Dynamic Relationship Marketing: Understating Relationship State Migration Strategies ABSTRACT

Dynamic Relationship Marketing: Understating Relationship State Migration Strategies ABSTRACT Using a multivariate hidden Markov model applied to a six-year longitudinal data set of 346 business-to-business relationships maintained by a Fortune 500 firm, the authors identify four latent buyer seller relationship states by modeling changes in each customer s levels of commitment, trust, dependence, and relational norms over time. Customers movements among relationships states are parsimoniously captured by three positive (exploration, endowment, recovery) and two negative (neglect, betrayal) relationship migration mechanisms. At each stage in the relationship, firms can actively tailor their relationship migration strategies to promote customer migration to higher performance states and prevent deterioration to poorer performance states. A counterfactual elasticity analysis also offers a comparison of the relative importance of different migration strategies at various stages in the relationship. This research thus moves beyond extant relationship marketing literature that focuses on relationship states to explicate relationship state changes, and it provides managerial guidance related to the strategies that are most effective at promoting long-term, interfirm relationship performance. 1

Understanding and managing customer relationships is fundamental to marketing. Accordingly, firms spend in excess of $12 billion annually on customer relationship management, in efforts to understand how to target and sell to customers across various relationship stages (Gartner Research 2012). A substantial body of research in the relationship marketing (RM) domain has proposed multiple relational constructs and frameworks to better understand the nature of the buyer seller relationship (Mullins et al. 2014; Samaha, Beck, and Palmatier 2014). Yet much of this literature treats relationships as temporally homogenous, implying that all relationships respond in similar ways to RM initiatives, independent of the relationship stages, or states, in which customers engage with the firm. However, recent research using hidden Markov models suggests the importance of acknowledging customer relationship states as a means to understand customer behavior (Luo and Kumar 2013; Netzer, Lattin, and Srinivasan 2008). This concept is particularly important in business-to-business (B2B) settings, where customer relationships take longer to develop, last longer, exhibit higher switching costs, and have a greater impact on outcomes than do relationships for firms operating in business-to-consumer settings (Zhang, Netzer, and Ansari 2014). However, little research has moved beyond identifying the performance implications of different relationship states or sought to understand dynamic RM strategies that seek to influence migration across relationship states to enhance exchange performance. Specifically, we propose using dynamic relationship marketing to understand the impact of state-specific relationship migration strategies. Buyer seller relationships evolve through distinct stages (Celuch, Bantham, and Kasouf 2006; Dwyer, Schurr, and Oh 1987; Jap and Ganesan 2000), and latent customer relationships can be described according to discrete states at any given point in time (Netzer, Lattin, and Srinivasan 2008). Hidden Markov models are useful for studying relationships in such a dynamic 2

setting, because they can uncover latent relationship states at each point in time, as well as customer migration patterns across those states. Although such research has greatly advanced knowledge of customer relationships, few studies recover the hidden relationship states directly (Luo and Kumar 2013); those that do so tend to rely on transactional data to infer underlying relational constructs or mechanisms. With the current study, we instead capture overarching dynamic relationship states using theoretically sound, actual relational state variables (commitment, trust, norms, and dependence), which we link to actual, account-level firm performance variables (account profitability and sales growth). We combine longitudinal survey and transactional data from 346 B2B relationships maintained by a Fortune 500 firm across a broad set of product categories. Our research thus contributes to RM literature in several ways. We advance RM theory by developing and testing a framework that describes the underlying, state-specific migration mechanisms responsible for prototypical transitions. All customer state changes (with non-zero probabilities) can be explained parsimoniously by three positive (exploration, endowment, recovery) and two negative (neglect, betrayal) migration mechanisms, which account for the level and direction of relational development contained within a customer portfolio. Each prototypical migration mechanism in turn consists of different patterns of relational variables (trust, commitment, relational norms, and dependence), reflecting the development and decline of each customer relationship. For example, neglect captures a pattern of slow state variable decay, due to passive inattention (lack of communication). Betrayal describes immediate, dramatic damage to state variables, due to purposeful actions (conflict, unfairness), and it is typically is accompanied by a strong emotional response and outcomes that are more difficult to reverse than those caused by neglect. 3

We hypothesize that particular relationship state-specific migration strategies influence each migration mechanism. Not only do we show that the optimal strategy must match the relationship state, but we determine that removing marketing actions at the wrong time can quickly undo prior relational development. Thus, we move beyond extant theory that has focused on describing relationship states to improve our understanding of relationship state migration. Whereas extant RM research tests how specific strategies drive performance but ignores differences in relationship states, thereby generating results that suffer from aggregation biases across states, we provide a dynamic RM perspective that clarifies the performance impacts of state-specific relationship migration strategies. With a post hoc elasticity analysis, we also provide managerially relevant insights into the marginal effectiveness of relationship state-specific migration strategies. The blanket application of relationship migration strategies, without regard to a customer s relationship state, can be highly inefficient. For example, both marketing communication and the product mix lose approximately 27% of their effectiveness when applied to high relational development states rather than lower ones. Seller investment is 3.2 times more effective for higher relational development states than is communication. In damaged, poor quality relationship states, the best recovery option is through compromise, a finding that contrasts with extant literature (Samaha, Palmatier, and Dant 2011). Finally, in our sample, relationship migration strategies applied to the appropriate customer relationship states can improve firm revenues by hundreds of thousands of dollars. 4

Understanding Customer Relationship States We first review specific relationship state variables, defined as relational constructs that determine each developmental state, before we summarize extant literature on relationship state conceptualizations (states), or the blend of state variables that capture the nature of buyer seller relationships. From these two main concepts, we employ a hidden Markov model (HMM) to infer relationship states, customer migrations across states, and RM strategies that drive relationship state migrations. We label the strategies that focus on promoting or suppressing state migrations to enhance performance dynamic relationship marketing. Relationship State Variables Different frameworks propose various state variables as critical for determining a relationship s state or stage, such as trust, commitment, dependence, and relational norms. These four state variables are the most frequently studied and capture alternative facets of the relationship. That is, a relationship state typically depends on both partners perspectives on the relationship (partner and self), where trust is more partner-focused and commitment is more selffocused (Garbarino and Johnson 1999). State variables also provide information about individual partners (trust, commitment) or the bilateral structure of the exchange (dependence, relational norms). These four state variables have different temporal characteristics, such that trust and dependence tend to change more quickly, whereas commitment and norms change more slowly (Jap and Ganesan 2000). Each state variable also provides nonredundant information about the relationship. Perhaps the most studied relationship state variable is trust, defined as confidence in an exchange partner s reliability and integrity (Morgan and Hunt 1994, p. 23). It is an otherfocused evaluation of the partner s integrity, reliability, and intentions. Trust in an exchange 5

partner enhances performance by inducing potentially risky but rewarding investments, in the belief that the partner will not act opportunistically (Palmatier, Dant, and Grewal 2007). It also facilitates shared information and resources. Finally, trust helps a partner assess the quality of a relationship and is a key predictor of performance (Fang et al. 2008). Customer commitment is an enduring desire to maintain a valued relationship (Moorman, Zaltman, and Deshpandé 1992, p. 316). This state variable reflects self-focused attitudinal facets of an exchange, such as dedication, personal identification with the partner, and a focus on long-term benefits over short-term alternatives (Garbarino and Johnson 1999; Morgan and Hunt 1994). As a global evaluation of the relationship with a temporal facet, which indicates long-term expectations of continuity, customer commitment is key to the long-term success of a relationship (Palmatier et al. 2006). A third widely studied state variable is customer dependence, which reflects the evaluation of partner-provided benefits, for which there exist few alternatives (Hibbard et al. 2001). Customers and sellers can create value by investing in relationship-specific assets, which increase dependence and exposure to opportunism (Wathne and Heide 2000). These assets are difficult to transfer without loss of productive value but often enhance performance. In addition, a high level of dependence can lock in a partner and prevent switching, even in the face of problems. Customer dependence is useful for assessing a relationship, in that it captures an immediate evaluation of structural constraints with an exchange partner. Finally, relational norms emphasize long-term concerns about a partner s prosperity, ensure equitable sharing of benefits and costs, reduce opportunism, and provide a normative governance structure, because they guide and regulate the standards of trade and conduct (Gundlach, Achrol, and Mentzer 1995, p. 81; see also Heide and John 1990). Relational norms 6

focus on the interactions between partners and reflect the history of these interactions. Norms facilitate decisions, and they are useful for evaluating the state of a relationship. Along with trust, customer commitment, and customer dependence, relational norms provide a multifaceted view of the relationship, including partner versus self, individual versus bilateral structure, and shortversus long-term perspectives, which together define the relationship state. Relationship State Conceptualizations Relationship state conceptualizations (states) generally specify a set number of developmental stages to describe unique combinations of state variables, outcomes, and processes and thereby identify each state. Extant research exhibits little consensus about the appropriate number or nature of specific states though, positing anywhere from two to six different options, all based on varying state variables, as we summarize in Table 1. In general, lifecycle frameworks tend to consist of three fundamental stages or states: an initial neutral or transactional state, a positive relational state (often divided into multiple substages), and a negative relational state. - Insert Table 1 here - In most frameworks, relationships start with low experience and norms (Jap and Anderson 2007). These initial states are characterized by low to moderate interpersonal interactions and correspondingly low levels of relational state variables. Partners become aware of each other, often with low levels of relational bonding, akin to what Anderson and Narus (1991) call transactional exchanges. In some situations, this state can exhibit moderate financial performance, due to repetitive purchases with low transaction costs. Thereafter, lifecycle frameworks diverge considerably. Positive relationship states contain as few as one stage or as many as three or more (Heide 1994; Wilson 1995). Yet 7

consistently, a relationship state(s) appears to encourage the development of shared purposes, values, and expectations and identify value-creating opportunities. This state is characterized by increasing levels of trust and commitment; increasing interdependence as partners make nonrecoverable investments; and the establishment and augmentation of relational norms as the customer and seller explore, expand, and build the relationship (Dwyer, Schurr, and Oh 1997; Jap and Ganesan 2000). When relational variables and investments expand, the states feature higher levels of relational and financial performance, beyond the initial transactional state. Whether the boundaries between a developmental state and a mature state blend smoothly or differ fundamentally remains an open question. The final category is the negative relationship state, which exhibits low levels of trust, commitment, and norms, often due to some relationship failure, with varying levels of dependence. A relationship with low state variables may terminate in subsequent periods if nothing holds the partners together. However, when one partner is highly dependent on the other, for lack of better alternatives, the relationship might persist, despite its underlying problems. Depending on the circumstances, performance in a negative state exhibits diverging levels, often determined by the level of dependence. Highly dependent negative states may result in performance on par with neutral states, whereas less dependent negative states likely dissolve relatively quickly and degrade performance. Dynamic Relationship Marketing: Drivers of Relationship State Migrations Because the predominantly determinist perspective that appears in extant RM theory offers little insight into either the strategies that managers can use to influence customers migration from specific states or how different migration strategies ultimately affect 8

performance, we seek to advance RM theory in two key ways. First, we develop a theoretical framework that describes migration mechanisms (exploration, endowment, neglect, betrayal, recovery), to clarify both positive and negative state-specific change. Second, we empirically test the strategies that we expect will influence migration mechanisms and the transitions across relationship states. Then we can infer the performance implications of applying relationship state-specific migration strategies. Using an HMM approach to model relationship lifecycles, we simultaneously identify a customer s relationship state, its most probable migration paths, and the effect of different strategies on customer migration (see Figure 1). Because each customer s state is determined by the unique mixture of state variables, the state migrations ultimately are determined by changes in these state variables. A relationship migration mechanism is the unique pattern of change in relational variables that leads to a prototypical transition between relationship states. We identify three positive relationship migration mechanisms (exploration, endowment, and recovery) and two negative migration mechanisms (neglect and betrayal) from extant literature. In Table 2, we describe the unique pattern of changes to the variables, relevant migrations, and effective strategies for each mechanism. - Table 2 and Figure 1 about here - Relationship-Building Migration Mechanisms Early development relationship states: Exploration. Buyer seller interactions with low relational development, such as those described by Dwyer, Schurr, and Oh (1987) as transactional, have low to medium levels of state variables. For relationships from this state to grow, both parties must develop stronger relational bonds through increased levels of commitment and trust. Repeated interactions that build relational governance (e.g., norms), 9

through positive evaluations of reciprocal behaviors, also are crucial for continued future relationship development (Cannon, Achrol, and Gundlach 2000; Heide and John 1992). Performance likely improves in lockstep with stronger relational development during this state change, representing significant value creation, such as that which results from sales growth. We refer to this period as an exploration relationship migration mechanism, because the partners explore the value creation potential of the relationship and demonstrate their willingness to share rewards and costs as their relational bonds strengthen (Dwyer, Schurr, and Oh 1987; Jap and Ganesan 2000). Identifying value-creating opportunities and building norms to share value are both key to this exploration migration mechanism and the continuation of the relationship, beyond arm s-length, one-time transactions. Extant literature suggests two strategies might enhance the effectiveness of exploration: communication and product mix. Communication refers to the amount, frequency, and quality of information shared between partners (Mohr, Fisher, and Nevin 1996). Through the exchange of information, partners identify value-creating opportunities. In addition, communication helps partners establish mutual goals and relational norms, necessary for governing the exchange to ensure an equitable division of value, which then increases confidence about investing for further value creation (Jap and Ganesan 2000). Although communication helps identify opportunities, a partner also must provide sufficient offerings to fulfill its partner s needs. Product mix captures the seller s assortment and the variety of desirable products available to satisfy customer needs, such that it can act on identified opportunities to create value (Morgan and Hunt 1994). As effective strategies to promote the exploration migration mechanism, communication and product mix should increase the likelihood that a relationship migrates from a low development 10

relational state to a higher development relational state, by helping partners create and fairly share value, with the mutual goal of transitioning to a mature, sustained high performance state. H 1 : (a) Communication and (b) product mix positively affect the probability of migrating from a transactional to a transitional state. Advanced development relationship states: Endowment. Transactional relationships still must navigate potential hurdles to further their relational development beyond arm s-length exchange. Although state variables may exhibit medium to high levels, migrations from lower to higher development states require the continued development of trust, customer commitment, and relational norms to persist in strengthening relational bonds. In contrast with exploration, relational norms are already present in sufficient levels in this case to ensure buyer seller governance. The crucial state variable needed to cement the relationships is dependence, which can lock in both parties to prevent switching. It arises when the seller can offer its customer benefits that would be difficult to procure from other sources (Hibbard et al. 2001). This migration mechanism, termed endowment, implies stronger relational bonding between partners, with a great increase in dependence as both partners make unrecoverable investments in the exchange to spur performance to its highest level. Past research indicates the key role of relationship investments, or time, effort, spending, and resources focused on building a stronger relationship (Palmatier et al. 2006, p. 138), in leveraging the value-creating capabilities of partners (De Wulf, Odekerken-Schröder, and Iacobucci 2001). Relationship investments increase dependence, because of the difficulty of redeploying resources to other exchanges, so they can help mitigate concerns of opportunism and support further investments (Heide and John 1990). Customer and seller investments help raise performance to its highest level by endowing the relationship with the resources needed to exploit any identified opportunities (Jap and Ganesan 2000; Palmatier et al. 2006). Moreover, the 11

positive interaction between customer and seller investments should accelerate the relationship transition to a communal state. When both parties provide resources, the opportunity to create complementary and unique value increases, with the potential to generate superior returns (Kozlenkova, Samaha, and Palmatier 2013). Thus, customer and seller investments are key strategies for enhancing the endowment mechanism by providing the resources needed to exploit opportunities, creating unique value propositions, and building a protective shield of interdependence. Both parties then engage in a mutually dependent, mutually satisfactory relationship that provides them with sustainably higher performances. Such a relationship is communal in nature. H 2 : (a) Customer and (b) seller investments positively affect the probability of migrating from a transitional to a communal state. H 3 : The interaction of customer and seller investments positively affects the probability of migrating from a transitional to a communal state. Relationship-Damaging Migration Mechanisms Passive damage: Neglect. Relationships with higher levels of development also can weaken, rather than continuing to strengthen or remaining in the same state. A negative but passive migration mechanism that describes state change is neglect, which captures a pattern of decay due to passive inattention rather than proactive negative activities. The change results from erosion in the state variables due to a failure to maintain the relationship. Often the relationship does not end, because neglect results in reduced social exchanges with the partner firm (but not necessarily reduced economic exchanges with them) (Ping 1999, p. 221). Marketing research on relationship decay and passive neglect is relatively limited, but social psychology literature identifies an absence of communication between partners as especially detrimental to relationships (Rusbult, Zembrodt, and Gunn 1982). Lower levels of 12

communication reduce trust, commitment, and norms but typically do not affect dependence, which is more of a structural constraint (Heide and John 1990; Koza and Dant 2007). The absence of communication thus results in the neglect mechanism, which increases the likelihood of a relationship returning to the transactional state. Similarly, reducing relationship investments removes the resources required for value creation and signals a lack of relational motivation (Rindfleisch and Heide 1997), which suppresses trust, commitment, and relational norms, consistent with the neglect mechanism (Palmatier et al. 2008). For example, Rusbult, Zembrodt, and Gunn (1982, p. 1239) argue that failing to invest resources is promotive of relatively destructive behaviors such as ignoring the partner. Thus, reduced communication or reduced customer and seller investments represent passive strategies for triggering the neglect migration mechanism, which increases the likelihood that a relationship will shift from a more developed relational state to a lower state, in that it removes the factors that helped improve the relationship. Because the passive decay of a relationship due to inattention or inaction is less emotional or purposeful, the relationship likely drifts into a neutral or transactional state, rather than a more negative resentful or damaged state. H 4 : Reducing (a) communication, (b) customer investments, and (c) seller investments increase the probability of migrating from a more developed relationship state to a neutral relationship state. Active damage: Betrayal. In addition to passive neglect, purposeful actions (or inadvertent actions perceived as purposeful by the customer) can significantly and immediately reduce relational state variables (Hibbard et al. 2001). When sellers actively undermine the customer relationship, it implies the negative migration mechanism betrayal, which involves an immediate, dramatic drop in the state variables, due to purposeful actions rather than passive neglect, typically accompanied by strong emotional responses (Hibbardet al. 2001; Samaha, 13

Palmatier, and Dant 2011). Relationships with very low levels of trust, commitment, and norms may persist though, because the continued exchange is held together by the medium to high levels of dependence. Often, only dependent relationships survive betrayal, because they lack of alternatives, despite the likely presence of anger and a desire to punish the betrayer (Eyuboglu and Buja 2007). This strong emotional aspect and resultant drop in state variables distinguishes the consequences of betrayal from those of neglect. According to extant literature, conflict and injustice represent two migration strategies that likely invoke a betrayal migration mechanism (Kaufmann and Stern 1988). Unresolved conflict, or disagreement between a seller and a customer as each party strives to achieve its business goals (Samaha, Palmatier, and Dant 2011 p. 102), can have detrimental impacts on relationship health, in that the mere presence of relationship conflict demonstrates that parties do not share mutual understanding and appreciation, and will thus undermine trust (Langfred 2007, p. 887). Injustice, or perceived unfairness in the ratio of outcomes to inputs, can be especially damaging in a relationship setting (Samaha, Palmatier, and Dant 2011). We further expect a synergistic interaction between conflict and injustice that accelerates the transition of a relationship to the damaged state, because perceptions of injustice suggest a partner s negative intentions (Campbell 1999). That is, perceived injustice adds emotionally powerful, negative attributions to any ongoing conflict. Because injustice is difficult to attribute to external sources other than the exchange partner, these perceptions exacerbate the damaging effects of conflict (Samaha, Palmatier, and Dant 2011). Thus, both conflict and injustice represent key triggers of the betrayal mechanism, which increases the likelihood that a relationship will shift from a welldeveloped relational state to a negative or damaged state, because it removes factors that underpin the exchange relationship and provoke significant negative emotional responses. 14

H 5 : (a) Conflict and (b) injustice increase the probability of migrating from a more developed relationship state to a negative relationship state. H 6 : The interaction of conflict and injustice enhances the probability of migrating from a more developed relationship state to a negative relationship state. Damaged relationship states: Recovery. The last migration mechanism is recovery, which describes the pattern of change in the state variables by which an exchange migrates from a negative to a neutral state. It necessitates a substantial improvement in commitment, trust, and norms; the only state variable holding the relationship from complete dissolution is likely the high level of dependence. Migration strategies to promote the recovery mechanism must help resolve the conflict and injustice that drove the initial migration to the negative state, as well as rebuild the damaged relational state variables. Just as communication helps identify value creation opportunities to promote relationship growth, it is critical to relationship recovery, because it can identify the root causes of conflict and inequity. Communication is necessary but not sufficient for recovery; after the problem is identified, partners must act on this information through compromise, or the resolution of conflicts by developing a middle ground on a set of issues based on the initial positions of both parties (Ganesan 1993, p. 186). Through compromise, partners directly address the conflict and inequalities that damaged the relationship. Thus, communication and compromise individually and synergistically represent migration strategies for promulgating the recovery migration mechanism, which increases the likelihood that a relationship will shift from a damaged to a transactional state, by identifying and correcting problems and rebuilding relational bonds. H 7 : (a) Communication and (b) compromise increase the probability of migrating from a negative state to a neutral state. H 8 : The interaction of communication and compromise enhances the probability of migrating from a negative state to a neutral state. 15

Methodology Sample A large Fortune 500 firm agreed to participate in this research and allowed access to its business customers (channel members) over a six-year period. The focal firm sells hundreds of products across several dozen categories (e.g., housewares, clothing, books, tools, automotive, consumer electronics, sports equipment, home appliances) and offers both branded and privatelabel items. Its approximately 1,600 suppliers conduct average yearly business with the firm of more than $600,000, and our study period revealed only 2% turnover among these suppliers, reflective of the relative longevity, stability, and short-term switching costs inherent to such B2B relationships (Jap and Ganesan 2000). This broad sample thus should generalize effectively to B2B channel relationships, more so than a sample based in narrower product categories. The data were collected through six consecutive annual surveys, then matched with the focal firm s sales records for each channel member. The constructs included in each survey remained the same, measured with identical items each year. In any particular year, an average of 57% of the channel members responded to the survey with usable responses. We thus defined our final sample as the 346 channel members that responded with usable questionnaires in at least five of the six years, so that we could effectively model potential relationship dynamics. We tested for nonresponse bias by comparing early and late responses for all waves of the study; these results suggested no differences (p >.05). In addition, multivariate analysis of variance tests of the study constructs and sales numbers indicated that responses by channel members that returned at least five of the six waves did not differ significantly from those of channel members that returned four or fewer waves (p >.05). Comparable tests of demographic 16

variables (age of relationship at time t) in our final sample against those of firms with four or fewer responses and nonresponders also suggested no significant differences across groups. The survey questions, as reported in the Appendix, remained the same every year. Trust, customer commitment, customer dependence, relational norms, and the profit assessments used five-point Likert scales. The seller provided objective financial measures of sales revenue and growth. All scales exhibited acceptable Cronbach s alpha values above.81, in support of their internal reliability. Individual confirmatory factor analyses for each year yielded acceptable fit indices (χ 2 (220) = 279.22 1009.40, p <.01; comparative fit index =.95.99; root mean square error of approximation =.03.05). Table 3 summarizes the descriptive statistics and correlations. Although these data are censored, in that the channel relationships existed both prior to and after the study period, we do not perceive a problem, for two reasons. First, we used an HMM framework to examine movement between states of any given relationship, rather than tracking the entirety of the relationship lifecycle. Thus, we model a sample of relationship state changes that tend to be generally infrequent in B2B relationships. Second, HMMs can use interval censored data without issue (e.g., Luo and Kumar 2013), because the Markov chain depends only on the previous state, not the entire history of the relationship. - Insert Table 3 here - Modeling Approach Using a multivariate HMM, we empirically infer latent relationship states from the timevarying levels of survey responses of the four state variables for each customer. The vector of state variables for customer i at time j is y ( t, c, d, n ), where t is the average of the trust items, and c, d, and n are commitment, dependence, and relational norms, respectively. The ij ij latent state at time j for customer i, ij ij ij ij ij ij Y y,..., Y = y i1 i1 ij ij ij, has four components: (1) the initial 17

latent state probabilities π i, which denote the customer ss initial state; (2) a matrix of transition probabilities among the states Ω i, j-1 j that explains how the customer moves between statess from one period to the next, as well as the effects of various migration strategiess on the transition; (3) a multivariate likelihood of interrelated state variables, conditional on the relationship state L ij s fis tij cij (,, d ij, n ); and (4) the customer s latent state probability at each ij time period. Initial state distribution. Let s denotee a latent relationship state ( s = 1,2,....,S ). Let be the probability that customer i is initially in state s in the first period of our data set, where is 1 is S 0 and s 1 is 1. We use S 1 logit-transformed parameters to represent the vector of initial state distribution. Markov chain transition matrix. The HMM transition matrix denotes the probability of a customer moving from one state to each other state, from one period to the next. We model the transitions as a Markov process: where PS ( s S s) ijss ij ij-1 is the conditional probability that customer i moves from state s at time j 1 to state s at time j, and where 0 itss' 1 ss, ', and s ijss ' 1. ' 18

The transition probabilities might be influenced by several factors (i.e., migration strategies) at time j 1. We therefore define each transition probability ijss as a function of migration strategies, and we use the logit specification to ensure 0 1. We define: itss ' x 'γ ij-1 is e 1 e ijss ' S 1 s 1 x ij-1'γ is, (1) where x is a vector of migration strategies affecting the transition between states, and γ is a ij-1 state-specific vector of response parameters that measure the impact of each migration strategy on the transition probability. Specifically, x includes all hypothesized migration ijss ij-1 strategies and interactions. HMM likelihood function. Conditional on being in state s at time j, a customer responds to the level of trust, commitment, dependence, and relational norms. These four responses are is unconditionally interrelated. If customer i at time j is in a latent state Sij s, we can factor the conditional discrete continuous joint likelihood L ij s using multivariate normal distributions to model the joint distributions on all four variables, as follows: L ij s =f is (t ij,c ij,d ij,n ij ), (2) where t c, d and n are trust, commitment, dependence, and relational norms, respectively. ij ij ij ij Considering the Markovian structure of the model, the likelihood of observing a set of joint customer responses at time J depends on all responses up to that event. The likelihood of observing the customer s responses over J periods ( Y, Y,..., Y ) is: i1 i2 ij L P( Y =y,..., Y =y ) π M Ω M...Ω M1, (3) ij i1 i1 ij ij i i1 i, 1 2 i2 i, J-1 T ij 19

where π is the initial state distribution, Ω is the transition matrix, M is a S S diagonal i matrix with the elements L from Equation 1 on the diagonal, and 1 is a S 1 vector of ones. To ensure identification of the states, we restrict customers responses to trust to be non-decreasing in the relationship states. Let ij s ij, j+1 b 01 i be customer i s mean-level response to trust in ij state 1, with the lowest trust. Then we impose the restriction... t01i t02i t0si by setting s exp( ) t0si t01i ; s = 2,, S. To avoid underflow, we scale the likelihood function in s 2 t0s i Equation 2 using the approach suggested by MacDonald and Zucchini (1997). Recovering the state membership distribution. We use a filtering approach (Hamilton 1989), to determine the probability that customer i is in state s at time j, conditional on the customer s history: P( S s Y, Y,.., Y ) π M Ω M...Ω L / L ij i1 i2 ij i i1 i, 1 2 i2 i, j-1 t s ij s ij, (4) where i,j-1 j.s is the s th column of the transition matrix i,j-1 j, and L ij is the likelihood of the observed sequence of joint decisions up to time j from Equation 2. Estimation Approach In our HMM, relationship states are determined by each customer s time-varying levels of survey responses pertaining to the four state variables (commitment, trust, norms, and dependence). It thus simultaneously captures customer seller relationship dynamics, as well as allowing customers to transition freely across different states. We estimate multiple models with different numbers of latent states and select the one that offers the best fit, according to the deviance information criterion (DIC), which accounts for additional model parameters. If the relationship states are truly dynamic, we expect a better fit from HMM than from static latent class segmentations. Thus, we compare the fit of our HMM against three-, four-, and five-state 20

latent class segmentation models, estimated on our longitudinal data set of 346 B2B relationships maintained by a Fortune 500 firm, which includes the corresponding customer financial performance. Results Number of States Compared with the latent class segmentations, all three HMMs have significantly smaller DICs, indicating the presence of nonrandom patterns of customer state-to-state migrations. The DICs for the three-, four-, and five-state latent class models are 18,992, 16,972, and 17,981, respectively significantly larger than the comparable values for the HMM DICs, at 14,023, 13,200, and 14,386, respectively. These fit criteria indicate that the four-state HMM fits the data significantly better than any other HMM or latent class models. We report the results in Table 4, which reveals the significantly different mixtures of state variables and outcomes for the four relationship states. In addition, Table 4 presents the migration path probability matrix. The diagonal of the matrix represents the mean probability of remaining in the same state (i.e., stickiness). The off-diagonal values indicate the probabilities that a customer in a given state, listed on the left, will migrate to a different state, listed at the top of each column. - Insert Table 4 here - Identifying Relationship States The first transactional state is where most relationships begin (M age = 4.0 years), characterized by low-medium levels of trust (M = 3.74), customer commitment (M = 3.85), and customer dependence (M = 2.58), as well as relatively low levels of relational norms (M = 2.46). This mixture of state variables captures a neutral, undeveloped relationship, with little relational 21

governance embedded in the exchange. Relationships in this transactional state exhibit moderate levels of profit (M = 3.65) and sales growth (7%), but it captures the most relationships across the four states (55%). It is unique, in that it exhibits the most heterogeneous state migrations. Relationships move to stronger relational states 35% of the time, or to weaker or damaged relational states 13% of the time, but they remain in the transactional state from one period to the next 51% of the time. The transactional state is evaluative, such that parties receive some value from their transactional exchanges but also explore opportunities to create more value before making significant investments. Following a positive migration trajectory, relationships move from the transactional to the next state, which exhibits medium to high levels of trust (M = 4.02), commitment (M = 4.24), and relational norms (M = 3.48) and medium levels of customer dependence (M = 2.83). Two unique characteristics help define this state. First, all state variables are higher than in the previous state, and the change in relational norms is three to five times greater than the change in the other state variables, which reflects a substantial increase in relational governance. Second, this state is the least sticky; in each period, 70% of the relationships transition to another state (61% strengthening, 9% weakening). We thus refer to it as the transitional state, with the recognition that most of these relationships are simply shifting from a transactional to the most developed relational state and are unlikely to remain in this state for more than one period. Profits (M = 3.76), and annual sales growth (22%) are higher than in the transactional state. The communal state exhibits the highest levels of trust (M = 4.75), commitment (M = 4.80), customer dependence (M = 3.75), and relational norms (M = 4.15). Strong relational development is reflected in the highest level of profit (M = 4.30), as well as good sales growth (10%). In terms of migration, the communal is the most sticky relationship state (66% remain 22

in the state each period), but if the relationship changes, it also is more likely that they move directly to the weakest relational state (19%), rather than just migrating to a neutral relational state, such as the transactional one (13%). Thus, transgressions appear particularly damaging in the communal state. Finally, the last relationship state is the damaged state, marked by low levels of trust (M = 2.58) and commitment (M = 2.73) and very low levels of relational norms (M = 1.60), though with medium to high levels of customer dependence (M = 3.29). The combination of low trust, commitment, and norms but higher dependence results in divergent relational and financial performance. Sales growth (2%) is the lowest, though even the lowest level of profit (M = 3.11) is still positive, likely due to customer dependence. Exiting the damaged state is difficult; 59% of the relationships remain stuck here in each period, and if they recover, they move only to the relationally neutral transactional state. Exchanges in the damaged state do not feature undeveloped relationships but rather negative ones. If not for the high dependence, many relationships would dissolve. Thus, our analysis identifies four unique relationship states connected by five prototypical migrations. Effectiveness of Relationship Migration Strategies: Dynamic Relationship Migration In Table 5, we present the effect of each hypothesized migration strategy on the probability of a state-specific change. Conditional on being in the transactional state, both communication (β =.47, p <.01) and product mix (β =.59, p <.01) significantly increase the probability of moving a customer to the transitional state, in support of H 1 and the underlying exploration migration mechanism. In support of the endowment mechanism, partners can increase the probability of advancing their relationship from the transitional to the communal state by making customer (β =.48, p <.01) and seller (β =.65, p <.01) investments, which 23

supports H 2. Furthermore, the positive interaction between customer and seller investments promotes the transition of partners to the communal state (β =.37, p <.01), in support of H 3. For the two relationship damaging mechanisms, we find that reducing communication (β =.55, p <.01), customer investment (β =.51, p <.01), and seller investment (β =.70, p <.01) increases the probability of shifting a relationship from the transitional to the transactional state, in support of H 4. Similarly, reducing the levels of communication (β =.24, p <.01), customer investment (β =.50, p <.01), and seller investment (β =.37, p <.01) increases the probability of shifting a relationship from the communal to the transactional state, providing further support for H 4. Worse still, the betrayal mechanism drives a relationship to the damaged state through conflict and injustice. As we predicted in H 6, conflict (β =.56, p <.01) and injustice (β =.82, p <.01) significantly degrade communal relationships to the damaged state; being perceived as unjust appears especially harmful. Similarly and in support of H 7, conflict (β =.39, p <.01) and injustice (β =.50, p <.01) increase the likelihood that a transitional relationship moves to the damaged state. The significant, positive interaction effect between conflict and injustice enhances the probabilities of migrating from the communal (β =.61, p <.01) and transactional (β =.40, p <.01) states to the damaged state, in support of H 8. Finally, we evaluate the recovery migration mechanism, which moves customers from damaged to transactional states. Contrary to our expectations, the direct effect of communication fails to increase the probability of transitioning out of the damaged state, so we must reject H 9a. Yet compromise (β =.64, p <.01) has a significant positive effect on the likelihood of migrating to the transactional from the damaged state, in support of H 9b. Moreover, we find that the interaction between communication and compromise is significant (β =.44, p <.01), in support 24

of H 10. That is, communicating is helpful to recovery only if combined with the more solutionoriented compromise strategy (i.e., talking without action does not help recover a relationship). - Table 5 about here - Relationship Migration Strategies Elasticities To understand the performance consequences of applying state-specific migration strategies, we ran a series of hypothetical scenarios with one standard deviation changes to factors that are within the control of the selling firm, such as communication or seller investment, while maintaining the status quo for all other factors. By examining the resulting changes in the mean transition matrices, we could calculate the elasticity of a migration strategy and thereby derive managerial insights into each dynamic relationship marketing strategy. In turn, we can understand both the relative effectiveness of a strategy across states, as well as the performance implications of applying such a strategy. Table 6 presents the results of our elasticity analysis for each relationship migration mechanism. For example, for exploration, we examine the effect of a one standard deviation increase in communication and product mix. We present the results as elasticities, such that for a 1% shift in the migration strategy of interest, we identify the corresponding percentage change in the probability of moving from one relationship state to another. - Insert Table 6 here - Exploration. With Table 6, we reveal that the most effective strategy for migrating a customer in the low development transactional state to the higher development transitional state is product mix (1.64), followed by communication (1.30). For comparison purposes, we also note that these two strategies are more effective for this migration mechanism than three other migration strategies: seller investment (.88), customer investment (1.21), or compromise (.15). 25

Likewise, both strategies have a greater marginal return in the exploration mechanism (i.e., relationships in the transitional state) than in the endowment mechanism (product mix =.16, communication =.35) or recovery mechanism (product mix =.19, communication =.16), that is, in the two other relationship-building transitions. Increasing communication or the product mix by 10% would yield $502,882 and $633,197 increases in revenue in the next period, respectively roughly equivalent to adding a new account with average value. Although each strategy induces a slightly greater increase in the probability of moving from the transitional to the communal state, they are less than 27% as effective when they apply to other states, which underscores the benefit of deploying communications and product mix strategies in low development relationships. Endowment. For relationships in the transitional state, the most effective migration strategy to move customers to the communal state is seller investment (1.13), which is substantially more effective than communication (.35) or the product mix (.16) for relationships with higher levels of development. Presumably, the information and value created in earlier states already are known to both parties. Applying seller investments also provides the most marginal benefit for transitional relationships compared with transactional (exploration.88), or damaged (recovery.15) relationships. A 10% increase in seller investment would yield a $436,742 increase in revenue in the next period. If the customer also is willing to invest in the relationship (.80), this strategy, though outside the selling manager s control, benefits both parties, as does a willingness to compromise (.82). However, seller investment is the key strategy for more developed, yet still nascent relationships, because it provides resources for a more successful exchange. 26

Neglect. Rather than increase the levels, we reduce the levels of a particular migration strategy, one at a time, to understand the effect of neglect on more developed relationships. Thus we find particularly high elasticities for any strategy, indicating the strongly detrimental effect of starving a customer relationship. For example, for relationships in the less stable transitional phase, reductions in seller investment (-4.95) and communication (-2.61) are substantially more likely to drive a relationship back to the transitional state than are similar reductions in seller investment (-.73) and communication (-.45) for relationships in the most developed, communal state. With 10% decreases in communication or seller investment, relationships in the communal state would yield $94,176 and $152,912 decreases in revenue in the next period, respectively. The greater the number of relationships in the firm s portfolio that have developed past, arm slength exchanges, the more essential ongoing maintenance appears to be for performance. Betrayal. To examine the other relationship-damaging mechanism, we increased the levels of conflict and injustice to determine their effect on driving more developed relationships in transitional and communal states toward the damaged relationship state. Again, relationships in the transitional state are less stable and more prone to damage than relationships in the communal state. For example, a transitional relationship exhibits more sensitivity to increases in conflict (.96 vs..71) and injustice (1.96 vs. 1.83) than a communal relationship. In this sense, injustice is 104% to 157% more caustic to the customer relationship than conflict. Accordingly, 10% increases in conflict and injustice would yield $199,207 and $511,560 decreases in revenue in the next period, respectively. Recovery. In the event that a customer relationship deteriorates to a damaged state, it can be difficult, though not impossible, to recover. A manager that attempts to salvage a poor relationship with greater communication (.16), product mixes (.19), and seller investments (.15) 27

is making inefficient use of the firm s resources. Instead, compromise (1.45) offers far more marginal return than the other strategies. However, in our sample, the average account size for a transactional relationship is smaller than the average account size for a damaged relationship, and accordingly, a 10% increase in compromise applied to damaged relationships would actually yield a $232,030 decrease in revenue in the next period. Rather than transactional (.15) or transitional (.82) relationships, compromise is more effective when applied to damaged relationships for moving a customer to a more developed state. Although clearly outside of the firm s direct control, a willingness on the part of the customer to mend the relationship through investment (.80) is beneficial, though still secondary to a compromise strategy. Thus, applying the strategies that fit relationships that are more developed, such as communication, seller investment, or effective product mix, is less appropriate for damaged or negative relationships. Discussion Using a hidden Markov model with six years of survey and transactional data collected from 346 relationships maintained by a Fortune 500, we capture dynamic relationship state changes using actual relational state variables (commitment, trust, norms, and dependence) and associate them with real, account-level firm performance variables (profitability and sales growth). This framework allows us to build on extant theory and managerial practice. Theoretical Implications This study advances RM theory in two key ways. First, we identify and define three positive (exploration, endowment, recovery) and two negative (neglect, betrayal) prototypical relationship migration mechanisms that parsimoniously account for observed customer state changes. Each positive migration mechanism consists of building relational variables at different 28

times and in different patterns, and it reflects the potential development and decline of a customer relationship. Relationships with lower levels of development in transactional states demand increases to relational variables, such as trust and commitment. More developed relationships instead must build mutual dependence and relational governance through norms. The two relationship-damaging mechanisms also add to understanding of relationship decline. As Jap and Anderson (2007, p. 272) note, decline is a separate phenomenon, unique in its own right, and deserves more systematic research. We show that neglect and betrayal both damage relationships but operate through different mechanisms. Neglect is passive inattention, such that partners withhold relationship-sustaining resources and move the relationship to a transactional state. Betrayal actively undermines relational equity, driving the relationship into a damaged state, such that it likely persists only due to customer dependence. Second, we posit and empirically test relationship state-specific migration strategies that we expect to influence each migration mechanism, then determine which do so most efficiently. For example, increased communication and product mixes are 73% and 91% less effective, respectively, when applied to higher development relationships rather than less developed, transitional relationships. The correct strategy thus must match the correct relationships state. In addition, the removal of marketing actions at the wrong time can quickly undo prior relational development. A reduction in seller investment in the transitional state, for example, is 4.3 times as likely to result in a negative migration to the transactional state than an identical increase is to produce a positive migration to the communal state. Thus, we move beyond extant theory that describes relationship states and provide further understanding of relationship state migration. The results show which dynamic relationship marketing strategies individually and synergistically are most effective for promoting or suppressing state migration. 29

Managerial Implications Whenever possible, managers should consider multiple facets of a customer relationship, because what may appear to be one type of relationship actually may be another, and the identification can have important implications for managerial action. For example, we find that different patterns of all four state variables are critical for determining unique state conceptualizations. Although trust and commitment tend to move in unison, divergent levels of relational norms and dependence are important for state identification. Frameworks that fail to include one of these critical state variables thus may be unable to identify or distinguish a transactional state from a damaged state across a portfolio of customers (e.g., Palmatier et al. 2013). Relying on purely transactional data to infer relationship states, though convenient, is risky. An example underscores this point: Damaged and transitional relationships appear to have similar account sizes ($687,000 and $670,000) in the short run, yet each demands substantially different treatment in terms of migration strategies. Therefore, managers must realize that account performance and relationship state do not always align intuitively, and developing marketing actions without looking deeply into the underlying relationship can lead to inefficient resource allocations. Our findings also suggest that, even though a relationship in a damaged state is roughly 50% more likely to remain damaged than to improve, it still can be recovered through compromise a result that contrasts with extant literature (e.g., Dwyer, Schurr, and Oh 1987; Luo and Kumar 2013). We make a case for investigating multiple facets of a customer relationships, then applying appropriate relationship state-specific strategies that can provide the best marginal returns. Limitations and Further Research 30

We also acknowledge some limitations of our research. First, we examine longitudinal relationships between a single large firm and its many customers. The generalizability of this analysis would increase with a broader sample of customer seller relationships from multiple firms across different industries. Further research should determine if there are fundamental differences in relationship lifecycles across industries or contexts, to clarify the most effective strategies for managing relationships in various environments. Second, we examine four important relationship state variables, but other state variables might capture additional relationship facets. For example, gratitude and reciprocity debts could be critical early in a relationship lifecycle. Third, we used a yearly sampling frame, a gap that might be too long to detect rapid relationship migrations. Although business relationships tend to be slow to develop (Jap and Anderson 2007), monthly or quarterly measurements might provide more fine-grained information about relationship dynamics, especially early in the relationship and during conflictladen periods. Fourth, we do not account for competitive relationships. Further research should expand our analyses by noting the possible influence of third parties and examining group-level relationship dynamics. 31

References Anderson, James C. and James A. Narus (1991), "Partnering as a Focused Market Strategy," California Management Review, 33 (Spring), 95-113. Campbell, Margaret C. (1999), "Perceptions of Price Unfairness: Antecedents and Consequences," Journal of Marketing Research, 36 (May), 187-99. Cannon, Joseph P., Ravi S. Achrol, and Gregory P. Gundlach (2000), When Does Vertical Coordination Improve Industrial Purchasing Relationships? Journal of Marketing, 64 (October), 52-65. Celuch, Kevin G., John H. Bantham, and Chickery J. Kasouf (2006), "An Extension of the Marriage Metaphor in Buyer Seller Relationships: An Exploration of Individual Level Process Dynamics," Journal of Business Research, 59 (May), 573-581. Crosby, Lawrence A., Kenneth R. Evans and Deborah Cowles (1990), Relationship Quality in Services Selling: An Interpersonal Influence Perspective, Journal of Marketing, 54 (July), 68-81. Dant, Rajiv P., and Patrick L. Schul (1992), "Conflict Resolution Processes in Contractual Channels of Distribution," Journal of Marketing, 56 (January), 38-54. De Wulf, Kristof, Gaby Odekerken-Schröder, and Dawn Iacobucci (2001), "Investments in Consumer Relationships: A Cross-Country and Cross-Industry Exploration," Journal of Marketing, 65 (October), 33-50. Dwyer, F. Robert, Paul H. Schurr, and Sejo Oh (1987), "Developing Buyer-Seller Relationships," Journal of Marketing, 51 (April), 11-27. Eyuboglu, Nermin and Andreas Buja (2007), "Quasi-Darwinian Selection in Marketing Relationships," Journal of Marketing, 71 (October), 48-62. Fang, Eric, Robert W. Palmatier, Lisa K. Scheer, and Ning Li (2008), "Trust at Different Organizational Levels," Journal of Marketing, 72 (March), 80-98. Ganesan, Shankar (1993), "Negotiation Strategies and the Nature of Channel Relationships," Journal of Marketing Research, 30 (May), 183-203. Garbarino, Ellen and Mark S. Johnson (1999), "The Different Roles of Satisfaction, Trust, and Commitment in Customer Relationships," Journal of Marketing, 63 (April), 70-87. Gartner Research (2012), Market Snapshot: Customer Relationship Management, Worldwide, ID:G00233998, (Accessed July 24, 2013). 32

Greenbaum, Howard H., Ellsworth J. Holden, and Lucian Spataro (1983), "Organizational Structure and Communication Processes: A Study of Change," Group and Organization Management, 8 (January), 61-82. Grégoire, Yany and Robert J. Fisher (2008), "Customer Betrayal and Retaliation: When Your Best Customers Become Your Worst Enemies," Journal of the Academy of Marketing Science, 36 (Summer), 247-61. Gundlach, Gregory T., Ravi S. Achrol, and John T. Mentzer (1995), "The Structure of Commitment in Exchange," Journal of Marketing, 59 (January), 78-92. Hamilton, James D. (1989), A New Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle, Econometrica: Journal of the Econometric Society, 57 (March), 357-384. Heide, Jan B. (1994), "Interorganizational Governance in Marketing Channels," Journal of Marketing, 58 (January), 71-85. and George John (1990), "Alliances in Industrial Purchasing: The Determinants of Joint Action in Buyer-Supplier Relationships," Journal of Marketing Research, 27 (February), 24-36. and (1988), The Role of Dependence Balancing in Safeguarding Transaction- Specific Assets in Conventional Channels, Journal of Marketing, 52 (January), 20 35. and (1992), "Do Norms Matter in Marketing Relationships?" Journal of Marketing, 56 (April), 32-44. Hibbard, Jonathan D., Frederic F. Brunei, Rajiv P. Dant, and Dawn Iacobucci (2001), "Does Relationship Marketing Age Well?" Business Strategy Review, 12 (4), 29-35. Jap, Sandy D. and Erin Anderson (2007), "Testing a Life-Cycle Theory of Cooperative Interorganizational Relationships: Movement Across Stages and Performance," Management Science, 53 (February), 260-75. and Shankar Ganesan (2000), "Control Mechanisms and the Relationship Life Cycle: Implications for Safeguarding Specific Investments and Developing Commitment," Journal of Marketing Research, 37 (May), 227-45. Johnson, Michael D. and Fred Selnes (2004), "Customer Portfolio Management: Toward a Dynamic Theory of Exchange Relationships," Journal of Marketing, 68 (April), 1-17. Joshi, Ashwin W. (2009), "Continuous Supplier Performance Improvement: Effects of Collaborative Communication and Control," Journal of Marketing, 73 (January), 133-150. 33

Kahn, Barbara (1998), "Variety: From the Consumer s Perspective," Product Variety Management, 10 (January), 19-37. Kaufmann, Patrick J., and Rajiv P. Dant (1992), "The Dimensions of Commercial Exchange," Marketing Letters, 3 (April), 171-185., and Louis W. Stern (1988), "Relational Exchange Norms, Perceptions of Unfairness, and Retained Hostility in Commercial Litigation," Journal of Conflict Resolution, 32 (September), 534-52. Koza, Karen L. and Rajiv P. Dant (2007), "Effects of Relationship Climate, Control Mechanism, and Communications on Conflict Resolution Behavior and Performance Outcomes," Journal of Retailing, 83 (August), 279-96. Kozlenkova, Irina V., Stephen A. Samaha, and Robert W. Palmatier (2013), "Resource-Based Theory in Marketing," Journal of the Academy of Marketing Science, 6 (April), 1-21. Kumar, Nirmalya, Lisa K. Scheer, and Jan-Benedict E.M. Steenkamp (1995), The Effects of Perceived Interdependence on Dealer Attitudes, Journal of Marketing Research, 32 (August), 348-356. Langfred, Claus W. (2007), "The Downside of Self-Management: A Longitudinal Study of the Effects of Conflict on Trust, Autonomy, and Task Interdependence in Self-Managing Teams," Academy of Management Journal, 50 (August), 885-900. Lewicki, Roy J., and Barbara B. Bunker (1996)). "Developing and Maintaining Trust in Work Relationships." in Trust in Organizations: Frontiers of Theory and Reach. Thousand Oaks, CA: Sage Publications, 114-39. Luo, Anita and V. Kumar (2013), "Recovering Hidden Buyer-Seller Relationship States to Measure the Return on Marketing Investment in Business-to-Business Markets," Journal of Marketing Research, 50 (February), 143-60. Lusch, Robert F. and James R. Brown (1996), Interdependency, Contracting, and Relational Behavior in Marketing Channels, Journal of Marketing, 60 (October), 19 38. MacDonald, Iain L. and Walter Zucchini (1997), Hidden Markov and Other Models for Discrete- Valued Time Series. New York: CRC Press. Mohr, Jakki J., Robert J. Fisher, and John R. Nevin (1996), "Collaborative Communication in Interfirm Relationships: Moderating Effects of Integration and Control," Journal of Marketing, 60 (July), 103-15. Moorman, Christine, Gerald Zaltman, and Rohit Deshpandé (1992), Relationships Between Providers and Users of Market Research: The Dynamics of Trust Within and Between Organizations, Journal of Marketing Research, 29 (August), 314-29. 34

Morgan, Robert M. and Shelby D. Hunt (1994), "The Commitment-Trust Theory of Relationship Marketing," Journal of Marketing, 58 (July), 20-38. Mullins, Ryan R., Michael Ahearne, Son K. Lam, Zachary R. Hall, and Jeffrey P. Boichuk (2014), "Know Your Customer: How Salesperson Perceptions of Customer Relationship Quality Form and Impact Account Profitability," Journal of Marketing, 78 (November), 38-58. Netzer, Oded, James M. Lattin, and Vikram Srinivasan (2008), "A Hidden Markov Model of Customer Relationship Dynamics," Marketing Science, 27 (March/April), 185-204. Palmatier, Robert W., Rajiv P. Dant, and Dhruv Grewal (2007), "A Comparative Longitudinal Analysis of Theoretical Perspectives of Interorganizational Relationship Performance," Journal of Marketing, 71 (October), 172-94.,,, and Kenneth R. Evans (2006), "Factors Influencing the Effectiveness of Relationship Marketing: A Meta-Analysis," Journal of Marketing, 70 (October), 136-53., Mark B. Houston, Rajiv P. Dant, and Dhruv Grewal (2013), "Relationship Velocity: Toward A Theory of Relationship Dynamics," Journal of Marketing, 77 (January), 13-30., Lisa K. Scheer, Kenneth R. Evans, and Todd J. Arnold (2008), "Achieving Relationship Marketing Effectiveness in Business-to-Business Exchanges," Journal of the Academy of Marketing Science, 36 (Summer), 174-90. Ping, Robert A. (1993), "The Effects of Satisfaction and Structural Constraints on Retailer Exiting, Voice, Loyalty, Opportunism, and Neglect," Journal of Retailing, 69 (Fall), 320-52. (1999), "Unexplored Antecedents of Exiting in a Marketing Channel," Journal of Retailing, 75 (Summer), 218-41. Rindfleisch, Aric and Jan B. Heide (1997), "Transaction Cost Analysis: Past, Present, and Future Applications," Journal of Marketing, 61 (October), 30-54. Ring, Peter Smith and Andrew H. Van de Ven (1994), "Developmental Processes of Cooperative Interorganizational Relationships," Academy of Management Review, 19 (January), 90-118. Rousseau, Denise M., Sim B. Sitkin, Ronald S. Burt, and Colin Camerer (1998), "Not So Different After All: A Cross-Discipline View of Trust," Academy of Management Review, 23 (July), 393-404. Rusbult, Caryl E., Isabella M. Zembrodt, and Lawanna K. Gunn (1982), "Exit, Voice, Loyalty, and Neglect: Responses to Dissatisfaction in Romantic Involvements," Journal of Personality and Social Psychology, 43 (December), 1230-42. 35

Stephen A. Samaha, Joshua T. Beck, and Robert W. Palmatier (2014), The Role of Culture in International Relationship Marketing, Journal of Marketing, 78 (September), 78-98. Samaha, Stephen A., Robert W. Palmatier, and Rajiv P. Dant (2011), "Poisoning Relationships: Perceived Unfairness in Channels of Distribution," Journal of Marketing, 75 (May), 99-117. Sirdeshmukh, Deepak, Jagdip Singh, and Barry Sabol (2002), Consumer Trust, Value and Loyalty in Relational Exchanges, Journal of Marketing, 66 (January), 15-37. Wathne, Kenneth H., and Jan B. Heide (2000), Opportunism in Interfirm Relationships: Forms, Outcomes, and Solutions, Journal of Marketing, 64 (October), 36-51. Wilson, David T. (1995), "An Integrated Model of Buyer-Seller Relationships," Journal of the Academy of Marketing Science, 23 (Fall), 335-45. Zaheer, Akbar and N. Venkatraman (1995), Relational Governance as an Interorganizational Strategy: An Empirical Test of the Role of Trust in Economic Exchange, Strategic Management Journal, 16 (June), 373 92. Zhang, Jonathan Z., Oded Netzer, and Asim Ansari (2014), Dynamic Targeted Pricing in B2B Relationships, Marketing Science, 33 (May-June), 317-337. 36

TABLE 1 Summary of Customer Relationship States Research Reference Methodology Number of Relationship States Relationship State Variables Relationship State Conceptualizations Conceptual Dwyer et al. (1987) Conceptual 6 Trust, commitment, dependence, norms Awareness exploration expansion commitmentdecline dissolution Rousseau et al. (1998) Conceptual 3 Trust Building stability dissolution Ring and Van de Ven (1994) Conceptual 4 Trust, dependence, norms Negotiation agreement execution assessmentdissolution Lewicki and Bunker (1996) Conceptual 4 Trust, identification Stage 1 (calculus and deterrence based trust), stage 2 (knowledge based trust), stage 3 (identification based trust), decline Heide (1994) Conceptual / partial empirical test 3 Dependence, norms Initiation maintenance termination Wilson (1995) Conceptual 5 Trust, commitment, social bonds, satisfaction, cooperation Johnson and Selnes (2004) Conceptual, using simulation Partner selection defining purpose settling relationship boundaries creating relationship valuerelationship maintenance 3 Trust, commitment, satisfaction Acquaintances friends partners Anderson and Narus (1991) Conceptual, using case studies N/A Dependence, norms Transactional vs. relational Empirical Jap and Ganesan (2000) Hibbard et al. (2001) Netzer, Lattin, and Srinivasan (2008) Cross sectional; age cohorts Cross sectional; age as covariate Longitudinal; hidden Markov model 4 Commitment, dependence, norms, satisfaction, bilateral investment, conflict 4 Trust, commitment, dependence, communication, shared values 3 Satisfaction, emotional connection, pride for subset of sample Exploration buildup maturity decline Quartile 1 (age: 1 96 months), quartile 2 (age: 97 160 months), quartile 3 (age: 161 236 months). quartile 4(age: 237+ months) Dormant occasional active Jap and Anderson (2007) Cross sectional; age cohorts 4 Trust, dependence, norms (goal congruence) risk seeking Exploration buildup maturity decline Palmatier et al. (2013) Longitudinal; latent growth curve N/A Commitment, commitment velocity Continuous dynamic state Luo and Kumar (2013) Longitudinal, HMM 3 None State 1 (lowest) State 2 State 3 (highest) 37

TABLE 2 Relationship Migration Mechanisms Migration Mechanisms Definition Pattern of Change in State Variables Description of Effect on Exchange Relevant State Transitions Migration Strategies References Exploration Establishing the conditions through which both parties are willing and able to develop the relationship. Stronger relational bonds through 8% to 14% increases in trust, customer commitment, and customer dependence. Significant increase in relational governance due to a positive evaluation of reciprocal behaviors as relational norms dramatically increase (47%). Exchange performance rapidly grows. Significant value creation through largest sales growth increase of any migration path. Transactional to transitional Communication. Product mix Dwyer, Schur and Oh (1987). Jap and Ganesan (2000). Morgan and Hunt (1994). Sirdeshmukh, Singh, and Sabol (2002). Mohr, Fisher, and Nevin (1996). Endowment Bilateral investment intended to bring mutual valuecreating opportunities to fruition. Relational bonding between partners as trust, customer commitment, and relational norms increase additional 11% to 15%. Additionally, large increase in dependence (3 5x more than other migrations) as both partners make non recoverable investments in the exchange. Cemented relational governance. Investments in the exchange enhance performance to its highest level (e.g., sales revenue, cooperation highest in the communal state). Transitional to communal Customer investment Seller investment Customer x seller investment De Wulf, Odekerken Schröder, and Iacobucci (2001). Heide and John (1990). Palmatier, Dant, and Grewal (2007). Kozlenkova, Samaha, and Palmatier (2013). Neglect Passive abuse indicating a lack of desire or resources to invest in the relationship. Passive, downward state change results from erosion in state variables (between 20% to 40% decay). Always ends in the transactional state due to reduced contact and reduced social exchanges with the partner firm (but not necessarily reduced economic exchanges). Communal to transactional Transitional to transactional Low communication Low customer investment Low seller investment Ping (1993, 1999). Joshi (2009). Koza and Dant (2007). Rusbult, Zembrodt, and Gunn (1982). Betrayal Actively undermining fairness and damaging relational governance. Immediate and dramatic state variable damage due to purposeful actions, typically accompanied by a strong emotional response Trust, customer commitment, and relational norms implode 40% to 60%. Relationship often only saved from termination by high dependence. Relational governance destroyed, with correspondingly low cooperation. Exchange performance weak, with low sales growth often anchored in dependence. Communal to damaged Transactional to damaged Conflict Injustice Conflict x injustice Grégoire and Fisher 2008 Hibbard, Kumar, and Stern (2001) Samaha, Palmatier, and Dant (2011) Dant and Schul (1992) Recovery Repairing of norms, rebuilding trust, and moving relationship out of damaged state. Entails substantial improvement in state variables, with trust, customer commitment, and relational norms typically needing greater than 40% improvement. Must overcome relatively "sticky" nature of damaged relationship. Conflict and injustice resolved to repair relational governance and improve cooperation Exchange performance improves from near 0% sales growth to moderate levels (6%) Damaged to transactional Communication Compromise Communication x compromise Ganesan (1993) Dant and Schul (1992) 38

Constructs M SD α 1 2 3 4 5 6 7 8 9 10 11 12 13 Relationship State Variables 1. Customer trust 4.16.69.94 2. Customer commitment 4.35.67.91.57* 3. Customer dependence 3.12.68.87.11.12* 4. Relational norms 3.59.81.93.52*.51*.11* Performance Outcomes 5. Profit 3.92.63.92.32*.38*.06.33* 6. Sales growth (%) 1.00 11 NA.02.01.04.03.01 7. Sales (thousands of $) 686 516 NA.21*.15*.02.17*.04.02 Migration Strategies 8. Communication 3.94.62.93.53*.44*.06.43*.19*.02.20* 9. Product mix 4.11.49.81.49*.40*.17*.45*.22*.01.02.38* 10. Customer investment 3.78.68.91.30*.31*.04.37*.26*.02.05.38*.33* 11. Seller investment 3.73.70.89.45*.44*.16*.49*.31*.06.14*.50*.29*.52* 12. Injustice 2.35.72.92.50*.37*.09.52*.27*.01.18*.52*.38*.40*.49* 13. Conflict 1.61.67.92.48*.44*.11.37*.22*.02.15*.45*.35*.30*.37*.36* 14. Compromise 3.54.63.92.45*.40*.13*.48*.28*.02.18*.41*.43*.30*.45*.34*.30* * Correlation is significant at the.01 level (two tailed). TABLE 3 Descriptive Statistics and Correlations 39

TABLE 4 Results: Hidden Markov Model of Relationship States State Name: Transactional Transitional Communal Damaged Relationship State Variables Trust (Mean) (1 5 scale) 3.74* 4.02* 4.75* 2.58* Customer commitment (Mean) (1 5 scale) 3.85* 4.24* 4.80* 2.73* Customer dependence (Mean) (1 5 scale) 2.58 2.83 3.75* 3.29* Relational norms (Mean) (1 5 scale) 2.46* 3.48* 4.15* 1.60* Performance Outcomes Profit (Mean) (1 5 scale) 3.65 3.76 4.30 3.11 Sales growth (Annual %) 7% 22% 10% 2% Sales revenue ($Thousand) 611 670 900 687 Relationship State Descriptives Percent of sample across all years (%) 55% 8% 20% 17% Relationship duration (Years) 4.0 5.4 7.5 5.8 Migration Path Probabilities To: Transactional Transitional Communal Damaged Transition from: Transactional state 50.9% 34.5% 12.7% Transitional state 8.6% 28.4% 60.9% Communal state 13.1% 65.6% 19.0% Damaged state 35.7% 59.4% Model Fit Comparisons Three State Four State Five State Latent class (DIC) 18,992 16,972 17,981 HMM (DIC) 14,023 13,200 14,386 Note: " " indicates cell not statistically different from zero. * p <.05. 40

TABLE 5 Results: Effect of Path Specific Migration Strategies Migration Mechanisms Migration Probability Effect of Migration Strategies on Migration Probabilities Exploration Communication Product mix Interaction Transactional to Transitional 34.5%.47*.59* Endowment Transitional to Communal 60.1% Customer investment.48* Seller investment.65* Customer investment x Seller investment.37* Neglect Communal to Transactional 13.5% Low communication.24* Low customer investment.50* Low seller investment.37* Transitional to Transactional 9.0%.55*.51*.70* Betrayal Communal to Damaged 19.4% Conflict.56* Injustice.82* Conflict x Injustice.61* Transactional to Damaged 13.1%.39*.50*.40* Recovery Damaged to Transactional 35.9% Communication.08 Compromise.64* Communication x Compromise.44* * p <.05. 41

Positive Migration Mechanisms Communication Product Mix Seller Investment Customer Investment Compromise Exploration Transactional to transitional (Increase) 1.30 1.64 0.88 1.21 0.15 Endowment Transitional to communal (Increase) 0.35 0.16 1.13 0.80 0.82 Recovery Damaged to transactional (Increase) 0.16 0.19 0.15 0.80 1.45 Negative Migration Mechanisms 1 TABLE 6 Results: Effectiveness of Relationship State Specific Migration Strategies Communication Seller Investment Migration Strategy Elasticity 1 Customer Investment Conflict Injustice Neglect Transitional to transactional (Decrease) 2.61 4.95 3.15 Communal to tranactional (Decrease) 0.45 0.73 2.30 Betrayal Transitional to damaged (Increase) 0.96 1.96 Communal to damaged (Increase) 0.71 1.83 The point elasticities were calculated by increasing or decreasing a single migration strategy (e.g., communication) by one standard deviation and examining the percentagechange in migration probability from one relationship state to another. 42

FIGURE 1 Overview of Relationship States and Migration Mechanisms Strong Relationship State Endowment Communal State Exploration Transitional State Weak Relationship State Recovery Transactional State Neglect Betrayal Poor Relationship State Damaged State 43

APPENDIX Constructs and Measures (Scale Sources) Item Loadings Relationship State Variables Trust (Crosby, Evans, and Cowles 1990) I can count on [Seller] to be honest in their dealings with me..87 [Seller] is a company that stands by its word..87 I can rely on [Seller] to keep the promises they make to me..88 [Seller] is sincere in its dealings with me..90 Customer commitment (Kumar, Scheer, and Steenkamp 1995) We continue to represent [Seller] because it is pleasant working with them..83 We intend to continue representing [Seller] because we feel like we are part of the [Seller] family..91 We like working for [Seller] and want to remain a [Seller] agent..82 We are a [Seller] agent because we like what [Seller] stands for as a company..85 Customer dependence (Kumar, Scheer, and Steenkamp 1995) If for some reason, our relationship with [Seller] ended We would compensate for it by switching our effort to other lines we carry. [R].59 It would be relatively easy for us to diversify into selling new product lines. [R].99 Relational norms (Kaufmann and Dant 1992) Even if the costs and benefits are not evenly shared between us in a given time period, they balance out over time..80 We each benefit and earn in proportion to the efforts we put in..83 My business usually gets a fair share of the rewards and cost savings in doing business with [Seller]..85 In our relationship, none of us benefits more than one deserves..79 Performance Outcomes Profit (Lusch and Brown 1996) As compared to other similar [Seller] agents, our performance is very high in terms of Sales growth..71 Profit growth..78 Overall profitability..82 Migration Strategies Communication (Greenbaum, Holden, and Spataro 1983) Communications are prompt and timely..74 Information provided is relevant for decision making..79 Communications are complete..90 The channels of communication are well understood..83 Product mix (Kahn 1998) [Seller's category] lines represent superior value for our customers..68 [Seller's category] lines have enough variety and assortment..66 [Seller]'s big ticket items provide a rich assortment to the customers..76 [Seller]'s big ticket goods represent superior value for our customers..78 Customer investment (Heide and John 1988) I have invested a lot of myself as a [Seller] s agent, in terms of... Time, beyond normally expected, in order to make [Seller] business successful..82 Effort, beyond normally expected, in order to make [Seller] business successful..80 Personal sacrifices (e.g., lost opportunities for other jobs, vacations, etc.)..88 Professional knowledge as a retailer..83 Seller investment (Zaheer and Venkatraman 1995) [Seller] has invested significant resources in providing me ongoing training..74 [Seller] has invested significant resources in providing me customized support..82 [Seller] has invested significant resources in improving personal relations between us..89 Injustice (Kumar, Scheer, and Steenkamp 1995) Our earnings from [Seller] business are fair given... The duties and responsibilities that I perform for [Seller]..87 What other [Seller] agents earn in markets similar to mine..59 What [Seller] earns from its sales through my business..90 The contributions I make towards [Seller]'s marketing effort in my market..80 Conflict (Kumar, Scheer, and Steenkamp 1995) OVERALL, I consider my relationship with [Seller] to be Full of ill will..77 Frustrating..75 Threatening..66 Hostile..67 Compromise (Ganesan 1993) In our disputes with [Seller], they usually... Try to find the middle ground between our position and theirs..83 Try to soothe our feelings and preserve our relationship by meeting us half way..87 Try to find a fair combination of gains and losses for both of us..91 Let us have some of our positions if we let them have some or theirs..66 Notes: All items were measured using five point scales, from 1="strongly disagree" to 5="strongly agree." [R]=reverse coded. 44