Case study — automotive multiplexing

226 L.O. Morgan, R.L. Daniels Journal of Operations Management 19 2001 219–238 identify the relevant breakpoint volume as a function of the primary decision variables y j k and s j t k .

4. Case study — automotive multiplexing

In this section, we first describe the analysis under- taken by DE and its automotive customer to evaluate automotive multiplexing as an alternative technology for mechanizing vehicle option content. The approach outlined is representative of the type of analysis per- formed to evaluate new technologies in the automotive industry, and is similar in focus to standard finan- cial analyses used for technology adoption decisions across industries. We, then discuss shortcomings of the standard analysis approach, motivate the need for a broader context for technology adoption decisions, outline disguised data obtained from DE, and discuss the relationship between problem structure and the efficiency of the associated solution process. Cost studies developed by DE for assessment of the multiplexing technology focus on identification of the level of content complexity where multiplexing is more cost effective than the conventional technology where again content complexity increases as more features are added to a vehicle. In the study, five lev- els of vehicle content complexity were identified for costing purposes, as outlined in Table 1. As the level of content complexity in a vehicle increases from one level to the next, additional LCM hardware is required for multiplexing; thus, the volume and cost of ad- vanced technology components required for mecha- nization are greater for higher content levels. Within a level of content complexity, specific subsets of options OP use the same LCM hardware, but may require a slightly different number of connectors and per- haps minimal additional wiring, yielding minor cost increases for increasing content within a level. The study examined both a high-volume and low-volume demand scenario in order to capture the impact of volume on cost. Consistent with industry estimates, DE engineers assumed an 80 experi- ence curve for the multiplexing technology, yielding LCM module costs that decrease by 20 each time production volume doubles. The standard experience curve relationship was applied to compute costs, i.e. C v = C 1 v − b , where C v represents the cost of the vth unit and b = 0.322, consistent with an 80 experi- Table 1 Levels of content complexity Level 1 Power windows Power locks Courtesy lamps AMFM cassette Cruise control Rear window defogger Air conditioning Level 2 All Level I features Six-way power seats Power mirrors Level 3 All Level II features Express down windows Auto door locks Illuminated entry Level 4 All Level III features Universal theft deterrent UTD Memory eight-way power seats Adjustable lumbar support Memory power mirrors Level 5 All Level IV features Easy close doors Express up windows Personalized remote keyless entry RK Heated seats Shoulder belt height adjustment Parking assist mirror adjustment Dual zone HVAC ence curve. Based on historical demand data, the cost study further assumed that 60 of the total volume requires Levels 1 and 2 content complexity, 25 Level 3, and 15 Levels 4 and 5. Two categories of unit cost elements are relevant for multiplexing, the cost for the LCM hardware and the cost of the conventional components required to connect LCM modules, and experience effects apply only to LCM hardware costs. Beyond unit produc- tion costs, the study identified investment costs, e.g. to modify vehicle designs and to develop the tooling required to produce LCM modules, that must be in- curred to manufacture multiplexed vehicles. The study also identified significant cost benefits associated with multiplexing that reflect cost synergies in engineer- ing, increased reliability, reduced warranty costs, and enhanced diagnostics for service and safety. In order to facilitate unit-cost comparisons with the incumbent technology, the study aggregated all investment costs and cost benefits over a 3-year planning horizon, and L.O. Morgan, R.L. Daniels Journal of Operations Management 19 2001 219–238 227 Table 2 Unit cost comparison results from Delphi cost study Level Low-volume High-volume L1 Penalty US 49 Penalty US 21 L2 Penalty US 125 Penalty US 29 L3 Penalty US 173 Penalty US 61 L4 Penalty US 91 Savings US 51 L5 Penalty US 48 Savings US 103 allocated them uniformly across an anticipated vol- ume of vehicles, resulting in a per unit cost penalty associated with multiplexing for each content level. The focus of the cost study was to determine whether multiplexing results in a cost penalty or cost savings over the conventional technology for each level of content complexity. Table 2 presents the results of the unit-cost comparisons generated in the multiplexing cost study. As indicated in Table 2, the multiplexing technology is cost-justified only for content Levels 4 and 5 under the high-volume sce- nario. The study concluded that while multiplexing should become increasingly attractive as the average level of content complexity in vehicles increases, the new technology should not be adopted at the time of the study. Rather, the study recommended that an implementation scheme be developed as part of the firm’s strategic planning process, and that the cost-effectiveness of multiplexing be evaluated for each new electronically controlled feature developed by the firm. Our discussions with management indicated that new technologies such as multiplexing are tradition- ally adopted in an incremental fashion to hedge the risks of large-scale investments in retooling without volume commitments. The incremental adoption ap- proach is consistent with the traditional unit cost-based focus of technology adoption cost studies in the indus- try, but has significant limitations that act to conceal the true profit implications of alternative technology adoption decisions. Frey 1991 provides excellent ex- amples of incremental adoptions of features in auto- mobiles based on his experiences with Ford Motor company, and discusses the need for product champi- ons to get new features and technologies into automo- biles. The model developed in this research provides a framework for developing the required support for promising new technologies. Several factors motivate a more thorough analysis of automotive multiplexing. First, given the experience curve rates assumed relevant for technologies like mul- tiplexing in the automobile industry, the cost of mul- tiplexing depends significantly on the total amount of LCM hardware used for content mechanization across all levels of content complexity. In turn, the cumula- tive production volume of LCM hardware is a func- tion of both the mix of OP offered by the firm across all levels of content, and the demand for those OP. In addition, hybrid policies that allow any combination of content levels to use multiplexing e.g. a selected mix of moderate- and high-content vehicles, should be evaluated to more accurately determine the poten- tial impact of adopting the advanced technology on the firm’s overall costs and profits. Finally, efforts by automobile manufacturers to restructure their vehicle divisions to eliminate duplication of effort and facil- itate coordination across similar vehicles Wall Street Journal, 1994, 1996b, c potentially increase the rel- evant volumes of LCM hardware, further decreasing the marginal cost of multiplexing. As discussed in Section 3, one factor important in the evaluation of new technologies is the experi- ence effect. While traditional analysis often includes explicit consideration of experience as, e.g. in the DE cost study, the lack of coordination with product mix decisions and the focus on incremental adoption i.e. failure to consider the implications of coordinated adoption across content levels prevent identification of the true cost and profit implications of alternative adoption decisions. In fact, because of the significant experience effect governing LCM hardware cost, the profit maximizing assignment of mechanization tech- nology k = 1 or k = 2 to a particular option package is strongly influenced by the degree to which other se- lected OP use the multiplexing technology, as shown below. 4.1. Implications of volume-related experience benefits for multiplexing The cost studies summarized in Section 3 indicate that multiplexing costs less than conventional mecha- nization for the highest content vehicles only when a sufficient volume of those high-content, multiplexed vehicles are produced and sold. The cost of using multiplexing for lower content vehicles is shown to 228 L.O. Morgan, R.L. Daniels Journal of Operations Management 19 2001 219–238 Table 3 Option packages by content complexity L1–L5 Level Option packages Economy vehicle Luxury vehicle Level 1 L1-I Courtesy lamps, AC, AMFM cassette OP1 – L1-II L1-I plus: rear window defogger, cruise control OP2 – L1-III L1-I plus: power windows, power locks OP3 – L1-IV All Level 1 options OP4 – Level 2 L2-I L1-IV plus: power mirrors OP5 – L2-II L2-I plus: six-way power seats OP6 – Level 3 L3-I L2-II plus: auto door locks OP7 OP1 L3-II L3-I plus: illuminated entry OP8 OP2 L3-III L3-II plus: express down windows OP9 OP3 Level 4 L4-I L3-III plus: memory power mirrors – OP4 L4-II L4-I plus: eight-way power seats, adjusstment lumbar support – OP5 L4-III L4-I plus: universal theft deterrent UTD – OP6 L4-IV All Level 4 options – OP7 Level 5 L5-I L4-IV plus: dual zone HVAC, heated seats – OP8 L5-II L5-I plus: shoulder belt Ht. adjustment, parking assist mirror adjustment, personalized remote key entry RKE – OP9 L5-III L5-I plus: easy close doors, easy up windows – OP10 L5-IV All Level 5 options – OP11 be higher than conventional mechanization, but this cost penalty is calculated under the assumption that volume-related experience is accumulated only within a content level, with different content levels evaluated independently. In practice, the marginal cost for us- ing multiplexing depends on the cumulative amount of LCM hardware used for all multiplexed OP. When cumulative volumes are explicitly factored into the analysis, using multiplexing on lower content vehicles may be attractive, even when the technology cannot be cost justified for individual low-content levels. To illustrate, suppose five OP are offered, represent- ing each of the five levels identified in Table 3. Fig. 1 depicts unit costs realized when either the conven- tional C or multiplexing M technology is used to mechanize all option content levels, under the assump- tion that all levels of option content use the same base LCM hardware for multiplexing mechanization. Each multiplexing cost curve Mv depicts the average unit cost incurred if an 80 experience curve is relevant and vehicle volumes are uniformly distributed across the five content levels. For a given content level, the Fig. 1. Relationship between content complexity and unit mecha- nization costs when base LCM hardware is sufficient for all com- plexity levels. L.O. Morgan, R.L. Daniels Journal of Operations Management 19 2001 219–238 229 Fig. 2. Relationship between content complexity and unit mech- anization costs when additional LCM hardware is required for higher complexity levels. set of multiplexing cost curves reflects the decreas- ing unit costs realized when the volume of vehicles manufactured with the advanced technology increases Fig. 2. Fig. 3. Economy vehicle mechanization costs under alternative policies. Figs. 3 and 4 show the unit costs and total costs associated with four policies that incorporate both mechanization technologies, based on 1997 demand forecasts for economy and luxury vehicles, respec- tively. These figures illustrate the importance of in- tegrating mechanization technology decisions across content levels. For example, in Fig. 3, the total cost of a policy that offers all economy vehicle content levels with the conventional technology CL1, L2, L3 is less than the total cost of mechanizing option content levels L1 and L2 conventionally and content level L3 with multiplexing CL1, L2ML3. Based on this comparison, multiplexing would not be adopted. However, if both content levels L2 and L3 use mul- tiplexing and level L1 uses conventional technology CL1ML2, L3, or if all three levels use multiplex- ing ML1, L2, L3, then the combined volumes are sufficient to reduce total cost below that for policy CL1, L2, L3. Similarly, note in Fig. 4 that if we ex- amine unit costs for luxury vehicle content level L3, we conclude that multiplexing is more costly than the conventional technology, even when all levels of con- tent are multiplexed, ML3, L4, L5. However, total costs are minimized under this mechanization policy due to the reduced costs for L4 and L5 that result from the contribution of L3 vehicles to total LCM volume. 230 L.O. Morgan, R.L. Daniels Journal of Operations Management 19 2001 219–238 Fig. 4. Luxury vehicle mechanization costs under alternative policies. 4.2. Demand and cost data for model application Two vehicle platforms are included in the analysis, corresponding to economy and luxury vehicles, re- spectively. Marketing information from DE indicates that option content for the economy vehicle ranges from L0 no significant option content to L3, while option content for the luxury model ranges from L3 to L5. A total of seventeen candidate OP are identified in the analysis to represent varying degrees of complex- ity across the five levels of option content outlined in Table 1. Table 3 specifies the OP associated with each Fig. 5. Distribution of demand across candidate OP for economy vehicle. Fig. 6. Distribution of demand across candidate OP for luxury vehicle. level of content complexity, and indicates the type of vehicle economy or luxury offered with each option package. Demand forecasts for the next model year were ob- tained for the specific economy D = 200,000 units and luxury D = 40,000 units vehicle platforms, with the anticipated distribution of vehicle demand across the designated levels of option content as indi- cated in Figs. 5 and 6. According to these estimates, 15 of all consumers purchasing an economy vehicle will select the base vehicle with no significant content represented by OP0 in Fig. 5. In addition to the base L.O. Morgan, R.L. Daniels Journal of Operations Management 19 2001 219–238 231 Table 4 Price, cost, and demand data for economy vehicle OPj p j US c j 1 US c j 2 US r j B j US Single-model demand D = 20000 Category demand D = 1M d j t d j t OP1 240 120 120 1.0 27 20000 100000 OP2 280 140 125 1.0 27 20000 100000 OP3 320 160 130 1.0 27 20000 100000 OP4 340 170 132 1.0 27 20000 100000 OPS 360 180 135 1.5 41 20000 100000 OP6 390 195 140 1.5 41 20000 100000 OP7 430 215 145 2.4 65 18000 90000 OP8 450 225 155 2.4 65 16000 80000 OP9 480 240 165 2.4 65 16000 80000 vehicle, n = 9 OP capturing the remaining 85 of forecasted demand are considered for the economy vehicle. The luxury vehicle will be offered with con- tent from levels L3–L5, with n = 11 OP OP1–OP11 in Fig. 6 considered in the analysis. We consider two demand scenarios for each vehicle category. In the first scenario, total annual demand is set according to the single-model estimates obtained for the economy and luxury vehicles. The second de- mand scenario extends the volumes to include multiple vehicle models in the same category, since automobile manufacturers typically offer a variety of economy and luxury vehicles e.g. General Motors GM of- fers several different economy vehicles, including the Pontiac Sunbird and Grand Am, Chevrolet Cavalier, Buick Skylark, and Saturn SL1, and has announced plans to develop a small-car group to exploit synergies Table 5 Price, cost, and demand data for luxury vehicle OPj p j US c j 1 US c j 2 US r j B j US Single-model demand D = 40000 Category demand D = 200000 d j 1 d j 2 d j 3 d j 1 d j 2 d j 3 OP1 430 215 145 2.4 65 4400 4000 3600 22000 20000 18000 OP2 450 225 155 2.4 65 3600 3200 2800 18000 16000 14000 OP3 480 240 165 2.4 65 3600 3200 2800 18000 16000 14000 OP4 520 260 175 3.4 92 3600 3200 2800 18000 16000 14000 OP5 700 350 205 3.4 92 3600 3200 2800 18000 16000 14000 OP6 920 460 235 3.4 92 3600 3200 2800 18000 16000 14000 OP7 1100 550 260 3.4 92 3600 3200 2800 18000 16000 14000 OP8 1300 650 290 4.0 108 3600 4400 4800 18000 22000 24000 OP9 1460 730 310 4.0 108 3600 4400 4800 18000 22000 24000 OP10 1620 810 330 4.0 108 3600 4000 4800 18000 20000 24000 OP11 1800 900 350 4.0 108 3200 4000 5200 16000 20000 26000 across similar vehicle platforms, Wall Street Journal, 1994. To capture the volume benefits gained by co- ordinating manufacturing across similar vehicles, cat- egory demand is set at five times the annual demand estimates provided for the individual vehicle models in each category. Clearly, the configuration of manu- facturing facilities and the level of coordination across vehicle divisions and brands are important factors in determining whether or not experience benefits are shared across similar vehicles. In Table 4, d j t denotes the annual demand for econ- omy vehicles with option package j in year t under each demand scenario. Similar demand information is provided in Table 5 for the luxury vehicle category, with demand figures reflecting a shift in luxury ve- hicle demand toward higher content levels for future years as anticipated by DE marketing research this 232 L.O. Morgan, R.L. Daniels Journal of Operations Management 19 2001 219–238 Table 6 Alternative product substitution rates a 90 substitution 80 substitution OPj is offered All 100 customers purchase OPj All 100 customers purchase OPj OPj is not offered 10 customers choose not to purchase vehicle 20 customers choose not to purchase vehicle 90 customers select similar alternative b 80 customers select similar alternative b a Suppose 100 customers have option package j OPj as their first choice. b If customer’s second-choice option package is also not offered, then the customer does not purchase vehicle. trend is not expected for the firm’s economy mod- els. Including demand trends in the model is signifi- cant, since the optimal product mix and volume may change over the planning horizon accordingly. In this case, the estimated shift in preferences toward higher content is conservative. By applying the model under alternative assumptions about changing customer re- quirements, additional insights into the potential im- pact of alternative technology adoption decisions can be generated. Revenue information is captured using disguised prices p j for each candidate option pack- age, as shown in Tables 4 and 5 for the economy and luxury vehicles, respectively. To be consistent with the framework of the cost study outlined in Section 3, we use a 3-year planning horizon in our analysis and ag- gregate the demands in Tables 4 and 5 into a single period demand volume. We examine two mechanisms for distributing demand across alternative OP when one or more candi- date OP are left out of the product mix, thereby allow- ing the sensitivity of customer demand to first-choice option content availability to be varied. In the first scenario, 90 of the customers with first-choice de- mand for option package j OPj on a given vehicle can be satisfied with a similar second-choice option package e.g. OPj ± 1 or OPj ± 2 if their first choice OPj is not offered. Thus, 10 of the de- mand for OPj is lost if OPj is not offered. Additional demand is lost if a second-choice option package is also absent from the firm’s product mix. Under the second scenario, 80 of customers will purchase the vehicle with an alternative option package if their first-choice option package is not offered, as illus- trated in Table 6. Table 7 shows the second-choice distribution of demand under each scenario for the economy vehicles, where α jj ′ is the percentage of OPj ′ ’s demand captured by OPj if OPj but not OPj ′ is offered. Second-choice distribution of demand for the luxury vehicles follows an identical pattern i.e. OP1, OP2, OPn − 1, and OPn follow the same pattern for either vehicle type, as do OP3 through OPn − 2. There is evidence to support the assumption that customer demand is not extremely sensitive to spe- cific content preferences. Recent efforts by GM to offer a simpler, more limited menu of OP won the ap- proval of customers in California Wall Street Journal, 1996a, and GM is expected to extend use of the new value pricing approach, citing customers’ preferences for more consistent pricing and a haggle-free buying experience. Customers’ willingness to sacrifice some degree of choice flexibility in favor of a simplified buying experience is also supported in the academic literature see, e.g. Purohit and Sondak, 1999. Tables 4 and 5 provide cost data for each option package listed in Table 3. Specifically, c j k represents the cost of producing and assembling conventional components for conventional k = 1 and multiplex- ing k = 2 mechanization of option package j, r j reflects the relative amount of LCM hardware needed to mechanize option package j with multiplexing, and B j is the net investment cost penalty as described in Section 3. Setup and holding costs can also affect the product mix and technology choice decisions. In this case, multiplexing results in significantly lower setup times to change over between OP. As discussed in Sec- tion 2, these efficiencies result from the simplified in- terface for option content mechanization, and become increasingly significant for higher content levels since the conventional technology requires selection and in- tegration of an increasing array of wiring and electron- ics for assembly of higher content vehicles. Fixed cost estimates that capture setup and holding cost implica- tions are provided in Table 8 for each total demand scenario. L.O. Morgan, R.L. Daniels Journal of Operations Management 19 2001 219–238 233 Table 7 Second-choice demand distribution OPj ′ OPj OP1 OP2 OP3 OP4 OP5 OP6 OP7 OP8 OP9 Scenario 1 α jj ′ OP1 0.00 0.80 0.10 0.00 0.00 0.00 0.00 0.00 0.00 OP2 0.45 0.00 0.40 0.05 0.00 0.00 0.00 0.00 0.00 OP3 0.05 0.40 0.00 0.40 0.05 0.00 0.00 0.00 0.00 OP4 0.00 0.05 0.40 0.00 0.40 0.05 0.00 0.00 0.00 OP4 0.00 0.00 0.05 0.40 0.00 0.40 0.05 0.00 0.00 OP6 0.00 0.00 0.00 0.05 0.40 0.00 0.40 0.05 0.00 OP7 0.00 0.00 0.00 0.00 0.05 0.40 0.00 0.40 0.05 OP8 0.00 0.00 0.00 0.00 0.00 0.05 0.40 0.00 0.45 OP9 0.00 0.00 0.00 0.00 0.00 0.00 0.10 0.80 0.00 Scenario 2 α jj ′ OP1 0.00 0.70 0.10 0.00 0.00 0.00 0.00 0.00 0.00 OP2 0.40 0.00 0.35 0.05 0.00 0.00 0.00 0.00 0.00 OP3 0.05 0.35 0.00 0.35 0.05 0.00 0.00 0.00 0.00 OP4 0.00 0.05 0.35 0.00 0.35 0.05 0.00 0.00 0.00 OP4 0.00 0.00 0.05 0.35 0.00 0.35 0.05 0.00 0.00 OP6 0.00 0.00 0.00 0.05 0.35 0.00 0.35 0.05 0.00 OP7 0.00 0.00 0.00 0.00 0.05 0.35 0.00 0.35 0.05 OP8 0.00 0.00 0.00 0.00 0.00 0.05 0.35 0.00 0.40 OP9 0.00 0.00 0.00 0.00 0.00 0.00 0.10 0.70 0.00 4.3. Problem simplification For problems involving only a moderate number of OP, the formulation can be solved using standard in- teger programming solvers. For larger problems, the cost structures associated with the conventional and multiplexing technologies permit the development of Table 8 Fixed cost data OPj Economy vehicle OPj Luxury vehicle D = 200000 D = 1000000 D = 40000 D = 200000 f j 1 US f j 2 US f j 1 US f j 2 US f j 1 US f j 2 US f j 1 US f j 2 US OP1 397251 198625 1986254 993127 OP1 194918 123277 974592 616386 OP2 463459 231730 2317296 1158648 OP2 182449 115391 912245 576955 OP3 529668 264834 2648339 1324169 OP3 194612 123084 973062 615418 OP4 562772 281386 2813860 1406930 OP4 210830 133341 1054150 666703 OP4 595876 297938 2979381 1489691 OP4 283810 179497 1419048 897485 OP6 645533 322766 3227663 1613831 OP6 373007 235910 1865035 1179552 OP7 675217 337608 3373084 1688042 OP7 445987 082067 2229933 1410334 OP8 666210 333105 3331049 1665525 OP8 618050 390889 3090252 1954447 OP9 710624 355312 3553119 1776560 OP9 694118 438999 3470591 2194994 OP10 734343 464440 3671717 2322198 OP11 815937 516040 4079686 2580220 a simplified decision rule for assigning the profit max- imizing mechanization technology to selected OP. The following result establishes the consequent relation- ship among the y j k decision variables, indicating that if option package j ′ ∈ L is offered with the multiplex- ing technology, then any OP j j ′ , j ∈ L offered on the vehicle will also use multiplexing. 234 L.O. Morgan, R.L. Daniels Journal of Operations Management 19 2001 219–238 Result: For content complexity level L, if y j ′ 2 = 1, j ′ ∈ L, then y j 1 = 0 for all j j ′ , j ∈ L. A proof is provided in Morgan and Daniels 1999. The result allows the following constraint to be added to the formulation: J L X h=j +1 J h,1 ≤ J L 1 − y j 2 , j ∈ L, L = L1, . . . , L5, 4 where J L denotes the index of the option pack- age with the highest content in level L. Constraint set 4 significantly reduces the number of product linesmechanization technology pairs that must be evaluated to determine the optimal solution. Specif- ically, without constraint set 4, the total number of possible solutions is a combination of the to- tal number of possible product lines and the two possible mechanization technologies that can be as- signed to each option package in the product line, or P n i=0 \ begin{array}{l}n\\i\\\end{array} × 2 i . For the luxury vehicle, the n = 11 candidate OP translate into 177,147 possible solutions. Constraint set 4 reduces the number of solutions within each content level to P n L i=0 n L i i + 1. For the luxury vehicle example with n = 11 OP in levels L3–L5, the number of possible solutions is decreased to Q L5 L=L3 P n L i=0 n L i i + 1 , or 46,080. Thus, the result reduces the solution space to 26 of its original size. Note that because the amount of LCM hardware required to mechanize option content increases across levels L1 through L5 as reflected by increasing values of r j in Tables 4 and 5, the model can be simpli- fied only within each content level. However, a more general technology adoption context might be char- acterized by a unit cost advantage for the advanced technology that is strictly increasing with product complexity across all products. In this case, constraint set 4 would apply to the entire set of candidate prod- ucts, and thus would have an even greater impact on the number of feasible solutions for the luxury vehi- cle example, the number of possible solutions would be reduced to 13,312, or 7.5 of the original solution space. Note also that for the more general structure, the model simplification result allows the problem to be decomposed into n smaller problems one for each value of j, each requiring 2n fewer integer variables. Such a problem separation would be useful for larger applications where problem tractability is an issue.

5. Insights into the technology adoption decision