Managing Knowledge Transfer in Software-Maintenance Outsourcing Transitions: A System-Dynamics Perspective

Abstract

The existing literature suggests that transitions in software-maintenance offshore outsourcing projects are prone to knowledge transfer blockades, i.e. situations in which the activities that would yield effective knowledge transfer do not occur, and that client management involvement is central to overcome them. However, the theoretical understanding of the knowledge transfer blockade is limited, and the reactive management behavior reported in case studies suggests that practitioners are frequently astonished by the dynamics that may give rise to the blockade. Drawing on recent research from offshore sourcing and reference theories, this study proposes a system dynamics framework to explain why knowledge transfer blockades emerge and how and why client management can overcome the blockade. The results suggest that blockades emerge from a vicious circle of weak learning due to cognitive overload of vendor staff and resulting negative ability attributions that result in reduced helping behavior and thus aggravate cognitive load. Client management may avoid these vicious circles by selecting vendor staff with strong prior related experience. Longer phases of coexistence of vendor staff and subject matter experts and high formal and clan controls may also mitigate vicious circles.

Appendix: Model Assumptions

1.1 Assumptions in Model 1

The following assumptions are made in model 1:

A1: The values of all rates, auxiliaries, and constants and the start values of stock variables are within a range of 0 (very low) to 1 (very high).

A2: The regression coefficients reported in Krancher and Dibbern (2012) reflect the strengths of the relationships of cognitive load with its antecedents.

A3: The inverted-U-shaped relationship between cognitive load and learning effectiveness obeys the following functional form, where a is a parameter indicating the sensitivity to high or low cognitive loads and 0.5 is assumed to be the optimal level of cognitive load:

$$learning\,effectiveness = e^{{ - a \cdot (cognitive\,load - 0.5)^{2} }}$$

A4: The following integral function describes the evolution of expertise (ex) in function of time t, where b is a parameter for adjusting the scale between learning effectiveness and expertise (learning rate base value):

$$ex(t) = \int\limits_{x = 0}^{t} {b \cdot learning\,effectiveness(x)} \,{\text{dt}}$$

A5: The vendor engineer is able to solve a task to the satisfaction of the client if the cognitive load is minor than or equal to .5.

A6: The following values have been chosen for the parameters: a = 16; b = 0.05.

1.2 Additional Assumptions in Model 2

Model 2 makes the following assumptions in addition to the assumptions made in model 1:

A7: Simple-to-complex sequencing adjusts task complexity so that cognitive load is closer to a medium level (0.5). Task complexity after simple-to-complex sequencing is therefore calculated as follows, where c indicates the magnitude of simple-to-complex sequencing:

$$task\,complexity_{afterSTCS} = task\,complexity_{beforeSTCS} + c \cdot (.5 - cognitive\,load_{beforeSTCS} )$$

A8: No help is provided if the cognitive load after simple-to-complex sequencing is below a medium level (0.5); else, help is calculated as follows, where d indicates the base line magnitude of help provided (help rate):

$$help = d \cdot (cognitive\,load_{afterSTCS} - .5)$$

A9: The following values have been chosen for the parameters c and d: c = .5; d = 1.

1.3 Additional Assumptions in Model 3

Model 3 makes the following assumptions in addition to the assumptions made in model 2:

A10: Control is the weighted sum of FCC and self-control, where f denotes the weight of self-control:

$$control = f \cdot selfcontrol + (1 - f) \cdot FCC$$

A11: Self-control is equally determined by expertise (ex) and ability trust:

$$selfcontrol = \tfrac{1}{2}(ex + abiltiy\,trust )$$

A12: Ability trust increases or decreases in function of cognitive load, where g denotes the sensitivity to latest cognitive load levels:

$$\begin{aligned} ability\,trust(t) = & (1 - g) \cdot ability\,trust(t - 1) \\ & + g \cdot \hbox{max} (0,\hbox{min} (1,2 - 2 \cdot cognitive\,load)) \\ \end{aligned}$$

A13: The following values have been chosen for the parameters f and g: f = .5; g = .1.