there is a third stream of safety buffer stock that can be replenished if there are not enough real or kanban orders to fulfill in a day. Through this combination of orders, production control has the tools to create a leveled schedule.

There is no need for additional external scheduling or planning beyond this one scheduling point. For the build-to-stock items, the needs of the customer (represented by the supermarket) are visible to everyone. The kanban are used to represent the inventory position and are effectively used to control the correct quantities. Kanban can be placed on a board, and visually represent an inverse relationship of the inventory—each kanban represents a reduced level of inventory. Build-to-order items can also be placed on the board so it’s clear what is being built to a real customer order, to replenish the supermarket, and to replenish safety stock. Setting priorities becomes visual and straightforward. When Toyota says, “Operators can schedule their own work,” this is what they mean. The operators are not performing traditional planning and scheduling—predicting what should be produced and when—they are simply using the information that flows to them from the visual system and a defined process governs the decision making.

How to Establish a Basic Leveled Schedule

Getting to a true heijunka schedule with a steady pitch multiple times in the day is what we would consider an advanced lean practice. Some minimum amount of leveling is needed in the stability phase (see Chapter 4) to even establish a basis for calculating a takt time and setting up basic flow. During the initial stages the pitch time is generally larger, often a daily time window, which creates a basis for stability, but it is not an impossible challenge. Attempting a smaller pitch prematurely may surface too many problems and create a system that is impossible to maintain.

In addition to the pitch time increment, the three aspects that will be leveled are:

1. Product volume, which is simply the quantity of a given product that must be produced in a specified period of time (the pitch).
2. Product mix, which is the proportion of the various models that are produced during the pitch increment, the quantity of A’s, B’s, C’s, and so
3. Product sequence, which is the order that the product volume and mix are It may be model by model, such as A, A, A, B, B, B, C, C, C, or part by part, such as A, C, A, B, A, C.

These three are listed in order of difficulty. Depending upon your starting point, you may need to begin by establishing a simple volume and mix leveling on a larger pitch time such as one shift or one day. We know that everyone is touting single-piece flow and sequenced heijunka as the epitome of lean, but that objective may be far off, depending upon the current condition of your

facility. After all, it has taken Toyota 50 years to achieve their current success, and in many cases they’re still striving to reach the epitome. The key is to stretch enough to make a great improvement, and to challenge your capabilities, but not so much that total failure results.

TIP: Identify the Most Important Items for the Greatest Benefit

It may not be practical to level all products, due to extremely low or sporadic demand of some items. Before beginning the analysis to identify specific products to level, it may be necessary to isolate variation (see Chapter 3) or to utilize an isolation technique we call “slice and dice,” which is discussed later in this chapter. Identify key products in key areas and begin with those that will provide the greatest benefit.

Begin with a review of the actual production or sales for each specific product over the previous 12-month period. This will provide a high, low, and average volume demand. The actual numbers can be plotted on a graph to get a visual representation, which is better than the plain numbers because it’s possible to see the “weighted average.” Simple highs and lows represent peaks, and a few peaks may skew the average. Plotting the actual numbers on the graph allows you to use your eye to see the most appropriate leveling point.

The final decision of the level volume is somewhat subjective. In general, Toyota selects a number that is approximately 80 percent of peak demand (unless the peak was an isolated event) because the gap between 80 and 100 percent could be filled using overtime (eight hours per week). The determination of leveled demand will be used to calculate takt time. In the previous chapter, we discussed the use of takt time as a design parameter. When determining the leveled demand quantity it is better to err on the side of a slightly higher demand if you are uncertain or uncomfortable with the 80 percent level. In reality, when you determine a quantity to serve as your assumed daily level demand it will either be too high, too low, or just right (not as likely). The problem is, it’s difficult to determine initially because of the variation that has been occurring in the production (a cloud). Once the process is stabilized, the clouds clear a bit and the correct level will become more evident and adjustments easily made.

TRAP: Avoid Analysis Paralysis

It’s easy to slip into “analysis paralysis” at this point if you try to determine the perfect leveling point. There are a few factors at work here that make a perfect selection virtually impossible. First, as they say about mutual funds, “Past performance is not a guarantee of future performance.” We are basing future plans on past results, but they will not be the same. Second, the law of large numbers means that the more data points are observed, the less influence any one point has on the overall total. When looking at yearly totals for production volumes, a random spike here and there has less effect on the overall average. In laymen’s terms this simply means looking at a large enough sample, so the “noise” in the data is filtered out. Third, the information you’re analyzing may be flawed. It may not show the actual demand, but rather, the orders that are generated internally by an MRP system to fulfill “demand.” These order quantities are influenced by many factors, and the quantities do not necessarily reflect true demand. Finally, as you’ll see below, when you attempt to level the entire product mix, there will have to be some slight adjustments made to achieve an effective balance. Our tip is to select a level volume for each item that seems to be correct and get started on leveling the process. We guarantee that you’ll need to make adjustments no matter how carefully you do the analysis!

TIP: Identify the Multiple for the Level Pattern

The best pattern is based off of a multiple of two. This provides a consistent pattern of daily, every other day, every fourth day, and at most every sixth day. If the volume of an item is such that the daily demand times six is still too low to be practical, you either need to reduce setup times or shift the item to the “other” category until setup times are reduced. In our example, the pattern of every other day was established for the items that had roughly one-half the demand of the every day items, and the every four-day items had roughly one-fourth the demand of the daily items.

The first pass of leveling will remove a layer of waste associated with chasing the waves. This will provide additional capacity that was not available before. Many companies discover that the initial leveling effort allows them to “catch up” with orders, and that they are overproducing based on the initial assumptions. It is possible to either reduce the resources or to increase sales if possible.

Let’s look at a specific example. The data in Table 7-1 represents a simplified version of a real situation, but the concepts can be applied to more complex situations as well. In our example we will level 10 parts, designated A through J, each with varying demands. The “Other” items that are produced in the process had low volume requirements, an average of 125, and will not be leveled by individual part. The total daily volume for all products, including the “Other” items, is leveled. In fact the “Other” items and the quantities will vary, and it is possible to make adjustments by increasing or decreasing the total running time if the actual requirement is more or less than planned. This adjustment does not alter the leveling effect for items A through J.

Based on the volume requirements for the leveled items, a production pattern is developed to minimize the negative effects of changeover (the process has improved, but the time is still greater than desired—for now). Items A through C are produced every day (ED), and items D through F are produced every other day (EOD). Items G through J are produced every four days (E4D— yes, yes, we know the goal should be to produce every part every day, but we are not there yet!).

PartDaily Average Demand A 250 B 220 C 210 D 128 E 125 F 75 G 60 H 45 I 45 J 35 Other 125 Total 1,318

Table 7-1. Volume Requirements by Part

One potential pattern is shown in Table 7-2. The daily requirement of 1,318 was adjusted slightly to 1,325 just to round the numbers. As we said, this is irrelevant because there is variation in the quantity of the “Other” items. This pattern is more evenly spread and allows for production of some “Other” items daily, but on some days the quantity of the “Other” items is low. If average order sizes of “Other” items are typically greater than these amounts, another pattern may be considered.

Table 7-3 shows an alternative pattern that groups more of the ED and EOD items on the same day. The ED items are a given—they run every day. The EOD and E4D may be changed to suit the needs of the process. For example, the EOD items could all be produced on the same alternating days as in this example. There are other potential patterns as well. The objective is to achieve the best level volume across the pattern by row—leveled by individual item over a time window, and down the pattern by column—total volume, and mix per time increment (pitch). The production sequence is defined by following the patterns (A through J) in the order specified. The level across the time period is within a defined repeating increment. In our case, the pattern has a four-day repeating sequence and each item is leveled (the totals are equal) every four days. Toyota typically uses a monthly window for leveling, but it is based on a repeating multiple of one day. Note: The main vehicle assembly line has a repeating pattern on a short pitch frequency depending on the particular mix of vehicles produced, but the supporting operations that are producing to a supermarket or “selectivity bank” are producing to a different pattern that is a derivative of the primary pattern.

Day 1Day 2Day 3Day 4Day 5Day 6Day 7Day 8 A (ED) 250 250 250 250 250 250 250 250 B (ED) 220 220 220 220 220 220 220 220 C (ED) 210 210 210 210 210 210 210 210 D (EOD) 256 0 256 0 256 0 256 0 E (EOD) 0 250 0 250 0 250 0 250 F (EOD) 150 0 150 0 150 0 150 0 G (E4D) 0 240 0 0 0 240 0 0 H (E4D) 0 0 0 180 0 0 0 180 I (E4D) 180 0 0 0 180 0 0 0 J (E4D) 0 0 140 0 0 0 140 0 Other 59 155 99 215 59 155 99 215 Total 1,325 1,325 1,325 1,325 1,325 1,325 1,325 1,325 Goal 1,325 1,325 1,325 1,325 1,325 1,325 1,325 1,325

ED = every day; EOD = every other day; E4D = every four days.

Table 7-2. Possible Leveled Production Pattern

Day 1Day 2Day 3Day 4Day 5Day 6Day 7Day 8 A (ED) 250 250 250 250 250 250 250 250 B (ED) 220 220 220 220 220 220 220 220 C (ED) 210 210 210 210 210 210 210 210 D (EOD) 256 0 256 0 256 0 256 0 E (EOD) 250 250 250 250 F (EOD) 150 0 150 0 150 0 150 0 G (E4D) 0 240 0 0 0 240 0 0 H (E4D) 0 0 0 180 0 0 0 180 I (E4D) 180 0 0 0 180 0 0 J (E4D) 0 0 0 140 0 0 0 140 Other 0 225 0 325 0 225 0 325 Total 1,336 1,325 1,336 1,325 1,336 1,325 1,336 1,325 Goal 1,325 1,325 1,325 1,325 1,325 1,325 1,325 1,325

ED = every day; EOD = every other day; E4D = every four days.

Table 7-3. Alternative Leveled Production Pattern

Notice that in the alternative pattern the total in days one, three, five, and seven exceeds the daily goal. This is not a major problem since the amount is within reasonable limits (normally a maximum of 10 percent). In most cases when working with actual demands, the numbers don’t work out as evenly as this example. For the first attempt, get the numbers as close as possible. After you’ve had the opportunity to produce based on a level schedule, you will gain a clearer understanding of the true need and will adjust the pattern accordingly. It is much easier to calculate a leveled schedule than to actually produce according to the plan! At first it’s likely that you will discover many obstacles that prevent adherence to the schedule. These obstacles need to be systematically identified and corrected so stability can be achieved (track causes for missing the heijunka, and use the problem-solving method to eliminate them). The leveled schedule should now be considered the “voice of the customer.” It is not the true customer, but a defined agreement that represents the needs of the customer that have been smoothed for the benefit of your processes.

Since this is the “customer,” you should measure and track your ability to satisfy the customer. If at any time you are unable to achieve the volume, mix, or sequence that has been defined, it is equivalent to a “missed order” (and represents a dissatisfied customer, although you may not miss an actual order). You must train people to consider the heijunka as the voice of the customer and as a primary objective of the value stream.

Incremental Leveling and Advanced Heijunka

Congratulations! Having gotten to this stage in your lean journey, you’re ready for the real fun to begin. After processes are stabilized and connected, there is value stream flow, and improvements are standardized, you now begin the continuous improvement cycle. That’s right, you get to go through it all again, and again, and again, forever. The good news is that each successive loop through the continuous improvement spiral will be somewhat easier, since much of the foundational learning has been done and resistance to change overcome. Any changes made from now on will yield direct benefits for the entire process. In other words, instead of “pocket” improvements that do not affect the overall result, improvements now will influence the outcome of the entire value stream.

Now the bad news. From here on, the improvement process is a continuous cycle of “tightening” and refining the operations to achieve shorter lead times and greater degrees of flexibility and capability, push inventory levels down, and strengthen the long-term position of the business. Now, the results will be incremental in nature; that is, they will be of a predetermined amount because change to standardized processes can occur within a defined portion. Because of the system that has been created, the desired outcome is identified and the result will be assured.

The method will stress the value stream, and the weakest link will snap, creating instability. When the weak link is detected, resources are gathered to attack the issues. This cycle repeats over and over as shown in the continuous improvement spiral model in Chapter 3 (Figure 3-4). Each successive cycle uncovers decreasingly smaller problems. So it’s a good news/bad news scenario. The bad news is that the issues become more difficult to correct. The good news is that improvements in the process will be significant and your skill level will grow as the difficulty of issues increases.

Incremental Leveling

After the value stream is connected, the incremental tightening process is applied to specific points. Remember what happens to the value stream if the production rate of the pacesetter is changed? It establishes a new rate for every other process in the value stream. Now, if the leveled schedule product mix were adjusted, all processes would need to adjust to support the new mix.

This type of incremental leveling or squeezing of the value stream forces the improvement process. It’s a planned and controlled process that will incrementally drive continuous improvement in a specific manner. If inventory in the supermarket is reduced, for example, the effect on the supplying processes should be predictable. This may force a changeover more frequently, which forces the need for shortened changeover times. Each change in a standard element of the value stream will force the need for improvement and create a specific and predetermined result.

Points of Control

Within a connected value stream there are specific “points of control” that will influence other processes in the value stream. Because of the connected nature of the value stream, an adjustment to the point of control will require an adjustment to all processes that supply the control point. And since the point of control is the primary operation within the value stream that must be closely managed in order to create consistent output of the value stream, managing it allows you to effectively understand how to maximize the entire value stream.

One key point of control is the leveled schedule. It provides a standardized core that is used to establish takt time. The pacesetter process uses this takt time to establish a beat that will be followed by all other operations. Understanding the point of control allows managers to effectively troubleshoot operations and drive continuous improvement.

If the pacesetter consistently produces the desired volume of product and is capable of producing the correct product mix and sequence according to the leveled schedule, the value stream is consistently meeting the customer requirement (the next step would be cost reduction). If, however, the pacesetter is unable to fulfill the requirement of the leveled schedule, the first place to stand in the circle is at the pacesetter. From this vantage point it is possible to evaluate whether the pacesetter is being supplied properly. If not, look upstream to find the weak link. If so, look at the pacesetter to determine if he or she is being blocked by a downstream operation. (The rules forbid overproduction, so if a downstream process is blocking the pacesetter, it will be visibly evident.) The creation of visible connections allows quick identification of flow stoppages, simplifying management of the value stream.

Point of Control for Managing Inventory

The point of control for inventory management is the kanban. Reducing the number of kanbans within the system will reduce total inventory quantity.

TRAP: Use Inventory Reduction as a Yardstick for Success, Not a Goal

Many people pursue inventory reduction as a primary goal of lean activities. There are numerous ways to achieve this goal, including manipulation of the inventory. It is better to establish a goal to create connected flow and to use inventory as a measure of success. Kanban are used to control inventory, and it’s simple to measure the effectiveness of the process by regulating the kanban. Inventory control via kanban is standardized, and the possibility of false manipulation is reduced.

These reductions should be done systematically, either as improvements are made to the process or to force the need for improvements. The quantity of inventory needed to support a process can be used as a yardstick for your improvement efforts. Sustainable inventory reductions are an indication of a capable process.

Inventory turns can also be influenced by the kanban. If the part quantity per kanban (also the container quantity) is reduced, the kanban will “cycle” more frequently, moving inventory through the process at a faster rate. Reducing the quantity per kanban also provides a greater degree of flexibility in the replenishment process, and reduces the size of the work area and waste. Strange as it seems, having more kanban “in the system” is an advantage. For example, if the total inventory level of an item is 2,000 pieces, it’s better to have 20 kanban of 100 pieces each than two kanbans of 1,000 pieces each. It’s difficult seeing the demand with only two kanban in the system, and each time a kanban is returned, it must be filled immediately.

A Leveled Schedule Dictates Replenishment

In addition to the smoothing effect for all processes, heijunka establishes a “pitch” time. Because materials are being consumed at a standard rate during a defined pitch time, it’s possible to establish a defined process for material replenishment. Material replenishment is subordinate to the primary value adding operation; therefore, establishment of material replenishment “routes” or methods should not be attempted before creating a standardized “core” in the primary process.

The following example illustrates how a leveled schedule dictates the material replenishment needs and establishes a consistent requirement. This allows the standardization of work for material handlers, including routes that are completed during the pitch time or a multiple of the pitch. Material quantities are standardized, and the quantities per container may be adjusted to match the requirement per pitch. For illustrative purposes we assume that this process is capable of advanced heijunka and produces each item in the exact sequence, and that the total available work time is eight hours. Demand is 400 total pieces, and the ratios are shown in Table 7-4.

Part NameQuantity per 8 hoursRatio A 200 4 B 100 2 C 50 1 D 50 1 Total 400

Table 7-4. Quantity of Parts as a Relative Ratio

TIP: Set Your Pitch Based on Current Conditions

Unless you are well along in your lean journey, you will not likely set a one-hour pitch initially. We recommend progressing in halves. If you currently move material at a daily pitch (or it is not defined), start with a shift-by-shift pitch. Then incrementally reduce the pitch by one-half as the processes become more capable and refined.

Based on the quantity required and the ratios, the repeating heijunka pattern (which minimizes batching) would be: ABACABAD—ABACABAD— ABACABAD

The pitch time to repeat the pattern is determined by dividing eight hours by the demand of 400 pieces and multiplying by the number of pieces in a pattern (pitch):

28,800 seconds (eight hours) per day / 400 pieces = 72 seconds per piece And:

72 seconds per piece x 8 pieces per pitch =

576 seconds per pitch (9 minutes 36 seconds) or 6.25 pitch-cycles per hour.

Part Name

Ratio

Patterns per Hour

Hourly Requirement

Quantity per Container

Containers per Pitch

A 4 6.25 4 × 6.25 = 25 10 2.5 B 2 6.25 2 × 6.25 = 12.5 5 2.5 C 1 6.25 1 × 6.25 = 6.25 5 1.25 D 1 6.25 1 × 6.25 = 6.25 5 1.25

Table 7-5. Calculation of Containers Moved per Pitch

Let’s also assume that we want the material handler to move material every hour (the pitch for material replenishment). Table 7-5 shows the calculation of the number of containers that will be moved during each one-hour material replenishment cycle.

Based on the material movement requirement during a one-hour cycle time, it is possible to define standardized work, including the specific route of travel and other processes that will be serviced during the route.

Slice and Dice When Product Variety Is High

Heijunka seems straightforward enough when you are dealing with 5 to 10 products. But what happens when there are many different finished products? One company claimed to have 25,000 individual finished goods part numbers and insisted heijunka was impossible. How would it be possible to level with this kind of variety? We have to go through a process we call “slice and dice,” which is a method of dividing the whole into groups of products with similar characteristics (you may also think of this as “divide and conquer”).

The first “slice” separates products into value streams that have common products and processing steps. This grouping puts like items together and also reduces the overall number of items within the slice—the 25,000 may now only be 5,000. Think of your operation with the variety of products and processes in its entirety as a rectangle. The separation into value stream “families” with common characteristics and processing steps would divide the rectangle horizontally into slices (Figure 7-3). If the most important value stream overall is addressed first, the greatest benefit will be achieved from the effort.

If the slice is “diced” (Figure 7-4), the most significant items within the 5,000 are isolated, and the primary focus is reduced further. The “dicing” of the value stream also includes the selection of a portion of the stream (processes) to focus on during the initial lean efforts. Usually we look at the production volumes of all products to determine where the dices should be. When the product mix is diced in this manner we invariably find that the volumes fall into three groups: a top group with the most significant demand; a second group roughly one-half the volume of the first group; and a third group, one-half again lower than the second group (the volumes in the leveling example above represent a typical example). Generally, the first group is relatively small in terms of the quantity of part numbers but large as a percentage of total volume. (If you are thinking that this is the Pareto principle in action, you’re exactly right. This method allows you to isolate the “significant few” from the “trivial many.”)

Value Stream Family A Value Stream Family B Value Stream Family C Value Stream Family D Value Stream Family E Value Stream Family F Value Stream Family G

Figure 7-3. The operation sliced into value streams

Process 1

Figure 7-4. A value stream slice with a section diced

We began with 25,000 part numbers. The top 100 part numbers in terms of volume accounted for 35 percent of the total sales volume for the company! That is a significant reduction. An additional slice revealed that the volume for the number one item was 10 times greater than the fiftieth item. It was decided to focus on leveling production for the top 50 part numbers (out of 25,000). While looping through the continuous improvement spiral, we work on specific segments or layers (slices), and each successive pass through the cycle brings the addition of subsequent layers. After the first 50 parts are successfully leveled and the value stream is performing consistently, the next 50 parts (or more) will be initiated.

With the focus quantity reduced to 50, the magnitude of the effort is minimized and the benefit is maximized. Many people incorrectly assume that if it’s not possible to level everything, it’s not possible to implement heijunka. In reality, the question is a matter of simple math. Is it better to be stabilized zero percent of the time on 100 percent of the items, or to be stabilized 100 percent of the time on 25 percent of the items? This is not an all or nothing proposition.

As your operations develop greater capabilities, it becomes possible to consider leveling smaller and smaller quantities. It may never make sense to level all items. Consider the slice and dice: If 75 percent of all items are leveled—and therefore 75 percent of the total resource needs are leveled—the remaining 25 percent of the resource time (people and equipment) can be devoted to the “as needed” items. The raw materials may be shared between the leveled and nonleveled items, and the additional need can easily be factored into the material replenishment calculations.

Case Example: Leveling Workload in a Custom Cabinet Shop

The workload required at various operations in this company fluctuated greatly, depending on the product, which caused many problems, including poor quality (workers were frequently rushed), line stoppage, and unpredictable production schedules. Because of the custom nature of the product, it was assumed that standardized work for the processes was not possible.

When dealing with a situation of this nature, the apparent complexity can be overwhelming. There were many interconnected and interrelated issues resulting from the ripple effect of the workload (imagine the snake that eats the rat, and the bulge proceeds down the length of its body). As is often the case, the company had attempted to address the outlying issues (where the “problem” was realized), creating elaborate schemes to shift labor to the bulge, but the problem originated at the core. Intuitively, they understood this, but believed it was impossible to change because every item produced was different and the size of each order and the mix of components (cabinets, doors, drawers) varied significantly. They assumed that customers dictated the schedule and there was nothing they could do to level the workload.

The first step was to stop looking at the product as “part specific” or “job specific” and to look at it based on the work content and the effect that content had on the processes within the value stream. If you step back a bit, you can see commonalities either in the product itself or in the processing. In this case, we first identified that most “jobs” or orders had some common elements that affected the workload. The primary components were: cabinets, drawers, shelves, doors, miscellaneous parts, and trim. We also determined that there were a few characteristics common to all products that had an effect on the workload, primarily the type of finish. The finishes were in two categories: stains and solid colors. Further discussion revealed that within the two finish categories each had two additional separations. The stain colors had a burnished and unburnished option, and the solid colors were light and dark.

A review of the value stream revealed that the finishing line where product is stained or colored was the “pacesetter.” All products converged at the finish loading area and from that operation flowed on as a complete order. Leveling the workload at the pacesetter would serve to create a smooth workload to subsequent operations (including the finish operation) and provide leveled signals to all upstream feeder operations.

Again the question surfaced: How do you level the workload when the product is always different? By standing on the circle the answer was clear. The finish type, and the surface area to be finished, affected the workload. Workers confirmed that burnished stain jobs required much more effort than unburnished ones, and that dark-colored solid jobs were much harder than light ones because the solid colors had a “polished” finish. We also saw that parts with larger surface area required more time, as did many small parts with less surface area. It was becoming clear that creating a sequenced pattern with leveled mix would be the answer. But, again the question: How do you do this when every job is different?

This group, and especially the supervisor who had struggled with the issue for years, was not easy to convince. What we needed was a variable standard; that is, we would develop a standard with an

allowance for variation. The variation method would be defined so it was consistent (standardized variation).

Analysis of the production data revealed the ratio of finishes was 75 percent stained to 25 percent solid colors. The burnished jobs (the more difficult ones) accounted for approximately 25 percent of the total stain jobs. For solid colors, the split was nearly even, with slightly more lightcolored (easy) jobs. This allowed us to create a primary leveling factor for establishing mix based on the ratios of finish colors and types. Since the actual daily mix did not necessarily match the averages, there were secondary conditions added to the pattern. For example, the regular pattern was:

STU, STU, SOLL, STU, STU, STUB, STU, SOLD, STU, STU

STU = Stain, unburnished STUB = Stain, burnished SOLL = Solid, light color SOLD = Solid, dark color

But because the workload for solid light color and unburnished stain was similar, they could be substituted on the pattern. The goal was to create as consistent a workload as possible, while processing the correct ratio of each type job.

The second layer of the pattern was the individual components. The team identified that the trim work should always be the first item of any job because of the special processing needed. The small parts went at the end of a job because they tended to have a low workload and provided a “spacer” between jobs to allow for color changes, etc. In addition, two empty racks were sent through between jobs to provide an empty zone to prevent overspray from job to job. A pattern was developed that adequately mixed the size and surface area combinations of each job. Like the color application, some of the categories were similar and could be substituted as defined (the standardized variation).

The pattern for components was: trim—cabinet—doors—cabinet— drawers—shelves—doors—cabinets—drawers/doors—repeat as needed— miscellaneous parts—space—space (next job) trim . . .

Secondary rules were established based on the finish type (because of workload). For example, cabinets were placed two to a rack if small, and one if large (or one only of any size for burnished and dark fin ish). Doors were six to a rack for unburnished and light colors, four to a rack for burnished and dark colors. The same logic was applied to drawers and shelves.

In this case example, the production volume was difficult to define. The number of pieces, racks, and jobs all had variation. The company

had a goal for 25 jobs per day, so we set that as the volume target, even though the total amount of work varied. This variation, however, was handled by slight adjustment of the total work time each day and did not affect the workload balance throughout the day. The mix included two layers—the primary mix based on finish, and the secondary mix of components. The primary determination based on finish provided the correct mix to meet customer orders and workload, and the secondary provided the correct mix for workload. Sequence of the orders by finish helped to balance the workload, as did sequencing the components.

These changes were the foundation for establishing standardized work and flow. Balancing the workload reduced the amount of line stoppage and smoothed flow throughout the rest of the operations. Future activities that connected operations reduced the “pile-ups” that frequently occurred.

In a custom environment, it’s difficult to find an accurate measurement for performance. There is always an element of variation that will skew any measure. In this case, a longer view had to be developed, with the idea that over a wider time window (one month) the variation would be equalized. In other words, month over month we could begin to see improvement in performance as measured by total hours required versus total sales dollars. When performance was viewed over a sixmonth period, the variation was equalized even further, and there was a noticeable mean shift.

Leveling Is an Enterprisewide Process

The single most common reaction we get when we try to teach leveling to companies is: “Sales has their own incentives, and sales always comes first in this company. They sell whatever they can, and we in manufacturing are expected to build it, and sales can change by 100 percent or more from week to week.” When we’ve examined the data more carefully, we typically find that actual demand is much smoother than what manufacturing sees.

In one office furniture manufacturing company that built a large variety of different filing cabinets, customer orders to the plant were unstable. Yet the corporate policy was 100 percent build to order, and manufacturing was constantly fighting fires to build whatever orders came in. This led to huge amounts of inventory at every stage of production and no clear takt time anyplace in the process. When asked what lead times they gave customers who ordered filing cabinets, the answer was six to eight weeks out. So the manufacturing plant was working like crazy to build orders that were all over the place, but there were six to eight weeks of nonvalue activities in the pipeline. Why couldn’t that time buffer be used to level the schedule? If there was inventory of finished filing cabinets, at least for the high-volume cabinets, that six-to-eight-week lead time could be reduced and level the schedule, creating a more efficient process. In fact, the plant reorganized around three product family value streams, used some finished goods inventory to level the schedule, freed up one-fourth of the plant for new business, and dramatically reduced overall inventory, lead time, and total cost.

To accomplish what appears to be a logical plan is not as easy as it sounds. The furniture manufacturer had to change the way sales people placed orders. They had to change the distribution process and the way production control scheduled the plant. These are all governed by different functional groups who had been doing things a certain way for decades. Nobody believed the new system could possibly work, and all predicted disaster. Overcoming this resistance required a strong vision of a future state and a lot of top management support.

Frequently, sales groups work to incentives based on sales targets by month or quarter. Such incentive systems lead to lumpy sales patterns with serious discounts to move product at the end of the bonus period. At Toyota, sales is aware of the importance of a leveled schedule in production. While, even at Toyota, production often complains about what sales does to them, there is a lot more cooperation than we typically see in other companies. This cooperation is encouraged by top management who understand the implications of sales patterns on the leveled schedule that is the foundation of TPS.

Thinking in systems terms and enterprise terms is just plain hard. And learning to think in value stream terms is the most critical in leveling the schedule— the foundation of lean systems.

Case Study: Leveling the Schedule in an Engineering Organization

Most knowledge work is inherently lumpy. And you cannot parcel out a schedule in units the way you can in a manufacturing process.

Nonetheless, Toyota has found a way to level the workload in engineering new products to a far greater degree than its competitors.

First, you have to get some basic stability in the process. Toyota has developed a stable development process in which there are clear stages, and each one takes a standard amount of time and engineering hours.

Second, this allows Toyota to set up a planned schedule at the beginning of the program and stick to it. Roughly, Toyota freshens cars every two years and issues a major new version about every four years.

Knowing this, not all cars are completely overhauled in the same way. This is spread out so that roughly one-quarter of the launches are overhauled in one year.

Third, within a vehicle program, Toyota has a clear profile of manpower over the program. The program definition phase starts off with a small number of senior engineers, the program ramps up to a peak and then comes back down to a relatively small number of engineers through launch. Again, this is based on the stability Toyota has in the process. Many of its competitors send an army of people to the plant when they launch. Toyota has such a well-planned process and does enough high-quality engineering in the concept stage that its launches are smooth and most engineers are on to another program.

Fourth, Toyota takes care of the peak of the program by drawing on its affiliates. This includes closely linked contract firms that provide technicians and computer-aided design specialists at peak times. It also includes affiliated companies like suppliers and Toyota Auto Body, which send engineers at the peaks. This allows Toyota to keep the core engineers on staff and bring in the rest flexibly. Standardized design processes and designs help Toyota engineers and affiliates come in and out of the program and contribute seamlessly.

Fifth, Toyota staggers the release of a lot of engineering information. For example, its competitors often provide a batch of all body data released at once to die engineers who then process all of this data into the design of stamping dies. Toyota releases body data as parts are developed and released directly to die design, which releases data as it goes to die making. There is a clear understanding of what body parts can be released early, before the rest are complete. This creates something like a onepiece flow and is much more level than releasing large batches of part designs.

Reflect and Learn from the Process

Basic leveling of production volume and model mix is necessary to establish process stability and continuous flow. Using your current state value stream map as a guide, identify the operations that continue to struggle with meeting the expectation.

1. Are these operations being affected by external customer variation?
a. Does the daily requirement change?
b. Determine the extent of the fluctuation (show the daily demand on a line graph). Variation of greater than 10 percent must be reduced.
c. Identify current methods for aligning resources (people, material, machines) to these fluctuations, and your effectiveness in meeting the requirement (measurements of efficiency and customer delivery).

2. Establishing a “level schedule” requires up-front effort, and diligence to sustain.
a. Evaluate the effect of the variation and decide whether leveling the product flow would be beneficial.
b. Are you willing to make the effort to eliminate problems that currently prevent you from producing smaller quantities more frequently and consistently?

3. If you’re producing a product to stock, establish a finished goods supermarket to absorb the true customer variation.
a. Determine the average daily volume demand for your products.
b. Determine a pitch time for each product. The highest 10 to 20 percent of products by volume (maybe more) should be set for daily production.
c. Determine the repeating time pitch for the other products, and create a “pattern” in which to produce the product. Consider the mix of products required and the sequence to produce them for balanced flow.

4. Your leveled schedule becomes a standard for operation. Measure your effectiveness in achieving the standard and correct obstacles that prevent consistent ability to achieve the schedule. Note: Do not change the plan because the process is not capable. Correct the weakness.

5. As a process continues to achieve higher levels of capability, it is necessary to incrementally raise the bar. Evaluate your value stream, and reflect on the following questions:
a. Do you know where the “point of control” is within your value stream?
b. Are you measuring and managing the point of control?
c. What changes at the point of control will impact the entire value stream?
d. How will these changes affect the value stream (where will the chain break)?
e. Can you implement corrective action to the weakness in the value stream prior to forcing the change?

6. Leveling is necessary to provide a “standard core” to which all resources are aligned. Build these additional elements based on your level schedule process:
a. Material replenishment: All material supply within the facility is based on the consistent requirement at each process. This dictates the material replenishment pitch and is the basis for a replenishment strategy, including consistent replenishment from suppliers.
b. People: The level schedule becomes the basis for determining takt time, which is necessary for standardized work. Establish standardized work for all processes, and determine the required number of people.
c. Equipment: Standardized work is also the basis for equipment requirements. Align the required equipment to the people and work based on the level schedule.