Customization: Impact on Product and Process Performance

Manufacturing capability has often been viewed to be a major obstacle in achieving higher levels of customization. Companies follow various strategies ranging from equipment selection to order process management to cope with the challenges of increased customization. We examined how the customization process affects product performance and conformance in the context of a design‐to‐order (DTO) manufacturer of industrial components. Our competing risk hazard function model incorporates two thresholds, which we define as mismatch and manufacturing thresholds. Product performance was adversely affected when the degree of customization exceeded the mismatch threshold. Likewise, product conformance eroded when the degree of customization exceeded the manufacturing threshold. Relative sizes of the two thresholds have management implications for the subsequent investments to improve customization capabilities. Our research developed a rigorous framework to address two key questions relevant to the implementation of product customization: (1) what degrees of customization to offer, and (2) how to customize the product design process.


Introduction
Customization strategy is increasingly being pursued by industries to gain competitive advantage (Pine 1993).A fairly high degree of product customization is prevalent in almost all industry segments today.The popularity of customization stems from a marketing desire to achieve a better fit between customer needs and company's product offerings.Customer satisfaction and market performance are expected to increase with this improved fit.
A key factor is whether the manufacturer can facilitate higher customization levels without escalating costs.Pine (1993) introduces the notion of "Market Turbulence Map," which is a function of several market environment factors, to assess the value of shifting to a mass customization strategy.His work characterizes customization as a products and services strategy that responds to varying market performance requirements.
Mass customization thus has been linked to market performance.It also impacts significantly the manufacturing functions as discussed in the manufacturing strategy and manufacturing flexibility literature.For example, Wheelwright and Bowen (1996) present a manufacturing strategy framework that connects market performance through products and services to the capabilities of the manufacturing function.Likewise, Safizadeh et al. (2000) examine the trade-offs between costs and customization, and between cost and design quality that are inherent to the firm's process choice decisions.Gupta and Somers (1996) state that flexibility enables manufacturers to produce customized products without sacrificing cost-efficiency and product effectiveness.Gupta and Lonial (1998) describe the links between process structure complexity, and product structure complexity, manufacturing strategy and product customization.Clearly, recent studies strongly indicate that customization of products is an important component of manufacturing strategy initiatives (Schmenner 1997;Clark 1996).Also, achieving customization capability is viewd as an important factor driving the capital investments in manufacturing flexibility (Gupta and Somers 1996).
A variety of strategies and methods have also been proposed to enhance manufacturing flexibility and capabilities (Upton 1994).Our work builds on past research.For example, Swaminathan (2001) presents a framework to facilitate customization using standardized operations utilizing process and product modularity.Points of customer involvement in product design, manufacturing and the delivery process are another critical variables that impact product and process modularization.Duray et al. (2000) propose customization strategies based on the point of customer involvement.Gilmore and Pine (1997) propose customization strategies, which are based on the degree of changes to product design and product representation.Swaminathan and Tayur (1988) and Lee and Billington (1994) present late product differentiation techniques to mitigate the negative impact of customization on supply chain performances.Pine (1993) highlights the need to investment in flexible manufacturing systems, information technologies, and modular product design in order to facilitate customization.In this paper, we focus on the strategies that have been proposed specifically to enhance the capabilities of manufacturing to facilitate customization.
As established in the above overview, a large body of research has been reported on acquiring manufacturing capability to facilitate customization.However, there is a paucity of work on the efficacy of customization from a front-end process perspective.The willingness to pay a premium price for customized products largely depends on the higher performance quality of the product (Chakravarty and Kumar 2002).The achievement of higher product performance through tailoring the product to unique customer needs is essential for generating the demand for and bringing about the success of customization.Customization involves the identification of individual customer's needs and then subsequently satisfying the needs through appropriate product modifications.An example of this is seen in Fisher (1994), wherein the customization process at National Bicycle is described in detail.Bicycles at National Bicycle are designed by acquiring customer specific parameters through automated measuring technologies in the retail outlets.In such cases, the customer provides the key technical parameters of the product such as individual physical dimensions, body posture, and specification of the drive trains.This allows National to suitably modify standardized frame designs and manufacturing pro-cesses to provide the specific performance and conformance sought by customers.
However, as Zipkin (2001) states, elicitation of customer requirements is hard and often customers have trouble articulating their requirements.Contrary to the popular viewpoint, higher degrees of customization may not always result in higher performance and customer satisfaction due to complexity of the customization process.The complexity of the customization process can potentially limit the degree of to which customization is beneficial to customers (Zipkin 2001).Managing the linked engineering and manufacturing capabilities (both product and process interactions) during customization is equally important to ensure conformance quality (Voss and Winch 1996;Bordley 1993).We investigate these complex aspects in our field study.Specifically, our research investigates two key unexplored research questions identified by Ramdas ( 2003): ( 1) what degrees of customization to offer, and (2) how to customize the product design process.We investigate both these questions from a quality perspective.
The objective of our research is primarily to examine the efficacy of product customization in light of the complexity of the customization process.We investigate how customization affects the two key dimensions of quality: performance and conformance (Garvin 1987;Chakravarty and Ghose 1993).Specifically the following two questions are analyzed.First, does the complexity of the customization process limit the degree of customization that is beneficial from a product performance point of view?Here we find from our empirical analysis, that up to a limit, which we define as the mismatch threshold, customization does improve the product performance.Moreover, beyond this threshold, customization is found to be detrimental and it significantly escalates mismatch errors (decreases product performance).Second, does the manufacturing capability limit the degree of customization from a product conformance point of view?Here again, we find that up to a limit, which we define as the manufacturing threshold, customization does not create any adverse impacts.Beyond this second threshold, customization strains the capabilities of the manufacturing systems thereby decreasing product conformance.Flexible manufacturing systems will increase the manufacturing threshold.However, the relationship between the manufacturing threshold and flexible manufacturing systems is not examined in this paper.Further, we also examine the relative sizes of the two thresholds (mismatch and manufacturing) and derive implications on process investments to improve customization capabilities.
Our field study examined these issues at a large manufacturer of industrial iron and steel rolls.The research site was a designed-to-order (DTO) industrial component-manufacturing environment.The physical and metallurgical specifications of the rolls were unique to each order.Product customization was implemented by allowing the customer to choose a set of metallurgical parameters.The number of metallurgical parameters specified by the customer was a proxy for the degree of customization sought in the rolls.The product modifications necessary for customization resulted in a tension between the customer's performance requirements and the in-house capability of the manufacturing processes at the roll manufacturer to deliver conformance.A shortfall in meeting customer requirements led to mismatch errors.Likewise, a shortfall in the incapability to conform by the manufacturing process led to manufacturing errors.Either of these shortfalls led to roll breakages in the field at the customer's site.These breakages are classified into either mismatch errors or manufacturing errors.
We investigated the various drivers of both these two types of failures, mismatch errors, those where the customization did not meet the performance desired by the customer, and manufacturing errors, those where the manufacturing process did not achieve conformance.The impact of these drivers was analyzed deploying a competing risk hazard function model.The model parameters were estimated using the primary data.The estimated model quantified the relative magnitude of both types of failures.In particular, the following key managerial insights emerged: • A mismatch threshold exists.Customization up to this threshold, improves performance quality and decreases mismatch errors.However, beyond this threshold, further customization significantly increases mismatch errors.Customization beyond this threshold is thus counterproductive from a performance viewpoint.• Another threshold defined as the manufacturing threshold also exists.Customization up to this threshold does not decrease manufacturing quality.Beyond this threshold, however, an increase in customization escalates manufacturing errors.• At our site, the mismatch threshold was found to be smaller than the manufacturing threshold.Thus the mismatch threshold was far more restrictive from a customization point of view.This implies that the manufacturer should investigate opportunities to invest in front-end process technologies such as product configurators in Design-to-Order (DTO) environments.The above findings have important managerial implications on the implementation of customization.The recent drive to enhance manufacturing capabilities and cope with the adverse impact of customization seems to have extended the manufacturing threshold.In such cases, the manufacturing process is not restrictive.However, the front-end product customi-zation process and its associated complexity limit the degree of customization that is beneficial to customers.The degree of customization in such cases is limited by customization eliciting mechanisms.This study finds strong empirical evidence to support Zipkin's (2001) viewpoint that elicitation and the customization process itself often impose limitations on customization.Our research thus developed a rigorous framework to answer two key unexplored research questions identified by Ramdas ( 2003) from a quality perspective.First, we explained the concepts of mismatch threshold and manufacturing threshold.Identifying the drivers of these thresholds can help companies decide the degree of customization to offer.Second, we underscored the importance of modifying the drivers of the customization process.Further, the results of our research provide another plausible explanation for the Gupta and Somers (1996) counter-intuitive findings of a negative relationship between process flexibility and company's performance.Manufacturing flexibility is only one of the dimensions of the firm's customization capabilities.Achieving higher manufacturing flexibilities may not always directly translate into higher market performance due to limitations of the front-end process.In such cases, the interdisciplinary front-end process also becomes a critical part of the customization capability development.
The rest of the paper is organized as follows.In Section 2, we develop a conceptual framework and formulate hypotheses in the context of our research site.In Section 3, we describe the econometric analysis.In Section 4, we report the results and discuss the key managerial implications.In the final section, Section 5, we conclude by summarizing the results, discussing the limitations of the study and identifying opportunities for future research.

The Research Context and a
Conceptual Framework

The Research Site
The research site was a Pittsburgh based manufacturer of iron and steel rolls.These rolls were designed to order based on the profile of a customer's rolling mill.
Besides size and other physical characteristics, the rolls were designed to have metallurgical properties for a variety of grades of alloy steels and rolling campaigns of the customer.Customers were given choices to specify certain parameter values in the metallurgical specification such as alloy composition, hardness, molding, and heat treatment variables.Some customers assumed this responsibility more than others and tailored their needs by selecting specific values matching their unique application requirements and mill conditions.The manufacturer, recognizing his process capability relative to customers' requirements, subse-quently chose all the remaining parameters not set by the customer.Customization of metallurgical parameters was thus important to both parties, customer and the roll manufacturer.The metallurgical parameters determine roll performance for the customer and the manufacturing performance for the manufacturer simultaneously.Shortcomings surfaced as field complaints when the rolls broke or wore out prematurely.We classify such failures as either manufacturing errors or mismatch errors.Manufacturing errors arise when the manufacturer's production process is not capable of conforming to specifications set by the customer.In contrast, mismatch errors are the failures caused by non-manufacturing factors such as design errors.The failure detected in the field triggered a battery of metallurgical tests that determined the causes of failures into two categories: manufacturing or mismatch errors.Ambiguity in the classification occurred in less than 5% of the cases leading to either arbitration or protracted negotiations.

Dimensions of the Designed-To-Order
Process There are two drivers in the customization process: degree of customization and product complexity.We formally define these drivers in the context of the roll manufacturer.
2.2.1.Degree of Customization.As explained earlier, the customer participated in the customization process by specifying a subset of the metallurgical parameters.The total number of parameters varied depending upon the roll family and ranged from 9 to 22.The customers were given a choice to specify a subset of these parameters (30% to 60%) depending upon their specific needs and condition of the mill.The roll manufacturer incorporated values specified by the customer and determined the remaining set considering his internal process capability and the customer's requirements in best way possible.With such a choice scheme, we operationalize the degree of customization as follows:

Number of metallurgical parameters set by the customer Total number of metallurgical
parameters for the roll family Even within a roll family, some customers chose to participate more than others and specified a larger set of parameters values.The data revealed a wide range of customizations for the same product family.Figure 1 depicts the various degrees of customizations for a typical roll family.

Product Complexity (PRODX).
Rolls were ranked on a 1 to 5 scale to represent product complex-ity.The metallurgical group at the manufacturer established this ranking.The product complexity ranged from 1.2 to 4.8 and exhibited an average of 3.28.The product complexity ranking was jointly driven by the total number of parameters in the specification and the interdependencies among these parameters from metallurgical and mechanical standpoints.This product complexity measurement is consistent with the design engineering literature.Product complexity has three main elements: (1) the size of the design problem, that is the number of design parameters, (2) correlation among the design parameters, that is the interdependencies, and (3) variability in the design parameters (El-Haik and Yang 1999;Suh 1990).The product complexity represented here captures the level of difficulty associated with the design-engineering problem.
In summary, the CUSTOM variable measures the degree of design flexibility utilized by the customer for the given roll family.However, rolls from different roll families with the same CUSTOM value could vary in terms of design complexity.The PRODX variable accounts for the differences in design complexities arising from the number of parameters and their interdependencies.

Roll Failures and Customization
We discuss next the impact of customization on roll failures detected in the field.As explained earlier, metallurgical tests firmly attribute the failures to either manufacturing or mismatch modes.The manufacturing errors result from poor conformance quality (manufacturing defects), where as the mismatch errors result from poor product performance caused by nonmanufacturing reasons.The impact of customization on both these failure types is hypothesized next.

Customization and Mismatch
Errors.We conjecture that with increased customization, mismatch errors first decrease up to the mismatch threshold.Beyond this threshold however, mismatch errors start escalating rapidly.The rationale for this trend is as follows.The initial gains from allowing the customer to select parameters can be attributed to the customer's deeper knowledge of his/her rolling operations.
The customers also have a better understanding of the sensitivity of specific roll parameters to their individual rolling mill conditions.This knowledge resides more with the customer than with the roll manufacturer.Thus eliciting these values from the customer results in lower misapplications.Von Hippel (1998) has highlighted this aspect in CTI (Computer Telephony Integration) and ASIC (application-specific integrated circuit) production environments.Typical failures that are reduced in region 1 (Figure 2) are listed in Table 1.
For a particular product with given number of parameters, a higher degree of customization means the customer has to select a larger set of parameter values.As the number of parameters established by the customer increases in region 2 (Figure 2), the inter-dependencies among parameters become less manageable.This aspect has been emphasized in the literature related to House-of-Quality and product-process interactions (Hauser and Clausing 1988;Chakravarty and Ghose 1993;Bordley 1993).This difficulty leads to higher chances of design errors during the customization process.We therefore conjecture that the customer is able to leverage gains from customization only up to the mismatch threshold.Beyond this threshold, the complexity associated with the interdependencies dominates and the failure rate rises negating the earlier gains.The typical failures in region 2 (Figure 2) are listed in Table 1.The relationship between the degree of customization (CUSTOM) and the mismatch errors is summarized in the following hypothesis.
Hypothesis 1.The probability of a mismatch error first decreases with an increase in the degree of customization up to a threshold level called the mismatch threshold.Beyond this threshold, the mismatch errors increase with higher degrees of customization.

Customization and Manufacturing Errors.
The manufacturing capability at a roll manufacturer typifies a flexible manufacturing system that can handle a variety of roll families.As MacDuffie et al. (1996) points out, such systems provide a range of flexibility, where variety does not cause any detrimental effects up to a certain level.We also integrate concepts from Swaminathan (2001), who underscores the importance of product and process modularity.In our context, modularity in customization is the ability to accommodate a range of choices in the metallurgical parameters set by customers within the constraints of the manufacturer's standard routings.The manufacturer can achieve metallurgical modularity by allowing the customers to select their own specific values for a set of metallurgical parameters.The manufacturer adjusts the remaining parameters to deploy their own standard routings to ensure conformance quality.However, customization beyond a threshold makes it harder to manage product-process interactions and strains the capability of manufacturing processes (Chakravarty and Ghose 1993;Bordley 1993).Standardized routings developed for product families cannot accommodate the highly customized metallurgical specifications without sacrificing conformance quality.Thus, manufacturing errors escalate if degree of customization exceeds the threshold level as shown in region 4 (Figure 2).Typical failure modes in this region 4 are listed in Table 1.This leads to the following hypothesis: Hypothesis 2. The probability of a manufacturing error increases with the degree of customization only beyond a certain threshold level called the manufacturing threshold.
Product complexity is defined as a property of an object with many interwoven element, aspects, details,  or attributes that make the whole object difficult to understand (Suh 1990).With higher levels of product complexity, the interdependencies among the product parameters are far more difficult to manage.Additionally, the management of product-application mapping and product-manufacturing process mapping becomes far more difficult to manage with products of higher complexity (El-Haik and Yang 1999).The deterioration of performance and conformance with increased degree of customization occurs much sooner with higher product complexity.It suggests that the thresholds mismatch as well as manufacturing as portrayed in Figure 2, are more restrictive with higher level of product complexity.This is reflected in the next hypothesis.
Hypothesis 3.Both the customization threshold and manufacturing threshold decrease with higher level of product complexity.

Variable Definitions and the Conceptual Model
For our study, we chose a random sample of over 3,000 rolls sold to 150 customers over a span of 2 years.We collected primary data at the roll level on the following variables.The transaction data such as shipment date and customer's rolling mill attributes were obtained from the MRP system.Other records for the MRP transactions were obtained from various data sources in the company.Description of the variables and their data sources are described below: • Degree of Customization (CUSTOM): This is the ratio of number of metallurgical parameters set by customer to the total number of metallurgical parameters required to specify roll in the product family.
• Failure Time (FTIME): It is the time between the shipping date and complaint date as measured in days.The MRP system provided shipment dates where as the customer complaint documents provided the complaint date.

Empirical Model Specification: Competing
Risks Approach The objective of our empirical analysis is to quantify the impact of design-to-order (DTO) process on the probability of roll failure in either mode.Specifically, we model separately the impact of the degree of customization (CUSTOM) and product complexity (PRODX) on the probability of both mismatch error and manufacturing error.As noted earlier, for each roll, the data contained information on whether the roll failed during the observation period, and in the case of roll failure, the cause and time of failure (mismatch or manufacturing).
The following data characteristics influenced our econometric model specification: (1) the dependent variable is a measure of the duration from one state to another, i.e., the duration between product shipment to product failure, (2) the presence of multiple causes of roll failure; in particular roll failure due to a mismatch error or manufacturing error, and (3) the presence of observations on rolls which did not fail during the observation period.In these cases, there is an upper truncation on the failure time.We employed the competing risk hazard function formulation (Kalbfleisch and Prentice 1983;Cox and Oakes 1984), which is routinely used in reliability and quality engineering to analyze product failure data (Juran and Gryna 1993).
The competing risk hazard function model involves a specification for failure time distributions.We have chosen the Weibull hazard functions, which are defined as: (3) where j ϭ 1 for Mismatch and j ϭ 2 for Manufacturing.
The competing risks hazard function model is estimated through maximum likelihood method (Kalbfleisch and Prentice 1983;Cox and Oakes 1984).

Incorporating the Impact of Explanatory Variables and Interpretation of Parameters.
As shown above, the Weibull density function is characterized by the two parameters ␥ j and j .We use the proportional hazard specification to incorporate the impact of explanatory variables on the hazard.The j for mis- match errors and manufacturing errors are specified as

Incorporating Mismatch and Manufacturing Thresholds.
As hypothesized earlier, the likelihood of a roll failure due to mismatch errors decreases with an increase in CUSTOM up to a threshold point, which we defined as the mismatch threshold ().However, for values of CUSTOM above , an increase in CUSTOM increases the failure probability.We incorporate this aspect by modifying equation ( 4): The mismatch threshold is not directly observable in the data.Therefore, the mismatch threshold () needs to be estimated endogenously.To capture these unobserved differences, we employ a random effect specification.This is accomplished by assuming to be Beta distributed with parameters and where Notice that the parameter is a function of product complexity as shown in (7).Therefore the mismatch threshold () is driven by product complexity.The expected mismatch threshold (( )) is the expectation of the Beta distribution, which is given by: Similarly, for the manufacturing errors we hypothesized, that up to a threshold level, any increase in CUSTOM does not increase manufacturing errors.We have defined this threshold to be the manufacturing threshold ().For values of CUSTOM above this threshold, however, increase in CUSTOM increases the failure probability.We incorporate this aspect by modifying equation ( 5): Similar to the mismatch threshold, we define the manufacturing threshold () to be Beta distributed with parameters and where The expected manufacturing threshold ( ) is the expectation of the Beta distribution, which is given by: The probability density functions can be calculated once the parameters of the hazard functions are estimated.The hazard function model specification and the expected impact of the explanatory variables on the probability of failures are summarized in Table 2.

Results and Managerial Implications
4.1.Specification Tests and Discussion of the Findings The maximum likelihood estimates, the log likelihood values of proposed models and specification tests results are presented in Table 3. Incorporation of the mismatch threshold and manufacturing threshold improves the explanatory power of the model significantly.The likelihood ratio test strongly supports the presence of the thresholds and rejects the baseline models.The significance of the mismatch threshold () can be observed through the significance of the parameters that influence the parameter of the beta distribution.Also, the parameters ␣ 1L and ␣ 1U are very significant.These results together provide a strong support for Hypothesis 1.The significance of the manufacturing threshold () can be observed through the significance of the parameters that influence the parameter of the beta distribution.Among the parameters of manufacturing errors, the parameter for CUS-TOM (␤ 1 ) is positive and significant, indicating an adverse impact of customer's technical specification on manufacturing errors beyond the manufacturing threshold ().These results strongly support Hypothesis 2.

Sensitivity of Mismatch and Manufacturing
Errors to the Degree of Customization The estimates of parameter values in the competing risk hazard function model are presented in Table 3. Figures 3 and 4 show the sensitivity of roll failure to varying degree of customization computed from the estimated parameter values.The curves reconfirm the hypothesized impacts of degree of customization and product complexity on the roll performances.
Consider the case of mismatch errors shown in Figure 3, which shows the probability of roll failure occurring within a year due to mismatch error by the customer.Two curves (for PRODX ϭ 4 vs. 2) are computed to show the adverse impact of product complexity.In both cases, initially, we observe that product performance gains are achieved by allowing the customer to select certain parameters.Thus mismatch errors decrease with higher levels of customization in both cases.These decreasing trends continue up to the mismatch threshold, beyond which the adverse impact of customization dominates and the mismatch errors start increasing.The graphs also highlight the added bur-den of product complexity.Two detrimental impacts are observed.First, rise in the error rate occurs much earlier as seen from the lower mismatch threshold (for PRODX ϭ 4 vs. 2).Second, the rise in the error rate after the threshold is steeper confirming that interaction of degree of customization and product complexity are far more severe after the threshold.
The managerial implication of the above phenomena is that customization should be encouraged only up to a point.Seeking information from the customer, who has a better knowledge of the rolling mill conditions, can significantly reduce mismatch errors.However, permitting too many choices to the customer results in a higher-level of mismatch.The complexity in the specification of the parameter values becomes overwhelming to the customer due to interdependencies.This is consistent with Zipkin's (2001) description of the limitations of elicitation process and its adverse impact in situations of high degrees of customization.In complex products, the adverse impact is far more pronounced.This suggests that the manufacturer should offer a limited choice set of parameters to the customer in cases of complex products.
Likewise, manufacturing errors are dormant up to a threshold as shown in Figure 4.The manufacturing system and the standardized process routings of the manufacturer can accommodate the customization up to the manufacturing threshold.Any changes by the customer beyond this threshold rapidly escalate the failures due to manufacturing errors.Product complexity aggravates the failure rate as shown by the comparison of two curves (PRODX ϭ 4 vs. 2).The threshold decreases and the failures escalate at a steeper rate for higher product complexity.The managerial conclusion from Figure 4 is that investments in flexible manufacturing and process engineering have allowed the roll manufacturer to permit customization up to a certain degree.However, there is a substantial price to be paid for over taxing the manufacturing systems to higher levels of customization.

Restricting Threshold-Is It Mismatch or Manufacturing?
The existence of both thresholds raises an important issue concerning which of the two thresholds is more restrictive.The size of the two thresholds is a function of product complexity as characterized in equations ( 8) and (11).Figure 5 presents a plot of the size of the mismatch and manufacturing thresholds for varying levels of product complexity using our estimated model.Our field site represents a manufacturing environment with high product complexity.The average product complexity was 3.28, which is to the right of crossover point A. We observe that in such cases mismatch threshold is lower than the manufacturing threshold for the complexity levels to the right of crossover point "A" in Figure 5.However, this also suggests that if the roll complexities were to be low and to the left of A, the relative sizes of thresholds would flip.The manufacturing threshold would then becomes more restrictive.
The managerial implication is that with low levels of product complexity, customization is hampered by the ability of the manufacturing systems to handle the choices made by the customer.From an investment standpoint, our study underscores the need to invest in flexible manufacturing system and process engineering for environments represented by the region to the left of the crossover point A. However, we recom-mend investment in eliciting technologies and frontend process technologies, to the right of the crossover point A. Capabilities of the front-end process technologies rather than the manufacturing flexibility, limit the level of customization achievable right of crossover point A. In such cases, front-end process technologies such as configurators, can assist the customer in handling a larger set of interdependent parameters set.
Our results provide another plausible explanation for the unexpected Gupta and Somers (1996) findings of a negative relationship between process flexibility and company's performance.Manufacturing flexibility is only one of the dimensions of customization capabilities.Achieving higher manufacturing flexibilities may not directly translate into higher market performance due to the limitations of the front-end process.Enhancement of the capabilities of the front-end processes is very essential to fully realize the benefits of higher manufacturing flexibility in complex product environments.Our research thus identifies other variables that influence the relationship among customization capability, manufacturing flexibility, and market performance.We want to highlight the fact that we have carefully examined the benefits of flexibility and manufacturing strategy initiatives from a customization capability point of view.However, we stress that manufacturing flexibility and manufactur- ing strategy literature address a number of other issues and objectives that are not specifically considered in this research (Clark 1996;Skinner 1996;Hayes and Pisano, 1996;Gupta and Somers, 1996).
We also want to highlight the fact that product complexity only partially explains the positioning of the two thresholds.A number of other factors can also drive the positioning of the two thresholds.Table 4 presents a list of potential drivers of the thresholds based on the literature (Ramdas 2003;Von Hippel 2001;Thomke and Von Hippel 2002).We have not included these drivers into our model due to limitation of the data.Understanding the impact of these drivers on the thresholds would be valuable in better designing of the customization process.

Conclusions
The success of customization depends upon providing low costs of customization while achieving higher product quality simultaneously.In this research, we have analyzed the impact of customization on product quality at a manufacturing plant for an industrial component.The research site was a design-to-order production environment where each order was custom designed.We found that if errors were made during the elicitation of customer requirements, the product redesign to match the requirements and manufacturing led to higher failures in the field.These failures, classified either as mismatch or manufacturing, were driven by two factors: degree of customization and product complexity.
A competing risks hazard function model was estimated using extensive data on field failures.Mismatch failures were found to decrease initially by allowing the customer to make choices, establishing a case where customization was beneficial.The data also revealed a threshold, which we defined as mismatch threshold, beyond which customization was detrimental.Mismatch errors increase significantly if customization was carried to levels beyond the threshold.Prod-uct complexity reduced the threshold and also amplified the negative impact of higher levels of customization.Customization also impacted manufacturing errors beyond a threshold.Up to this threshold, the manufacturing systems were flexible to cope with the customization choices made by the customer.The existence of the two thresholds and their relative sizes has implications for investment choices in improving the customization capabilities of the roll manufacturer.Our research thus developed a rigorous framework to address the two key unexplored research questions identified by Ramdas ( 2003) from a quality perspective: (1) what degrees of customization to offer, and (2) how to customize the product design process.
We exercise caution in generalizing our results and point out some limitations of this research.First, we have validated the concepts based on the data from one manufacturing firm.Examination of the customization processes and their impact on quality in other environments is necessary to further confirm and replicate our results.Second, the research site represents an environment in which the components are expensive and the errors are rather objectively assessed.However, customer expectations may determine the mismatch curve in a product environment where product performance is more subjective.Finally, we have used probability of errors as the measure of product performance because of the unavailability of cost data.The relative positions of the curves and their shapes might change substantially if costs of field failures were modeled.
To sum up, we have developed a rigorous framework to analyze the impact of customization on quality in an industrial component market based on primary data.The insights of this study have a major implication on the implementation of customization from both marketing and manufacturing viewpoints.Our research, we believe, is one step towards a better understanding of the drivers of enhanced customization performance.
Figure 1Customization at the Research Site Figure 2Impact of Customization on Field Failures

•
Cause of Failure (CAUSE): This variable indicates the cause of failures.The resolved failure modes were classified into mismatch and manufacturing errors.This is a categorical variable (0 -not yet failed, 1-mismatch failure, and 2-manufacturing failure).The failure mode data was obtained from the internal documentation maintained by the engineering department.• Product Complexity (PRODX): This variable indicates the metallurgical complexity of the rolls on a scale of 1 to 5. Failure times for mismatch errors and manufacturing errors are the dependent variables in the model.The objective of our empirical model is to understand the relationship between the degree of customization (CUSTOM) and product complexity (PRODX) on the probability of a mismatch error and manufacturing error.The relationship between the dependent and the independent variables are expressed as follows: Probability of a Mismatch Error ϭ f͑CUSTOM, PRODX͒ (1) Probability of a Manufacturing Error ϭ f͑CUSTOM, PRODX͒ (2) Figure 3 Sensitivity of Mismatch Errors to CUSTOM and Product Complexity Figure 5 Sensitivity of Mismatch and Manufacturing Thresholds to Product Complexity

Table 4 Drivers of Mismatch Thresholds
Customization: Impact on Product and Process Performance