TOWARDS MORE EFFECTIVE SERVICE MANAGEMENT DECISION MAKING: DESIGN AND APPLICATION OF AN OPTIMIZATION FRAMEWORK IN A FRONTLINE EMPLOYEE MANAGEMENT CONTEXT

1
TOWARDS MORE EFFECTIVE SERVICE MANAGEMENT
DECISION MAKINGDESIGN AND APPLICATION OF AN
OPTIMIZATION FRAMEWORK IN A FRONTLINE EMPLOYEE
MANAGEMENT CONTEXT

• PROF. DR. SANDRA STREUKENS
• HASSELT UNIVERSITY
• FACULTY OF APPLIED ECONOMICS
• DEPARTMENT OF BUSINESS STUDIES

2
OUTLINE

• INTRODUCTION
• A primer in services marketing
• What we do (not) know
• Research objective
• Importance of this study
• MODEL DEVELOPMENT
• Overview of conceptual model
• Model development
• Estimation and calibration of the decision-making
  model
• Estimation of the behavioral model
• Example application
• DISCUSSION
• Implications
• Limitations and further research

3
INTRODUCTIONA primer in services marketing

• Services are processes (van Looy et al. 2003)
• Pure services services accompanying
  goods/products
• Flight on an airplane
• Consulting with an accountant
• Haircut
• Attending a university
• Training for a new manufacturing system
• Service delivery involves a game between people
  (employee-customer interaction)
• In services the service employee plays a crucial
  role

4
INTRODUCTIONWhat do we know

• The key to an effective service organization
  starts with managing employees perceptions
  regarding their own organization (Schneider and
  Bowen, 1993 Rogg et al. 2001)
• More specifically, ample empirical evidence for
  the positive relationships between employee
  perceptions, customer evaluative judgments, and
  financial performance (de Jong et al. 2004a
  Schneider et al. 1998 Kamakura et al. 2002)

5
INTRODUCTION What we do not know

• Despite the large body of knowledge regarding
  service management, there are hardly any
  practical decision making models that make use of
  this research.
• One the other hand, OR scholars call for the
  development of service decision making models
  that infuse behavioral data in their (so far)
  purely mathematical model (Bretthauer, 2004
  Boudreau et al. 2003).

6
INTRODUCTIONResearch objective

• To develop and demonstrate a practical and
  versatile decision-making tool that assists
  managers in evaluating and optimizing service
  improvement initiatives in an economically
  justified, yet behavioral oriented manner.
• More generally, the aim is to design a
  decision-making tool that assists managers in
  evaluating and optimizing decisions regarding
  soft measures (perceptions) using hard
  modeling.

7
INTRODUCTIONImportance of this study

• We live in a service economy
• Currently, services make up approx. 75 of the
  GDP in Belgium and of all workers approx. 70
  works in the service sector.
• Service managers should be increasingly results
  oriented
• (1) slow growth mature markets
• (2) increasing (inter)national competition.
• Customers become an increasingly scarce resource
  being pursued by an increasing number of service
  providers

8
INTRODUCTIONImportance of this study

• CONTRIBUTION TO THE ACADEMIC LITERATURE

9
MODEL DEVELOPMENTConceptual model

• AT A MACRO LEVEL

10
MODEL DEVELOPMENTConceptual model

• AT A MICRO LEVEL

11
MODEL DEVELOPMENTModeling revenues A
behavioral approach

• MODELING REVENUES A BEHAVIORAL APPROACH

12
MODEL DEVELOPMENTModeling Revenues A
Behavioral Approach

• GENERAL
• Behavioral approach is rooted in the SPC
  literature
• Operations researchers call for the infusion of
  perceptual data in decision making
• Employee Customer Revenues Chain
• A key role for employee well-being climate,
  service climate, and customer evaluative
  judgments
• The effects of the behavioral approach on
  investment profitability is reflected by link 1
  in the conceptual model
• CUSTOMER EVALUATIVE JUDGMENTS
• Customer evaluative judgments are predictors of
  financial performance (Kamakura et al. 2002)
• Pivotal constructs here are perceived quality,
  customer satisfaction, and behavioral intent
  (Cronin et al. 2002)

13
MODEL DEVELOPMENT Modeling revenues A
behavioral approach

• SERVICE CLIMATE
• One of the most relevant contributors in the
  forming favorable customer evaluative judgments
  (de Jong et al. 2004a)
• EMPLOYEE CLIMATE
• Employee climate is a key determinant of service
  climate (Parker, 1999)
• Dimensions rewards orientation, means emphasis,
  goal emphasis, management support, workgroup
  support, and interdepartment service (Burke et
  al. 1992 Schneider et al. 1998)
• Generalizable across settings (Kopelman et al.
  1990)
• Can be effectively influenced by targeted
  investments (Harter et al. 2002)
• An overview of the literature underlying these
  links is available upon request

14
MODEL DEVELOPMENT Modeling revenues A
behavioral approach

• THE REVENUES FUNCTION
• Using the approach developed by Streukens and de
  Ruyter (2004) we conclude that all relationships
  in our behavioral model are linear
• Hence, revenues vary as a linear function of
  changes in employee well-being dimensions and can
  be compactly expressed as

15
MODEL DEVELOPMENT Modeling revenues A
behavioral approach

• PARAMETERS REVENUE FUNCTION

16
MODEL DEVELOPMENTModeling effort (In)direct
effects

• MODELING EFFORT (IN)DIRECT EFFECTS

17
MODEL DEVELOPMENTModeling effort (In)direct
effects

• INVESTMENT EFFORT AND PROFITABILITY
• A positive indirect effect (i.e. link 2 in
  conceptual model)
• A negative direct effect (i.e. link 3 in
  conceptual model)
• INDIRECT EFFECT
• Investment effort ? employee perceptions ?
  customer perceptions
• ? revenues ? profitability (all positive
  relationships)
• DIRECT EFFECT
• Profits Revenues Investment effort

18
MODEL DEVELOPMENTModeling effort Indirect
effects

• Modeling the effect between investment effort and
  level of input variables
• Decision calculus approach
• ADBUDG-model developed by Little (1970)
• ABDUDG is simple, robust, easy to control,
  adaptive, as complete as possible, and easy to
  communicate with (Little, 1970 p.466)
• ABDUDG adheres to Blattberg and Deightons (1990)
  50-50 rule

19
MODEL DEVELOPMENTModeling Effort Indirect
effects

• THE ADBUDG MODEL

Level input variable
Investment effort
Level of when
Level of when
Shape parameter
Shape parameter
20
MODEL DEVELOPMENTModeling effort Direct effects

• Requires an estimate of the total investment
  effort
• As refers to the monetary investment
  regarding input variable the total direct
  investment effort equals
• The term is the investment effort needed
  to maintain the current level of the various
  input variables
• To capture the direct effect of investment effort
  in our approach the total investment effort needs
  to subtracted from revenues (i.e., link 3)

21
MODEL DEVELOPMENTProfit function

• PROFIT FUNCTION
• PROFIT OPTIMIZATION
• Profit optimization crucial decision making theme
  in services (Zeithaml 2000).
• The above profit function will serve as an
  objective function is an optimization framework.
• Optimization of the profit function is subject to
  several constraints.

22
MODEL DEVELOPMENTProfit function

• CONSTRAINT 1
• Total investment effort cannot exceed a pre-set
  budget or spending limit (Budget constraint)
• CONSTRAINT 2
• Non-negativity constraint investment effort

23
MODEL DEVELOPMENT Profit function

• CONSTRAINT 3
• Relationship between investment effort and the
  input variables
• CONSTRAINT 4
• The level of input variable after implementation
  of the investment strategy should be at least
  equal to its starting level

24
MODEL DEVELOPMENTOverview

• OVERVIEW DECISION MAKING APPROACH

25
MODEL DEVELOPMENTEstimation behavioral model

• EMPIRICAL STUDY
• Estimation revenue formation process (i.e.
  employee-customer-revenues chain)
• Actual data on employee perceptions, customer
  evaluative judgments, and revenues
• Please note that all scale items used in this
  study are available upon request!

26
MODEL DEVELOPMENTEstimation behavioral model

• SAMPLING
• Employees and business customers from an
  internationally operating firm in office
  equipment.
• Census of 250 employees in 28 teams (on average
  n8 per team). Effective sample size n 169.
• Random selection of 1500 customers meeting the
  following criteria (1) active in retail setting
  (2) at least 24 month customer (3) at least two
  times contact with service employees during last
  12 months. Effective sample size n 499. (Min. 5
  customers / team Max. 38 customers / team).

27
MODEL DEVELOPMENTEstimation behavioral model

• EMPLOYEE SURVEY
• Despite the fact that researchers agree upon the
  positive relationship between employee climate
  and service climate, there exists no measurement
  scale for employee climate (Parker, 1999).
• Careful investigation of the theoretical contents
  of the employee climate constructs (work of Burke
  et al. 1992 Schneider et al. 1998). Find
  existing validated scales that cover the contents
  of the constructs

28
MODEL DEVELOPMENTEstimation behavioral model

• EMPLOYEE SURVEY
• Rewards orientation (4 items), Boshoff and Allen
  (2000).
• Means emphasis (4 items), Iverson (1992)
• Goal emphasis (4 items), Sawyer (1992)
• Management support (7 items), House and Dessler
  (1974)
• Work group support (7 items), Beehr (1976)
• Interdepartment service (5 items), adapted from
  Schneider et al. (1998)
• Service climate (8 items), Schneider et al.
  (1998)
• All constructs measured on a 9-point Likert scale

29
MODEL DEVELOPMENTEstimation behavioral model

• CUSTOMER SURVEY
• Perceived quality (9 items), self designed cf.
  Rust et al. (1995)
• Overall satisfaction (1 item), Anderson et al.
  (1997)
• Behavioral intent (2 items), Zeithaml et al.
  (1996)
• All constructs measured on a 9-point Likert scale
• FINANCIAL DATA
• Internal company records on each customers sales
  history (i.e. revenues). Data covering a 12
  months period after the questionnaires were sent
  out.
• DATA LINKAGE
• Employee perceptual data , customer perceptual
  data, and customer financial data were linked by
  means of the customers unique client number.
  Providing client number on questionnaire
  incentive.

30
MODEL DEVELOPMENTEstimation behavioral model

• ASSESSMENT PSYCHOMETRIC PROPERTIES
• Partial Least Squares (PLS) estimation
• For the employee data the 1-to-10 parameter to
  sample size ratio was not met (cf. Raykou and
  Widaman, 1995 Bentler and Chou, 1987).
• Both reflective and formative were employed in
  our study.
• UNIDIMENSIONALITY
• First eigenvalue greater than 1 criterion (cf.
  Tenenhaus et al. 2005)
• All reflective scales met this criterion
• INTERNAL CONSISTENCY RELIABILITY
• For all reflective constructs ? gt 0.70 (cf.
  Nunnally and Bernstein, 1994)

31
MODEL DEVELOPMENTEstimation behavioral model

• CONVERGENT VALIDITY
• Tested for all reflective scales
• All loadings significant and gt 0.50 (cf. Anderson
  and Gerbing, 1988)
• All average variance extracted value gt 0.50
• CONTENT VALIDITY
• Key validity type for formative scales
• Scale designed to cover all relevant aspects of
  the construct (cf. Jarvis et al., 2003)
• Magnitude and significance of the loadings
  defining the formative relationships evidence
  relevance of the indicators (cf. Diamantopoulos
  and Winklhofer, 2001)

32
MODEL DEVELOPMENTEstimation behavioral model

• DISCRIMINANT VALIDITY
• Correlations between construct pairs did not
  include an absolute value of 1 in their 95
  confidence intervals (both reflective and
  formative scales).
• Average variance extracted gt squared value
  correlation coefficient (only for reflective
  scales).

33
MODEL DEVELOPMENTEstimation behavioral model

• COMPLEX DATA STRUCTURE
• Employee part employees nested within teams
• Linkage part customers are nested within teams
• Customer part between-person structure
• RESULTING ANALYSIS STRATEGY
• Employee part 2-level HLM (cf. de Jong et al.,
  2004a b)
• Linkage part 3-level HLM (cf. de Jong et al.,
  2004a b)
• Customer part SUR
• ANALYTICAL SOFTWARE
• HLM models estimated in Mlwim
• SUR model estimated using SAS PROC SYSLIN

34
MODEL DEVELOPMENTEstimation behavioral model

• ASSESSING THE DATAS SUITABILITY FOR HLM
• Interrater-agreement r(WG) (cf. James et al.,
  1993)
• Intra Class Correlation ICC(1) and ICC(2) (cf.
  Bliese, 2000)
• All three measures provide justification for
  aggregation of the data

35
MODEL DEVELOPMENTEstimation behavioral model

• 2-LEVEL HLM EMPLOYEE PART

36
MODEL DEVELOPMENTEstimation behavioral model

• 3-LEVEL HLM LINKAGE PART
• Level 1 perceived service quality (m qual01
  qual09)
• Level 2 individual customer (i 1 499)
• Level 3 team which serves customer (j 1-28)

37
MODEL DEVELOPMENTEstimation behavioral model

• SUR MODEL CUSTOMER PART

38
MODEL DEVELOPMENTResults behavioral model

• EMPLOYEE PART
• At the individual level rewards orientation (b
  0.23) goal emphasis (b 0.13) management
  support (b 0.23) work group support (b
  0.10) and interdepartment service (b 0.20)
  have significant impact on service climate.
• At the group level none of the hypothesized
  antecedents has a significant impact on service
  climate
• LINKAGE PART
• Service climate has a positive and significant
  impact on all quality dimensions (qual01 (b
  0.90) qual02 (b 0.74) qual03 (b 0.76)
  qual04 (b 0.57)qual05 (b 0.39) qual06
  (b 0.36) qual07 (b 0.40) qual08 (b
  0.42) qual09 (b 0.50))

39
MODEL DEVELOPMENT Results behavioral model

• CUSTOMER PART
• Perceived quality has a positive and
  significant impact on overall satisfaction (b
  0.73)
• Behavioral intentions is positively and
  significantly influenced by perceived quality
  (b 0.19) and overall satisfaction (b 0.51)
• Behavioral intentions has a positive and
  significant impact on revenues (b 1092.80)
• OVERALL
• We find empirical support for an
  employee-customer-revenues chain of effects
• Using these empirical results we can determine
  how much revenues vary as a function of changes
  in the employee climate perceptions
• We have insight in the revenue part of our
  decision making model (i.e. link 1)

40
SERVICE MANAGEMENT DECISION MAKINGExample
application

• EXAMPLE ILLUSTRATION OF DECISION MAKING MODEL
• An exact description of the investment actions
  and the involved costs and profits were not
  allowed to be made public by the company at which
  we collected data.
• Hence, fictive numbers are used demonstrating the
  decision making model (i.e. regarding link 2 and
  link 3)
• This is no problem, as in contrast to the
  empirical study described above the figures on
  the investment actions are completely company
  specific and do not allow for making
  generalization to other settings.

41
SERVICE MANAGEMENT DECISION MAKINGExample
application

• DECISION MAKING MODEL
• Determining optimal level investment effort
• Calculation rate of return (ROI)
• Determining optimal allocation of the investment
  efforts
• Assessing the robustness of the optimal solution
  (risk)

42
SERVICE MANAGEMENT DECISION MAKINGExample
application

• INVESTMENT STRATEGY
• Emphasis on revenues expansion rather than cost
  reduction (cf. Rust et al. 2000).
• In line with the customization-standardization
  trade-off explained by Anderson et al. 1997)
• The literature shows that revenue expansion,
  customization, and satisfaction are related
• Focus on defensive strategy (cf. Fornell and
  Wernerfelt 1987, 1988)
• Thus, maximize profitability through increasing
  revenues from existing customers

43
SERVICE MANAGEMENT DECISION MAKING Example
application

• OPTIMIZATION FRAMEWORK REVENUE FUNCTION
• Parameters di and y(0)i follow directly from the
  empirical study
• d1(ROR) 436.74 d2(GEMP) 246.86 d3(MSUP)
  436.74 d4(WGS) 189.89 d5(IDS) 379.78
• y(0)1(ROR) 5.60 y(0)2(GEMP) 5.12
  y(0)3(MSUP) 5.10 y(0)4(WGS) 5.68 y(0)5(IDS)
  3.97
• The value for parameter yi is determined via the
  ADBUDG function
• N 10,000

44
SERVICE MANAGEMENT DECISION MAKINGExample
application

• OPTIMIZATION FRAMEWORK REVENUE FUNCTION
• Some background info on calculating the di
  parameter
• Assume the following (a-cyclical) model
• The impact of variable yi on rev (i.e., di) is
  the sum of all paths connecting yi and rev
• Thus, ?1 (ß1 ß6)(ß1 ß5 ß7)(ß2 ß7) and
• ?2 (ß3ß6)(ß3 ß5 ß7)(ß4 ß7)

45
SERVICE MANAGEMENT DECISION MAKING Example
application

• OPTIMIZATION FRAMEWORK COST FUNCTION
• Calibration by means of the 4 standard ADBUDG
  questions
• If effort is reduced to 0 what will than be the
  evaluation regarding the input variable? This
  provides the value for parameter ai. The value of
  ai is typically the lowest value of the scale
  used to assess the perceptions regarding . In
  this case 1.

46
SERVICE MANAGEMENT DECISION MAKING Example
application

• OPTIMIZATION FRAMEWORK COST FUNCTION
• If effort approaches infinity what will be the
  value of the input variable? This answer provides
  the value for parameter bi. The value of bi is
  typically the highest value of the scale used to
  assess the perceptions regarding . In this case
  9.
• Regarding input variable i what is the current
  level of effort and to what evaluation does that
  lead?
• If compared to the current situation effort is
  doubled to what level of input variable would
  that lead?
• Questions 1 and 2 restrict function to meaningful
  range
• Questions 3 and 4 determine shape of the function
  (S-shaped or concave)

47
SERVICE MANAGEMENT DECISION MAKINGExample
Application

• OPTIMIZATION FRAMEWORK COST FUNCTION
• Having calibrated the ADBUDG functions for the
  various input variables (ROR, GEMP, MSUP, WGD,
  and IDS) automatically provides all input for the
  total level of investment effort (i.e., direct
  effect or link 3)
• METHODOLOGY
• Solving the optimization framework
• Non-linear programming using AIMMS

48
SERVICE MANAGEMENT DECISION MAKINGExample
application

• OPTIMIZATION ANALYSIS
• Investments remain feasible when the derivative
  of the objective function is positive
• Optimum of objective function is reached when its
  derivative is equal to zero
• Optimum of objective function is maximum level
  profitability
• Derivative profit function

49
SERVICE MANAGEMENT DECISION MAKINGExample
application

50
SERVICE MANAGEMENT DECISION MAKING Example
application

• RATE OF RETURN
• OPTIMAL SOLUTION
• Investment effort 23,000,000
• Profits 7,298,500
• Rate of return 31.71

51
SERVICE MANAGEMENT DECISION MAKINGExample
application

• OPTIMAL ALLOCATION
• Effort level and allocation of effort are equally
  important matters in making investment decisions
  (Mantrala et al. 1992).
• Question now is how to allocate the optimal
  effort level to indeed obtain the maximum level
  of profitability
• Guidance regarding the allocation of the
  investment effort can be directly obtained from
  the relative magnitudes of the derivatives.
• Remember that the partial derivative of the
  profit function with regard to yi reflects the
  change in profits obtained by investing an
  additional monetary unit in variable yi.

52
SERVICE MANAGEMENT DECISION MAKINGExample
application

• OPTIMAL ALLOCATION
• Thus, optimal allocation starts with directing
  all efforts to the input variable with the
  highest partial derivative
• Note that as investments are subject to
  diminishing returns, the partial derivative
  decreases
• When the highest partial derivative equals the
  second highest derivative, optimal allocation is
  obtained by spreading effort over the various
  alternatives as follows

53
SERVICE MANAGEMENT DECISION MAKINGExample
application

• OPTIMAL ALLOCATION

54
SERVICE MANAGEMENT DECISION MAKINGExample
application

• OPTIMAL SOLUTION
• What the various amounts mean in practical
  investment actions (e.g. Specific reward system)
  can be derived from the ADBUDG function

55
SERVICE MANAGEMENT DECISION MAKINGExample
application

• ROBUSTNESS / INVESTMENT RISK
• All investments are characterized by uncertainty
  regarding the projected outcome
• This uncertainty or variability concerning the
  projected outcome is referred to as risk
  (comparable to the definition of risk in finance)
• Robustness assessment by means of sensitivity
  analysis. That is how does the optimal solution
  respond to changes in the model parameters?
• Robustness assessment by means of calculation
  switching values. That is, how much can a
  coefficient drop until the investment results
  become economically infeasible / negative?

56
SERVICE MANAGEMENT DECISION MAKINGExample
application

• ROBUSTNESS SENSITIVITY ANALYSIS
• Numerical experiments
• Deviation in coefficient (5, 10) and the
  resulting percentual change in optimal solution.
• ROBUSTNESS SWITCHING VALUES
• Solving the optimization framework to determine
  per coefficient when profitability becomes zero.
• Note that the robustness is assessed by altering
  the di parameter in our decision making approach

57
SERVICE MANAGEMENT DECISION MAKINGExample
application

• RESULTS ROBUSTNESS ASSESSMENT

58
DISCUSSION

• Decision making model that allows to evaluate the
  financial consequences of service improvement
  initiatives in an economically sound manner,
  whilst guarding the firms key assets its
  employee and customers
• The model merges knowledge from service research
  with mathematical rigor
• Profit maximization
• Allocation of investment effort
• Risk analysis
• Integral empirical assessment employee-customer-re
  venues chain

59
LIMITATIONS AND FUTURE RESEARCH

• Cross sectional approach dynamic analysis
• Inclusion of customer characteristics
• Retention of customers and acquisition of new
  customers