Value from relationships a supplier perspective? Time Source: Anton 1996 1 Customer Defection and Win-back A causal analysis Prof. Dr. Doreén Pick, Marketing-Department, Freie Universität Berlin 12th, October 2010 2010_10_11
Customer defection due to discussions about salary dumping Regarding a study conducted by the market research firm GfK, the profit of German discount pharmacy Schlecker has decreased by 16 percent within the first four months in 2010. More than one million customers have defected. Source: o. V. (2010) in: Spiegel Online vom 05.06.2010, verfügbar unter: http://www.spiegel.de/wirtschaft/unternehmen/0,1518,698966,00.html 3 Many industries are facing customer defection. And they generate defection by themselve. Telekom: Customers massively defect (Focus, August 2006) Every third Eon customer thinks about switching (Welt, October 2007) Vattenfall loses customers (taz, July 2009) More and more firms try to win new customers by specific offers such as switching rewards. But isn`t this impacting their existing customers? 4
1. Basics about customer defection and win-back management 2. Investigation of willingness to return Ph.D. thesis 3. Procedure for structural equation modeling 4. Summary 5 1. Basics about customer defection and win-back management 6
Definition Customer defection is a conscious decision of a customer to end a relationship with a firm. This relationship can embrace all contracts or just one. This relationship termination can be done verbally or written, both is an explicite communication with a firm. 7 Reasons for customer defection can be threefold. Competitor-related Customer-related Firm-related Main reasons Dissatisfaction (customer) Satiation / Variety Seeking (customer) Offers and alternatives (competitor) Service failure (firm) Ending of contracts (firm) Source: Sauerbrey/Henning (2000); Pick (2006): EMAC Conference. 8
But, customer also defect strategically. Customer objectives in defection process 1. Avoidance of financial loss of an un-needed contract a. Monthly costs in subscriptions, also in case of non-usage b. Defection as Time-bridging ; 37.9 % as the main defection reason 2. Gain of extra financial value from a product or service a. Defection-experienced customers b. Re-negotiations 9 Why should firms identify their customers defection? Financial costs - Loss of sales, contributions margins, market share - Loss of potential for sales, margins, market share - Costs of customer acquisition - Costs of customer win-back - Etc. Non-financial costs - Costs of negative WOM by lost customers (up to 40 negative comments) - Loss of efficient customer relationship management - Etc. Source: Pick, D. (2008): Wiederaufnahme vertraglicher Geschäftsbeziehungen. Eine empirische Untersuchung der Kundenperspektive, Gabler Verlag, Wiesbaden, here p. 33 f.; von Wangenheim (2005). 10
Customer win-back management consists of 5 phases. 1 2 3 4 5 Definition of objectives (direct goals, indirect goals) Defection analysis (identification, reasons, segmentation of lost customers) Win-back activities (dialogue, offer, timing) Win-back controlling (rate of win-back, win-back costs, benefits and return on winback) Follow up processes (win-back knowledge, integration into customer retention) Source: Pick, D./Krafft, M. (2009): Status quo des Rückgewinnungsmanagements, in: Link, J./Seidl, F. (Hrsg.): Kundenabwanderung. Früherkennung, Prävention, Kundenrückgewinnung, Gabler Verlag, Wiesbaden, p. 119-141, here p.126. 11 What drives defected customers to return to a former firm? Antecedents of the willingness to return Will customers return? Influence of win-back offers 12
General willingness to return The general willingness to return is an intention of a former customer to reactivate a former relationship with the firm by expecting no win-back activities from the firm. 13 Simple causal model Attitude 1 Attitude 2 Intentions Behaviour Attitude 14
2. Investigation of willingness to return Ph.D. thesis 15 Research framework Overall satisfaction A Locus C + - Commitment + Stability Controllability Customer Knowledge Perceived control of behavior GWR Reactivation costs - Switching experiences Attractiveness of Alternatives Return behaviour Variety Seeking B Specific WR 16
Two-step investigation online survey and experiment 1 2 Survey Subscription examples Win-Back offers/ incentives Win-Back contacts/ dialogue page 2 Generation of a data set to build a causal chain for willingness to return and real return (win-back) 17 Selected results of the structural equation model (n = 543) with PLS Path coefficients and level of significance Satisfaction GWR: 0.047 Commitment GWR: 0.252*** Reactivation costs GWR: -0.223*** Locus GWR: 0.082** Age GWR: 0.054** *: p<0.10 **: p<0.05 ***: p<0.01 18
Interpretation Main driver of general willingness to return are - Commitment - Reactivation costs Resulting managerial implications - Increase commitment to stabilize the relationship and consequently increase the probability of return - Lower the perceived reactivation costs (communication channels toward the customer, etc.) 19 3. Procedure for structural equation modeling 20
Steps for estimation of structural equation modelling 1. Generate hypothesis 2. Classify variables 3. Conceptualize structural model 4. Conceptualize measurement model 5. Estimation by Lisrel or PLS Steps for Lisrel include: Generate path diagram, generate equations 6. Evaluation of results (metrics) 21 Step 1: Generating hypotheses - Hypotheses are assumptions about (causal) relationships between variables. - Hypotheses are the answers of research questions. - If., then. Sources of hypotheses 1. Theory, such as Theory of Cognitive Dissonance or Attribution theory 2. Conceptual and empirical literature 3. Own ideas (plausibility) 22
Step 2: Classifying variables - Which variables (factors, constructs) are relevant to explain the - direct and - indirect relationships toward the dependent variable? - What linkages have these variables? - How many variables are necessary in the research framework (risk of low R 2, explanation power)? - Logical test: Are there really causal relationships or just correlations, such as number of deaths and orange sales? 23 Step 3: Conceptualize structural model Error terms 24
Step 4: Conceptualize measurement model - Measurement as a process to identify a phenomenon, which can not be seen or otherwise perceived, e.g. black box of consumer, satisfaction of customers, purchase intention - Main aspect: How can you describe what satisfaction of customers is? - Development of new scales or usage of established scales from empirical work (basis: good criteria, A journal, etc.) - 2 types of scales - Reflective variables - Formative variables 25 Difference between formative and reflective measurement of variables Wine Walking Prosecco Speaking Being drunken Beer Driving formative reflective 26
Where to get scales? Example: Behavioral Intentions (from very low to very high on a 9-point scale): - The probability that I will use this facility s services again is low/high. - The likelihood that I would recommend this facility s services to a friend is low/high. - If I had to do it over again, I would make the same choice. 27 Evaluation of reflective measurement models (criteria) Analysis Level Criteria Requirements Explorative factor analysis Confirmatory factor analysis Content validity Reliability Convergence validity Discriminance validity Global criteria (via AMOS, Lisrel) AVE Correlation Fornell-Larcker criteria GFI AGFI NFI RMR Legend: r 2 quadrated correlation between variables Kaiser-Meyer-Olkin (KMO) Explained variance Cronbach Alpha Item-To-Total correlation Indicator reliability Factor reliability 0.5 50% 0.7 (more than 4 indicators) Cronbach Alpha 0,7, Elimination of the item with lowest correlation 0.4 0.6 0.5 < 0.9 AVE > r 2 0.9 0.9 0.9 < 0.1 28
Measurement example from GWR model Variable Label of item Indicator reliability Construct reliability AVE ALTATTR2 0.589 0.850 0.535 ALTATTR3 0.582 Attractiveness of alternatives ALTATTR4 0.492 ALTATTR5 0.648 ALTATTR6 0.360 Explorative Factor analysis KMO: 0.807 Explained Variance: 62.45% Cronbach Alpha: 0.842 Global Criteria GFI: 0.992 AGFI: 0.976 RMR: 0.130 NFI: 0.985 29 Evaluation for formative measurement models Multicollinearity = statistical phenomenon in which two or more predictor variables in a multiple regression model are highly correlated. Test for Multicollinearity - Pairwise correlations - Variance Inflation Factor (VIF), (< 10, < 8, <3) 30
Overall SEM Measurement model of latent exogen variable Measurement model of latent endogen variable Structure model Formative model Reflective model δ 1 δ 2 Indicator λ 11 x 1 ξ 1 Indicator x 2 Indicator y 1 Indicator y 2 λ 21 π 11 π 12 γ 11 η 1 ζ 1 γ 21 β 21 η 2 ζ 2 λ 32 λ 42 Indicator y 1 ε 1 Indicator y 2 ε 2 Reflective model Measurement model of latent endogen variable 31 Step 5 and 6: Estimation and evaluation of SEM by Partial Least Square (PLS) - There are no global criteria for variance-based models such as PLS. - This is different to covariance-based models such as AMOS, Lisrel. Local criteria are: - Path coefficients and their significance level - R² (> 0.4) - f²-effect size (> 0.35 huge, > 0.15 medium, ~0.02 small) 32
Screenshot of PLS SmartPLS freeware, available under: www.smartpls.de 33 4. Summary 34
Challenges - SEM can not test causality with 100 percent certainty. - The procedure just estimates a statistical dependence by covariates and correlations. This correlations are necessary but do not have to be sufficient. - To proof the causal relationship between variables, theories have to be taken into consideration. - Decision for covariance-based (e.g., Lisrel) or variance-based (PLS) models: different heuristics, such as sample size, existence of several formative variables (see Reinartz, Haenlein, and Henseler 2009) - Linear regression vs. SEM: If there are mediating and moderating effects, linear regression can not measure the impact on the dependent variable(s). 35 Thank you very much for your attention. Contact: Prof. Dr. Doreén Pick Marketing-Department Freie Universität Berlin Tel.: +49 (30) 838 54547 E-Mail: doreen.pick@fu-berlin.de 36
Structure and results of the win-back experiment publishing house Structure n No. of returner Rate of winback Dialogue by Phone (m=1) E 111 O 1 X 1 O 8 79 12 15.2% E 112 O 2 X 2 O 9 77 0 0% E 113 O 3 X 3 O 10 79 4 5.1% Dialogue by Printmailing (m=2) E 121 O 4 X 4 O 11 77 0 0% E 122 O 5 X 5 O 12 78 0 0% E 123 O 6 X 6 O 13 76 0 0% Control group K 3 O 7 77 0 0% O 14 E lmn : Experimentgruppe, l für die Art der Gruppe (l = 1, 2), (m = 1, 2), n für das Angebot (n = 1, 2, 3) O p : Messung der Befragung und des Rückkehrverhaltens, (p = 1,...7: Befragung) (p = 8,... 14: Rückkehrverhalten) X q : Experimentmaßnahme als entsprechender Stimulus (q = 1,., 6) K 3 : Kontrollgruppe für die Befragungsgruppe 37 38
Equation system for Lisrel Structure model Measurement model of the latent exogene variable Measurement model of the latent endogene variable Source: Kreis, H. (2010), Presentation in doctoral colloquium, Berlin, June. 39