Architecting Option Content

This paper presents an approach to determine the proper number of levels required on independent product architectural attributes, given their ability to generate added revenue through more direct targeting to smaller segments, and given the added costs of doing so. This is done in as simple and readily implementable manner as possible, making use only of conjoint data and cost estimates. From this, the order in which to consider added breakouts across the different attributes are prioritized. From this, for any minimum level of profit worth considering, a set of attribute levels to offer on each architectural attribute can be selected. Then, for any selected set of attribute levels to offer, the most effective product family using those levels is determined from the permutations. * Corresponding Author : 77 Massachusetts Ave, Room 3-449B, Cambridge, MA 02139. knotto@mit.edu Introduction Product families based upon a common product architecture are often used in many industries to leverage the common systems while yet tailoring some aspects of a product to more adequately match the product with a customer’s application. Yet, even in industries with very slow technology and market change rates, known demand volume, and known costs, there is no formulation of the optimal portfolio architecture to maximize profit. In this paper, we develop this formulation, given market variety demands as modeled through conjoint data and complexity cost coefficients for added levels of offering on attributes. The formulation presented requires data with modest requirements to develop or estimate. We derive our model from basic assumptions about both the market and costs of production/development. From the derivation, a methodology is developed with two fundamental analysis components. First, for a given minimum profit level to bother with, the best permutative set of architectural attribute levels is determined. Then, when restricted to this set of levels, the product family constructed from that set that most effectively meets the market demand is determined. We operate with design-independent product architecture attributes. Two architectural attributes are designindependent if the numbers of levels to offer of each does not impact the design of the other. For example, when system architecting an automobile, the passing acceleration, interior price, and seat height are design-independent. On the other hand, primary customer need variables such as towing capacity and passing acceleration are not, since both depend on the design choice of engine size. Engine size and vehicle weight are design independent (though perhaps there are rough interval constraints between them), and cover the concerns of passing acceleration and towing capacity. Design independence is fundamentally concerned with and determined by the product architecture, not on any statistical market considerations. Design independent variables are the ones that the initial systems engineering of a product family must operate with when determining numbers of levels on architecture attributes. The traditional approach to determining the variety content capability of a product family architecture is to qualitatively flow down variety requirements from a market segment attack approach. That is, a marketing group determines segments to attack, and defines ideal product attributes for each segment that should be constructed. Typically, this does not consider design, production and supply chain constraints, and so a negotiation ensues to define the actual product attributes. Our approach here is to do these activities quantitatively. We do this in two steps using the underlying market data – first we determine the number of levels on each attribute, then second determine the offered family from this set. To understand this, consider the more common market segment attack approach. Analytically, the idea is to develop conjoint data of market share, cluster analyze it to define market segments, the centroids of which generally define Otto, Architecting Option Content 2 architectural attribute targets for design teams. Unfortunately and commonly, however, this gives rise to excess product architecture complexity. For example, consider two attributes. An effective segmentation may result in three clusters, and so three products are to be developed, each with three separate sets of architectural attributes. This generally forces the design team to develop a product portfolio capable of 9 permutations of the architectural attributes – 3 levels over 2 attributes. Therefore, with this approach, excess complexity enters. Further, the product portfolio tends to grow, since the platform design can relatively easily accommodate the nine product configurations, they are often offered, when in fact seven are not particularly worthwhile. The result is that companies offer overly complex product families, when a smaller set of levels on some architectural attributes would have been more profitable. Documented common examples of such product proliferation include film (Jaime, 1998), electronics assembly (Mosher, 1999), photocopiers (Zamirowski, 1999), vehicle option content on everything from rear end differentials to interior content (Roberson, 1999), and commercial aircraft models (Weir, 2000), among many others. No industry has an effective set of analyses to simultaneously manage both revenue from variety and costs from complexity. Related Work The development of product families built on product platforms and shared modules has been the subject of much recent research. Meyer and Lehnerd (1997) have done extensive case studies on platforms, pointing out their advantages and challenges, and demonstrating their ability to save costs. Other researchers such as Sanderson and Uzumeri (1995) and Henderson and Clark (1990) have also shown that the use of platforms has given companies an edge on the number of products they can offer and on their profitability over their competitors. Other management research has shown different approaches on managing the planning and use of platforms (Wheelwright and Clark, 1992; Erens and Verhulst, 1996; Robertson and Ulrich, 1998; Pedersen, 1999; Pulkkinen et al., 1999). To determine marketshare for various products, conjoint analysis has long been studied and developed. Ben-Akiva and Lerman (1997) provide a useful reference for conducting conjoint studies, as does Urban and Hauser (1993). Sawtooth Inc. (2000) has multiple software tools for forming conjoint studies to question customers to determine buying preference structures, to which product attributes and positioning can be optimized. These works, however, ignore the costs of offering the combinations of levels on each attribute. For example, to meet three different products over 2 attributes, often three levels are demanded on each attribute, resulting in an architecture that is actually capable of nine different products. Portfolio complexity thereby grows. On the other hand, conjoint studies do provide a good estimate of marketshare, given the utility functions of a representative sample of the market. Given a set of offerings, software models exist to calculate market share, based upon the orderings of the offerings for each sample point (customer) in the market model. Most related to the work here are efforts to determine a most effective set of product offerings, given conjoint data. Moore et al (1999) consider conjoint data and cost coefficients for each attribute to offer multiple levels. They consider whether to platform a variable (1 level) or not (multiple levels), and rank order the attributes. In this work, we extend these thoughts to multiple levels and define a break out sequence. In the engineering design literature, one can find several design and manufacturing strategies for offering variety that begin with commonality metrics (Martin and Ishii, 1997; Kota and Sethuraman, 1998). Krishnan et al. (1998) developed network models to design families of products that are measured along a single performance criterion. Optimization approaches have been developed by Gonzalez-Zugasti et al. (1998) and Nelson et al. (1999) to design product platforms and families of variants. Another optimization approach is used by Fujita et al. (1999) for designing a family of products from catalogs of existing swappable modules. These approaches are all downstream component form development exercises as compared to the portfolio architecture work here. We next develop an analysis to define the profit extractable from any choice of levels on architectural variables. We do this by first developing a revenue model, subsequently a cost model, which then defines a profit model. Then we develop this into a method to analyze the profit capability of any breakout choice. Given this choice, we then define the portfolio to offer from the set of level permutations. Profit from Variety Determining the ideal family based upon profit maximization will involve two different models, a revenue model based upon conjoint data, and a cost model based upon cost-of-complexity factors of production. Otto, Architecting Option Content 3 Revenue from Variety Revenue from a Customer Consider a market S of customers sk. Define the price that customer k will pay as a function of desired architectural attribute levels τik as Pk, where each architectural attribute is denoted xi. Expanding Pk in a Taylor series about k τ v , ( ) ( ) ∑ ∑ + − ∂ ∂ + − ∂ ∂ + = =