[Return to papers] [Return to home page]

Reducing Flow Time in Aircraft Manufacturing

Jackson S. Chao
Boeing Company
P. O. Box 3707, MS 03-26, Seattle WA 98121 2207

Stephen C. Graves
A. P. Sloan School of Management
MIT, E40-439, Cambridge MA 02139

The assembly of aircraft is a labor-intensive process that exhibits a significant learning-curve effect and that requires long flow times and costly work-in-process inventories. This paper describes the production context, the cost of flow time in this context, and some of the causes for the long flow times. We then develop an argument for a firm to use improvements in labor productivity to reduce flow times. Boeing has implemented the recommendations from this research and has obtained significant benefits from reducing flow times.

October 1992, revised February 1994, January 1996, January 1997

The authors wish to acknowledge MIT’s Leaders for Manufacturing program for their overall support of this research, and to thank Boeing for creating the opportunity. Special thanks go to David Fitzpatrick and Fred Farnsworth at Boeing, for their insights, knowledge and support throughout the conduct of this research. We would also like to thank Al Drake and Tom Kochan from MIT for their support and guidance. Finally we thank the referees for their helpful comments and advice, and Kal Singhal for directing us to the line-of-balance reference.


[Return to papers] [Return to home page]


In this paper we report the results from an internship performed by the first co-author at The Boeing Company, the world’s most successful airplane manufacturer. The internship was conducted as part of MIT’s Leaders for Manufacturing Program, and ran from June 1990 through December 1990 in the New Airplane Division (now known as the Boeing 777 Division). The charge for the internship was to study the final assembly process of another Boeing airplane to discover lessons from its manufacturing and to make specific recommendations for the 777 program.

After learning about aircraft manufacturing the project quickly focused on manufacturing flow time, and the questions of what does it cost, why is it so long, and what needs to happen to affect it. Doing things fast is a common theme in the best manufacturing practices and is advocated at length in the manufacturing literature, e. g., Dertouzos et al. (1989), Goldratt and Fox (1986), Hayes et al. (1988), Schmenner (1988), Stalk and Hout (1990). This research examines how these ideas apply to aircraft manufacturing and provides a case study for addressing flow-time issues.

The main lessons from the research are (i) the importance of recognizing and quantifying the cost of flow time in aircraft manufacturing; (ii) the consideration of reducing flow times, rather than head count, to realize the productivity improvements from the learning curve; (iii) the differences in impact from how flow-time reductions are enacted, i. e., whether a reduction is pushed back or pulled through the production process; and (iv) the value from a data-driven examination of the impact of system variances on labor content and flow time.

We believe this study is a good example of action-oriented research which considers real operations with real problems, and attempts to develop, extend and apply new concepts. From this research, we can extract manufacturing principles that may be of generic value to other manufacturing contexts. Finally, this research can lead to significant, measurable impact for the company under study.

The rest of the paper is organized into four sections. We first describe the organization and planning for the manufacture of an aircraft. Next we discuss the costs of flow time and show how to quantify these costs; we propose a strategy for flow-time reduction based on productivity improvements from learning. We then present regression analyses that relate system variances to the direct labor content and discuss the implications for a longer-term strategy for flow-time reduction. We conclude with the impact on Boeing from this work.

[Back to top]


In this section, we describe some of the key concepts in the planning of airplane manufacturing. Specifically, we describe the organization of the manufacturing processes for airplane assembly and the methods for planning and scheduling the assembly operations. At a company like Boeing, planning is an enormously complex activity and involves hundreds of people with thousands of person-years of experience. Needless to say, we will at best give a high-level overview of the methodology, concepts and planning tools.


The assembly of an airplane entails a synchronized series of manufacturing processes which are organized as a network of concurrent and merging flows. These manufacturing processes are organized into a network of work centers or departments, known as control codes at Boeing. These work centers, staffed with varying numbers of line employees, have responsibility to perform pre-assigned tasks within the manufacturing process. The operations performed by these work centers vary from tasks as simple as finishing the surface of an airplane wing to tasks as complex as integrating the major body sections of the entire airplane. For example, a work center might be responsible for joining the completed left and right wings to the wing stub section of the airplane fuselage (wing-stub join).

The manufacturing flow time for a work center is the elapsed time (in work days) planned for a work center to perform and complete its required tasks. The flow time is the length of time that an airplane (or subassembly) will remain at a work center before moving to the next work center. Different work centers within the manufacturing sequence can have (and will have) different flow times.

The production cycle time is the elapsed time (in work days) between consecutive job completions for a work center or between airplane deliveries for the entire manufacturing system. Unlike manufacturing flow time, all work centers within the manufacturing system operate with the same production cycle time (i.e., at the same rate). An airplane manufacturer operating at a three-day production cycle completes and ships an airplane from the production line every three days. Consequently, every work center must also complete work on an airplane every three days, no matter what the individual flow time of the work center is. Correspondingly, every three days, one new job enters each work center in the manufacturing process.

The number of job or tool positions required within a work center is the flow time divided by the production cycle time, rounded up to the next largest integer. So, a work center with an eight-day flow time and a four-day production cycle has two job or tool positions. The work center has eight days to complete its required tasks on each plane and ships a completed job to the next work center every four days. Similarly, a work center with eight days of flow time operating on a three-day production cycle requires three job or tool positions. The schedule has a plane entering the work center at three-day intervals and a plane exiting the work center at three-day intervals; however, the arrivals and exits do not occur on the same day. As a result the number of job positions occupied by planes will vary between three and two over the three day cycle.

The number one flow chart depicts the exact sequence of every work center in the airplane manufacturing process (see Figure 1); there is a new number one flow chart for each new airplane program, model derivative, or new production rate. In Figure 1, the length of the jobs equals the flow time for the work center.

The determination of the flow time for a work center depends on the manufacturing work statement and the crew size for the work center. The manufacturing work statement details the exact tasks and the sequences in which these tasks must be performed for each job at a work center. At Boeing, the Industrial Engineering department estimates the direct labor input required to complete the tasks in the manufacturing work statements. They also estimate the learning curve for a work statement, which prescribes how the work content should decrease with experience.

For each work center, a crew-size study determines the minimum, maximum and optimal crew sizes based on detailed examination and planning of the work content in a work center. The (so-called) optimal crew size is the number of workers at the work center that minimizes the direct labor input per job.

Figure 1: Sample Number One Flow Chart

(WC## denotes work center)

An Illustrative Example

Suppose that the number one production unit (i.e. the very first airplane) will require eight hundred labor hours to assemble a plane at a given work center. The traditional planning methodology would set the crew size to minimize the direct labor per job; suppose in this case the optimal crew size is ten workers per job. Then, with one shift per day, the flow time for the work center is 800 hours/(10 workers * 8 hours per worker-day) = 10 days.

If the production line were to operate on a five-day cycle, we require two tool positions (10 flow days/5 day cycle). So, the work center will initially have twenty workers (ten workers per position) working on two jobs for ten days each. Every five days a new job will move into the work center and a completed job will leave.

Suppose the flow time is not a multiple of the cycle time. For instance, if the cycle rate is one plane every four days and the flow time is ten days, then there must be three positions. A new plane will enter the work center every four days and a completed plane will exit every four days: if new planes arrive to the work center on days 1, 5, 9, ...., then planes complete the work center on days 3, 7, 11, ... , where the plane that arrives on day x completes on day x+10. As a consequence, half the time (two out of every four days) the work force is working on two planes, while the other half of the time they have three planes to work on.

If this work center is staffed with three crews of ten workers, one per position, then each crew will be idle two out of every twelve days because job arrivals and exits are not synchronized. Thus, although the work statement calls for 800 labor hours per plane, this staffing plan with "optimal" crew sizes will incur 960 hours per plane (10 workers per position * 3 positions * 8 hr./day * 4 days/plane). To avoid this inefficiency, industrial engineering will relax the assumption of a dedicated crew per position and examine alternative ways to schedule the work in the work center. They will try to vary the crew size at each position with workers moving from position to position in order to achieve high labor utilization. In this example, at least 25 workers are required to produce a plane every four days (25 workers @ 8 hours per day * 4 days = 800 labor hours/4 days), which is the target for industrial engineering.

Alternatively, industrial engineering will examine flow times that are multiples of the cycle time, and thus facilitate scheduling of the work, albeit likely with non-optimal crew sizes. For instance, in this example it may be possible to have flow times of twelve days and three crews, each with nine workers. This staffing level incurs 864 labor hours for each plane.

Line of Balance Technique

The planning of airplane manufacturing, as described here, is an example of Line-of-Balance (LOB) analysis; see Schonberger (1985) and Iannone (1967) for a description. According to Schonberger, the LOB method is appropriate for the control of "limited-quantity production of a large-scale item." The method entails four steps. The first step is to establish a delivery schedule or rate, e. g., a cycle time of one plane every four days. The second step is to develop a process plan, indicating the sequence of manufacturing steps and their lead times; Figure 1 is an example of a process plan. Within the process plan, one would identify control points; at Boeing, the completion of work at the work centers (known as control codes) are the control points. The third step is to monitor the progress at each control point against the delivery schedule, and the fourth step is to project what the future schedule is for each control point. At Boeing, these steps are performed through a variety of means. For a fixed-rate delivery schedule, each work center will "ship" its assembly or subassembly to the next work center according to a fixed schedule driven by material handling considerations; Boeing will monitor how often these shipments are done before all of the work has been completed at the prior work center. Projecting the future schedules for a work center is an issue only when there is a change in rate; when a rate change is being considered, there is an extensive plan made to roll out the change across the process. The description of this is beyond the scope of our paper.

[Back to top]


In this section we identify three primary components of flow-time cost and discuss how to evaluate these costs in a context such as at Boeing. We then discuss the implications to the traditional production planning methodology at Boeing, and conclude with recommendations for how to incorporate flow-time costs into the methodology for production planning.

Flow Time Cost Elements

We consider three types of flow-time costs: 1) inventory carrying cost, 2) revenue opportunity cost, and 3) variable tooling cost.

Inventory Carrying Cost

The inventory holding cost for carrying the work-in-process (WIP) inventory includes the opportunity cost for the money tied up in the inventory, plus storage costs, insurance, spoilage and obsolescence costs. Usually, this cost is computed as the product of the inventory value and an inventory carrying rate, which includes at least the opportunity cost for money. For instance, in Figure 2 we show for a typical airplane how costs accumulate over the flow time for completing the airplane. The inventory holding cost for an airplane is found by multiplying this curve by the inventory carrying rate and then integrating over the total flow time.

We see from Figure 2 that a reduction in flow times will change the cumulative product cost curve, and presumably reduce the inventory holding cost per plane. Thus, the first type of flow-time cost is the inventory holding cost for the work-in-process.

Figure 2: Cumulative Product Cost Curve for an Airplane

Revenue Opportunity Cost

In a market where there is substantial demand backlog for a company's product, there is a second type of cost called revenue opportunity cost. Boeing commercial airplane group maintains a multi-year order backlog. (For instance in late 1992, Boeing had an $90 billion, three-year order backlog.) Revenue opportunity cost is the potential revenue from collecting sales revenue earlier if a shorter flow time results in earlier delivery of orders. For example, at the time this research was done (1990), demand exceeded supply in the commercial airplane industry. An airline ordering a Boeing 747-400 in 1990 was not going to get delivery of the airplane until approximately 1997. With airline passenger traffic predicted to grow at over 4% annually for the next decade, airline customers were eager to take delivery of newly designed, fuel efficient airplanes as quickly as possible. Given this market environment, there were significant revenue opportunity benefits associated with shorter product flow time (and earlier product delivery).

Flow Through vs. Flow Back

The possibility for a revenue opportunity benefit depends on the system response to a flow-time reduction. Consider a work center with eight days of flow time and a four-day cycle time, and suppose it reduces its flow time by one day. In isolation, the one-day flow-time reduction at the work center brings about no tangible benefits. This is because the one-day reduction has simply created a one-day time buffer at the particular work center if there are no other schedule changes to the adjacent work centers. To realize the benefits of flow-time reduction, either the upstream work centers need to delay their schedules to absorb the one-day reduction, or the downstream work centers need to accelerate their schedules to avoid the creation of a one-day buffer. We term these responses as "flow back" or "flow through," respectively.

By "flow back," we push the one-day reduction back through all upstream work centers. If the specified work center now has a seven-day flow time, instead of eight days, then it can receive its inputs one day later and still meet the original delivery schedule. Thus, the output schedule for the immediate upstream work center can be delayed by a day, which allows it to receive its jobs one day later, and so on. That is, the one-day reduction in flow time allows all upstream work centers to shift their schedules by one day. The primary benefits are savings in inventory holding costs, as discussed in the previous section.

By "flow through," we pull the one-day reduction through all the downstream work centers in the manufacturing process. To accomplish this, all of the work centers downstream of the specified work center must compress their schedule by one day on the very first airplane when the flow through is to occur. That is, these work centers will receive this very first plane one day earlier than planned, namely three days after the previous plane instead of the normal cycle of four days. These work centers need to complete their jobs within their normal flow times, and thus, they deliver the plane to the next work center one day ahead of the original schedule. After this very first plane, the schedules for all subsequent work centers are thereby advanced by one day. However, since neither the flow time nor the production cycle time changes for these work centers, these work centers simply experience a one-day compression when the flow-time reduction is pulled through for the first airplane; after that, the work centers should continue to operate as normal, but one day ahead of the original schedule.

Flow Through Illustration

We illustrate in Figure 3 how we can pull a flow-time reduction through the manufacturing process. From the figure, we see that the manufacturing process is operating at a three-day production rate and consists of three sequential work centers, A, B, and C, with flow times of five days, four days, and five days, respectively. Note that for the first two jobs in the production schedule, a new job is started and a completed job is shipped out every three days. Suppose that there is an opportunity to reduce the flow time at work center A from five to four days.

Table 1 below lists the start and completion dates for each of the work centers for all five jobs. On job number four, where the one-day time buffer is actually taken out of work center A, work centers B and C had to accelerate their production schedules to "flow through" the flow-time reduction (the dates in parenthesis in Table 1 are the original start and completion dates for each work center under the previous, longer flow time). After the one-time schedule acceleration for job number four, work centers B and C settle back to their regular production pace, starting and completing each job one day ahead of the old schedule.

This analysis applies similarly to a more complicated manufacturing process involving an assembly network of work centers (e. g. Figure 1). Then the schedule needs to be accelerated for all work centers downstream of the specified work center, and for all work centers on branches that join the network downstream of the specified work center. This is necessary in order to move the delivery schedule forward by the amount by which the work center's flow time has been reduced.

Figure 3: Production Schedule to Illustrate "Flow Through" Concept


Work Center A

Work Center B

Work Center C

Start Date Cmpletn Date Start Date Cmpletn Date Start Date Cmpletn Date
Job 1 0 5 5 9 9 14
Job 2 3 8 8 12 12 17
Job 3 6 10 11 15 15 20
Job 4 9 13 (14) 13 (14) 17 (18) 17 (18) 22 (23)
Job 5 12 16 (17) 16 (17) 20 (21) 20 (21) 25 (26)

Table 1: Start and Completion Dates for Five Jobs in Production Schedule

Advantages and Disadvantages of Flow Through versus Flow Back

A company can choose to "flow back" or "flow through" the buffer created by the flow-time reduction. Flow back simply requires that upstream work centers start later; there is no compression of the schedule and implementation is far easier than flow through. Since there is no change to the delivery schedule for the final product, the only savings are the reduction in inventory carrying costs due to a shorter flow time. Flow through shifts the production schedule ahead by the length of the flow-time reduction, and achieves revenue opportunity cost savings as well as inventory carrying cost savings. By choosing flow through, a company will have to accelerate the production schedule for a pre-selected job in order to pull the time buffer created by the flow-time reduction through the manufacturing process. This requires both careful planning and additional manufacturing costs, such as overtime, to accomplish the acceleration of the pre-selected job. Once this is done, all subsequent jobs follow the original schedule, shifted forward by one day.

Calculating Revenue Opportunity Cost

Calculating revenue opportunity cost for an airplane program requires knowledge of the cycle rate, selling price of the aircraft, customer pre-payment factor (if applicable), and relevant interest rates or the firm's cost of capital. Consider an example for an airplane with a sales price of $50 million. Suppose that there is no pre-payment factor and that there currently is a multi-year backlog for this airplane. Customers make their payment at the time of delivery and are willing to accept (and pay for) early delivery. The company is operating at capacity and produces one airplane every four days.

Consider a proposal to reduce flow time by one day. If the company pulls the flow-time reduction through the manufacturing process, it will then ship product to each of its customers one day earlier. In terms of cash flow, this will enable the company to collect its $50 million revenue from each customer a day earlier than under the current, longer flow time. This shift in the revenue stream generates revenue opportunities for the company in the form of either simple interest or internal investments. For instance, at an annual interest rate of 10% and a working calendar of 250 working days per year, one (working) day of interest on $50 million is $20,000. Over the course of a year, with a four-day production cycle, the revenue opportunity from the one-day flow-time reduction amounts to $1.25 million. This savings of $1.25 million per year will recur for the life of the plane's backlog, or until the production rate is reduced. Furthermore, any additional reduction in the flow time that can "flow through" the process will generate revenue opportunity savings of $1.25 million per year for each day. And any flow-time reduction, regardless of whether it flows through or back, yields savings in inventory carrying costs.

Variable Tooling Cost

Variable tooling costs are especially important in a high capital, labor intensive manufacturing environment such as at Boeing. These costs result from the purchase and maintenance of production tools and equipment for the manufacturing process; we illustrate here how these costs depend upon flow times.

A work center with eight days of flow time and a four-day production cycle requires two tooling positions. If the production rate increases to a three-day production cycle, the number of required tooling positions increases to three and a new tool has to be purchased, say at a cost of $1.2 million. However, if the work center can reduce its flow time to six days, then the tooling requirement for the work center remains at two when the production cycle decreases from four to three days. Therefore, in this example, flow-time reduction from eight to six days saves $1.2 million in tooling costs.

Variable tooling costs (and savings) increase in fixed increments that depend upon the production cycle time and the flow time for the work center. In the example, reducing the flow time from eight to seven days does not impact the tooling costs when the production cycle changes to three days; the number of required tools remains at three. That is, a one-day flow time reduction has no variable tooling benefit in this example, whereas a two-day flow time reduction saves $1.2 million in variable tooling saving.

Intangible Elements of Flow-Time Cost

In addition to the three types of flow-time cost, there are intangible costs as well. Long flow times in the manufacturing process lengthen feedback on production problems and allow these problems to accumulate in work-in-process inventory. Because of this, these problems require more corrective efforts to resolve and more rework to restore the work-in-process inventory.

Long flow times also decrease a company's capability to respond quickly to shifting market demand. At Boeing, since all production is to customer order, long flow times limit the company's ability to respond to change requests from customers. Because of the long manufacturing flow time, Boeing will encourage their customers to decide exactly what they want long in advance of delivery. However, since this is not always realistic, Boeing will incur additional costs, due to the disruption of the normal manufacturing process, to accommodate the inevitable changes from customers.

Implications of Flow-Time Cost on Production Planning Methodology

There has been limited visibility and awareness of the costs of flow time in the planning methodologies practiced at Boeing. To appreciate this, we offer the following observations to provide some perspective.

At Boeing, adherence to schedule is paramount, due to the significant cost penalties for delays in airplane deliveries. The sequential nature of the manufacturing process dictates that upon completion of each production cycle, each work center must advance a job to the next work center in the manufacturing sequence. The delay of a single job within the sequential manufacturing process disrupts the work flow on the production line and postpones the delivery of every successive airplane by the length of the delay. If a job is not completed within the allotted flow time, the incomplete job is nevertheless moved on to the next work center so that all following airplanes in the production line can proceed to their next respective work centers. The late airplane will then have two separate crews working on it during the manufacturing flow time in the next work center. One of the teams will be the regular crew of the new work center, the other is a special crew from the previous work center sent over to complete all remaining incomplete tasks. Manufacturing management monitors very closely these incomplete jobs, called "travelers." Thus, the prevailing attitude within manufacturing is to meet the schedule and avoid having to move incomplete jobs.

In Boeing’s management accounting system, there has been little recognition of cost associated with manufacturing flow time. The lack of flow-time cost visibility, coupled with the importance of completing jobs to schedule (while maintaining the capability to manage unforeseen disruptions) and close management scrutiny on work-force head count, all contribute to the practice of managing the work-force head count, at the expense of manufacturing flow time. Consequently, as the total labor required within a work center decreases because of worker learning, the production planning methodology has relied heavily on head-count reductions to realize learning-curve benefits; at the same time the planning methodology maintains flow time to insure that work centers can meet the strict production schedules and accommodate unforeseen disruptions.

In the remainder of this section, we argue that there are tradeoffs between the work-force level and the flow times, and that these tradeoffs need to be explicitly examined as part of the production planning. In particular, we discuss two immediate implications to Boeing's production planning methodology that result from an awareness of flow-time cost.

Determination of Work Center Design Parameters

In determining the flow time and the staffing level for a work center, production planning needs to adapt its methodology to evaluate all relevant costs, rather than focus primarily on labor efficiency and capital investment.

For example, consider a work center with a ten-day production cycle, a twenty-day flow time, and with a total staff size of six. The labor input is 480 hours per job. The present operation minimizes labor input per job by operating with the optimal crew size while protecting the schedule by having two jobs in process for smoothing unforeseen disruptions. Now, consider a proposal to reduce two days of manufacturing flow time at the work center by adding two more workers so that the labor input is 640 hours per job.

The traditional production planning methodology would view the flow-time reduction as a bad proposal because of the increased labor cost per job. However, by incorporating flow-time costs, this proposal might actually be very beneficial since it reduces inventory carrying cost at the work center by two days, and depending upon the implementation, may result in revenue opportunity cost savings.

Realization of Productivity Improvements

In a manufacturing environment where there is significant worker learning, the labor input required within each work center decreases as a function of the number of airplanes produced (see Chao (1991) for an example).

As the labor hours decrease, the production planners have to decide how to utilize these productivity improvements. Their options are to reduce the number of workers at the work center, or reduce work center flow time, or a combination of both. Because of past emphasis on head count as the primary tool of cost control, and due to the lack of flow-time cost visibility, production planners have relied primarily on head-count reduction to realize cost savings from these productivity improvements.

Proposed Production Planning Methodology

We have proposed a new methodology for utilizing worker productivity improvements that evaluates the possibility of reducing flow time. We illustrate this approach with an example.

Suppose a work center initially requires 40 labor-days, and operates with ten flow days and two tooling positions to meet a five-day production rate There are 8 workers working in the work center, 4 for each tooling position. Through worker learning, suppose the average labor content decreases to 10 labor-days by unit 256; however, because of manufacturing variances, labor content varies from plane to plane, and can range up to 16 labor-days per plane. Assume that because of projected market demand, the production cycle rate is to increase to a two-day cycle. Table 2 lists three alternative scenarios of utilizing the productivity improvement benefits and their respective impact on flow time, labor head count and tooling positions.

From Table 2, we see that the three scenarios have drastically different labor content per job. In scenario one, where the work center has ten flow days and five job positions, the supervisor can shift workers between jobs (from easier jobs to harder jobs) and smooth the work variability between incoming airplanes [see Chao (1991) for more discussion]. In scenario three, where the work center has only two flow days and one job position, the supervisor must staff at a level capable of completing even the most difficult jobs within the production schedule; since the labor content per job can range up to 16 labor-days, the supervisor has to staff the work center with eight workers so that 16 labor-days are available per job. Scenario two is a mixture of the two extremes.

  Scenario 1 Scenario 2 Scenario 3
Flow Time 10 days 6 days 2 days
Cycle Time 2 Days 2 Days 2 Days
Tooling Positions 5 positions 3 positions 1 position
Staffing 5 workers 5-6 workers 8 workers
Avg. Input / Job 10 labor days 10-12 labor days 16 labor days

Table 2: Three Different Ways to Realize Productivity Improvements

To choose from these scenarios requires knowledge of the flow-time costs as they apply to this work center, plus the cost of labor. More often than not this evaluation favors the scenario that reduces flow time the most, at the expense of labor productivity. This is counter to prior practice, which has not considered flow-time cost and has focused on minimizing labor content.

Implications of Proposed Methodology on New Airplane Program

In a new airplane program, where facilities have not yet been built, the proposed production planning methodology has significant impact. In particular, realizing productivity improvements via flow-time reduction will result in significantly shorter flow times as the number of airplanes manufactured increases. Therefore, as the new airplane gains market acceptance and approaches maximum production rate, the shorter flow time will reduce the costs for facilities and tooling, as well as providing reduced inventory carrying cost and revenue opportunity cost. For a new airplane program, where new capital investments add up to hundreds of millions of dollars, the proposed production planning methodology can bring about significant program savings.

[Back to top]


In the previous sections, we quantify the cost of flow time and examine the tradeoff between flow time and labor content within aircraft manufacturing. In this section, we consider how to avoid this tradeoff by attacking both flow time and labor content simultaneously. In particular, we analyze the impact of system variances on labor input, and indirectly on flow times. We define system variance as the factors or elements within the manufacturing environment which affect the execution of baseline manufacturing operations. Examples of variances in the manufacturing environment are engineering changes, part shortages, job rework, part rejections, and various product options.

At Boeing, significant portions of total manufacturing labor input are attributable to system variances. In this section, we present a working hypothesis regarding the effects these variances have on manufacturing labor input. We use regression to test the validity of the working hypothesis and to estimate the effects of the variances on manufacturing labor input.

The regression analyses support the working hypothesis that relates actual labor input to a baseline work time and the effects of external manufacturing variances. The regression focuses improvement efforts on the high-impact variances instead of diverting attention onto all the variances. Eliminating or reducing these system variances leads to productivity improvements (lower labor input) which brings about three significant benefits: 1) lower direct labor cost, 2) decreased variable labor overhead, and 3) reduced flow-time cost. The improved labor productivity leads to reduced flow-time cost because a significant portion of the flow time is due to variance-related activities. If the level of these variances can be reduced, the efforts required to correct these variances will decrease and the associated flow time can be taken out without incurring additional risk to the schedule.

Working Hypothesis

We conjecture that for each work center there is a baseline work package that the work center is required to complete. Associated with this baseline work package is the baseline work time (BWT), measured in labor hours per job, which is a function of the complexity of the work to be performed and the number of airplanes manufactured thus far. The complexity of the baseline work package determines the initial time required to complete the tasks at the work center, while the number of units manufactured and the slope of the learning curve determine the actual baseline work time required for each airplane.

The actual manufacturing time spent by a work center to perform the required tasks differs from (usually greater than) the BWT. The workers at the work center, while working on the baseline work package, have to contend with external system variances such as engineering changes, part shortages, and rework that disrupt the process work flow and add extra work to the baseline work package. These system variances increase the labor input required by each work center to complete its operations.

We model the actual manufacturing time at each work center as the sum of the BWT and the cumulative effect of the various external system variances. We will test the validity of this working hypothesis by means of a multivariate regression analysis.

Data Collection Methodology

To test the validity of the working hypothesis, the first co-author studied an existing Boeing airplane program, which we call the 7A7 to protect its identity, and collected actual direct labor data and system variance data for fifty consecutive Boeing 7A7 airplanes.

Major Shops

During data collection, we uncovered a difference in the way that data for direct labor hours and data for the system variances are kept. The direct labor hours are recorded and stored at the work center level, whereas the system variance data are collected, aggregated, and reported at the "major shop" level. These "major shops", which are collections of work centers in the manufacturing process, are the major operational units of the manufacturing organization. The four major shops within the manufacturing sequence are: 1) Body structures, 2) Wing structures, 3) Join & Installations and Final Assembly, and 4) Field Operations.

To insure compatibility of data, we aggregated the labor hours data for the work centers by the four major shops. In addition, we aggregated all of the data for the four major shops to form a data set for analysis at the airplane level. This was done in order to get a macro view of the overall impact of system variance effects on manufacturing direct labor input.

The actual labor input for manufacturing each airplane consists of direct labor plus several categories of indirect labor, such as rework. Because the data for the indirect labor categories are collected on a monthly basis rather than on a plane-by-plane basis, our analysis only considers the impact of system variances on the direct manufacturing labor hours.

Description of Regression Analysis

A total of five separate analyses were run for the Boeing 7A7 program. The first analysis is the regression of direct labor hours against about 30 system variances for the total airplane, and provides an analysis of variance impact on the entire 7A7 manufacturing process. The other four analyses are for assessing variance impact on each of the four major shops in the manufacturing process We will report here only the results from the total plane regression; the regression results for the four major shops are similar to that for the total airplane but differ according to differences in the operating characteristics of these shops. (see Chao (1991) for details)

In order to build an accurate model of direct labor hours, the first author consulted extensively with senior managers and industrial engineers. This consultation helped to determine the relevant manufacturing variances to include in the regression analyses, as well as to insure that the results of the analyses made sense and were consistent with their experience. Initial efforts attempted to regress over thirty different system variance variables as independent variables against the dependent variable, direct labor hours. This approach resulted in unsatisfactory solutions because many of the independent variables were correlated.

From this experience, the first co-author conducted a new round of interviews with industrial engineers, manufacturing managers, shop superintendents and factory managers to ask them what variables had the most impact on the manufacturing labor in each of the four major shops. The input from these individuals helped to prioritize the list of variances to form the starting set of variables for the new statistical analysis.

To complement the new starting set of variables, we applied stepwise regression, which iteratively adds (or deletes) one variable at a time to the regression model, where the method selects the variable that will yield the largest reduction in the amount of unexplained variability in the model Using this method, along with the smaller starting independent variable set, we were able to develop a satisfactory model for each major shop and for the airplane as a whole.

Statistical Regression for Total Airplane

In Table 3 we give the results for the regression model for the total airplane, namely the independent variables, their coefficients and their standard errors. We do not report the intercept for the model, in that it represents an estimate of the direct labor hours for the first plane, and hence is proprietary. We will define the independent variables below.

The top-level airplane regression explains 96% of the variability in the dependent variable, direct labor hours (R2 = 0.96). Furthermore, all of the variable coefficients are consistent with our prior expectations.

Variable Coefficient (labor hour/occurrence) Std. Err
Customer introduction 2964 769.7
Part Shortage 3.6 1.0
Production Revision Request 276.7 28.3
Model 2 -2247.8 878.4
Defects 1.3 0.46
log2(x) of Unit number -47732.6 8117.8

Table 3: Regression Results for Total Airplane

The customer introduction variable is a binary variable, which denotes that a particular airplane is being delivered to a new airline customer. This usually requires quite a bit more direct manufacturing input because of the learning required to satisfy the custom specifications for the first airplane for a new customer In addition, during the customer introduction process, the airline customer is usually more exacting in inspections and thus requires more time during the acceptance process. From the regression, we see that a new customer results in nearly 3000 additional direct labor hours.

The part shortage variable denotes the total number of occurrences per plane where a part needed on the line is not available for installation. The production revision request variable is the number of requests generated by the manufacturing or engineering organizations to revise the manufacturing plan of an airplane; the variable only counts requests that require at least 100 labor hours and are subject to management review. These variables add to the manufacturing effort required to assemble and test the airplanes, as indicated by the positive regression coefficients.

The baseline airplane model of the regression, because of its popularity in the fifty plane sample, is the 7A7 model 3. The model 3, which is approximately thirty feet longer than the model 2, requires more assembly and integration time than the model 2. The model 2 variable is a binary variable that takes a value of one for model 2 and a value of zero for model 3. As expected, the regression model indicates that the model 2 airplanes require 2200 less labor hours to manufacture.

The defect variable counts the number of occurrences of correctable rejectable conditions on an airplane, as detected by the Quality Assurance department. Defects are usually considered to be relatively insignificant in terms of their overall effect on total manufacturing hours. From the regression model, we find that each defect adds only 1.3 labor hours. However, the appearance of defects as an explanatory variable in the regression suggests that defect rework labor is a significant part of the total direct labor hours expended in the manufacture of airplanes.

Finally, the Log 2 of Unit Number variable captures the learning effect. The independent variable is the log, base 2, of the cumulative production number of the plane; for instance the thirty-second plane produced by the line would have a value of five for this variable. The regression coefficient for this variable signifies by how much the direct labor hours per plane go down when the production count doubles. As expected, we see a strong learning effect for the total manufacturing labor input as a function of the number of airplanes produced.

The traditional learning curve model assumes that the labor content goes down by a fixed percentage when the cumulative production count doubles. Here, we assume that labor content goes down by an absolute amount when the cumulative production count doubles. For both the total airplane and the four major shops, this learning model works quite well in terms of model fit; furthermore, it permits us to maintain a linear model that can capture the effects from system variances.

Construction of Variance Pie Charts

From the statistical analysis, we can construct a "variance pie chart" to show the relative impact of system variances on direct labor hours. In particular we determine what percentage of the total direct labor hours are due to each of the system variances from the regression. For instance, from the regression for the total airplane, suppose that the average number of defects per plane is 100. Then, each plane had an average of 130 labor hours due to defects; if each plane had an average of 5000 direct labor hours in total, then defects are responsible for 2.6% of the direct labor. In this way, Boeing has used the regression results to determine the relative contributions of the system variances to the total direct labor and to identify the high impact "vital few" variances. We have found that displaying these percentages with a pie chart is a very effective way to highlight the most important system variances. Unfortunately, due to the proprietary nature of the data, we cannot include the variance pie charts for the four major shops and for the entire airplane final assembly process of the Boeing 7A7 airplane.

Strategy for Reducing Defects

The regression analyses show that, consistent with Pareto principle, a few variances account for the majority of the impact on the direct manufacturing labor input for the assembly of airplanes. Simply presenting these results, however, does not always point to particular strategies for reducing the level of these variances. We summarize here the results from further analysis that produced a strategy for reducing one of the significant variances, namely defects; see Chao(1991) for more details.

In the regression analysis, defects accounted for a significant portion of direct manufacturing labor input. To develop a strategy for reducing defects, we hypothesized that the engineering organization has a major effect on the level of defects and thus, plays an indirect role in determining the total direct labor input. That is, certain engineering activities such as engineering changes and engineering errors disrupt the normal work flow in a work center and increase the likelihood of worker error, i. e., defects. To test this hypothesis, we again employed a regression analysis to relate statistically two engineering actions to the number of defects. The value of this analysis was that it provided evidence to support the hypothesis, namely that two engineering variances were driving the number of defects. In particular, we found that engineering has an important indirect impact on direct manufacturing hours through the effects that the quality of engineering release has on high-impact variances such as defects.

[Back to top]


This research generated three major recommendations for Boeing. The first recommendation is to recognize the cost of flow time as part of total cost, and to incorporate this cost into decision making at all levels. The second recommendation is to implement a strategy for flow-time reduction throughout the manufacturing process. In the near term, this strategy is to consider flow-time reduction as one means for realizing the benefits from learning in the manufacturing process, and to evaluate the tradeoff from flow-time reduction versus the alternative of head-count reduction. In the long term, the strategy is to focus improvement efforts on eliminating system variances and improving the quality of engineering releases to manufacturing. The benefits from these longer-term activities will be less direct labor, less overhead and shorter flow times. The final recommendation is to adjust the incentive system to motivate flow-time reduction. This includes putting the cost of flow time in the operating budgets for the manufacturing divisions and including flow time as part of the performance objectives for all levels of management.

At the completion of the research, the first co-author presented the findings and recommendations to numerous management teams at all levels of the organization. The response from Boeing management was very positive; as the research was conducted with the cooperation of senior management and involved participation of numerous Boeing organizations, there was considerable support and ownership for the recommendations.

As a result of these meetings, a planning directive was issued to take specific actions on these recommendations. First, a second study was performed on the 7B7 program to determine the impact of manufacturing variances on labor productivity, and thus replicate what had been done on the 7A7 program. The finance department was assigned to quantify the flow-time cost for each flow day in the manufacturing process for the 7A7 and 7B7 programs. Finally, the manufacturing organization at the facility (which assembles both the 7A7 and 7B7) was asked to initiate and implement flow-time reductions for the two airplane programs.

From these efforts, Boeing has achieved significant benefits. The manufacturing organization has removed several days of flow time from the 7A7 and the 7B7 programs. These reductions came primarily from converting the productivity improvements from learning into flow-time reductions rather than head-count reductions. Some of the flow-time reductions have been pushed through the manufacturing process to permit the acceleration of deliveries of airplanes. Within the first year, Boeing delivered one additional 7A7 and one additional 7B7 and collected the net profits from these additional deliveries. These reductions have also contributed to reducing work-in-process inventories and have generated net savings in inventory holding costs of tens of millions of dollars over a four year period.

More important than these immediate benefits is the fact that these efforts are continuing and the paradigm at Boeing for manufacturing planning is changing. Factory management and industrial engineering continue to drive the efforts to reduce flow time in the factory. The manufacturing organization recognizes the cost of flow time and it is being integrated as part of their planning methodology and their performance management system. Finally, the first assignment for the first co-author upon joining Boeing was to assume responsibility for inventory reduction for the next Boeing commercial airplane, the 777; so these ideas and concepts have been applied to this new program.

[Back to top]


CHAO, J. S., 1991, "Analysis of Variance Impact on Manufacturing Flow Time," S. M. Thesis, Massachusetts Institute of Technology, Leaders for Manufacturing Program, Cambridge MA.

DERTOUZOS, M., R. LESTER, and R. SOLOW. 1989. Made in America, MIT Press, Cambridge MA.

GOLDRATT, A. and R. FOX. 1986. The Goal, Rev. Edition, North River Press, Croton-on-Hudson, New York.

HAYES, R., S. WHEELWRIGHT, and K. CLARK. 1988. Dynamic Manufacturing: Creating the Learning Organization, The Free Press, New York.

IANNONE, A. L. 1967. Management Program Planning and Control with PERT, MOST and LOB, Prentice-Hall, Inc. Englewood Cliffs NJ.

SCHMENNER, R. 1988. "The Merit of Making Things Fast," Sloan Management Review, Fall, pp. 11-17.

SCHONBERGER, R. J. 1985. Operations Management, Second Edition, Business Publications, Plano Texas.

STALK, G. Jr. and T. M. HOUT. 1990. Competing Against Time, Free Press.

Key Words: flow time reduction; aircraft manufacturing

[Back to top] [Return to papers]