Article Preview
TopIntroduction
Decision-making is a management competence (Shewhart, 1980; Laudon & Laudon, 2013) encompassing four main stages: (i) the intelligence to discover the organizational problem, then (ii) the design of potential solutions, afterward (iii) choose the best solution, and finally, (iv) implement the solution and check if it fulfills the desired goals. These traditional stages support many levels of organizational management, e.g., project management, operational management or middle management. Under complex organizational environments, those stages are not always completely feasible due to: (i) Inability to map the current operational observations with the current state where the organization actually is (Weber, 1987; Montiel & Bickel, 2012), e.g., when actors perform workarounds (Alter, 2014) and override the previous defined prescriptions then the manager need to collect more information to interpret what, in fact, was executed; and (ii) Incomplete observations (Cassandra, 1998), e.g., because it is too expensive to collect information, or, if the business processes are partially performed in paper by humans and partially machine-based.
In the majority of the situations, the management competence should support their decisions with partial information about the surrounding environment, also named, in the literature, as partial observable environments or asymmetric information (De Giovanni, 2017) or information imperfections (Amor et al., 2017). In line with this limitation, (Frank, 2014) states that to make enterprise models a versatile tool for supporting professional action in organisations, new research challenges are posed in order to widen the scope of modelling by adding further topics that also comprise concepts to support managerial decision making. This foreknowledge drives this research effort.
Weinberg (2001) divides systems in three distinct types: organized simplicity, for self-contained problems that could be treated with analytical tools; unorganized complexity, for complex systems, but yet sufficiently random in their behavior so that they are sufficiently regular to be studied statistically; and organized complexity where all the remaining organizational-wide problems fall. Moreover, organized complexity does not fit analytical nor statistically solutions. The author state that new formal approaches dealing with systemic perspective are needed to develop the knowledge on how to address organized complexity problems. This assertion is the essence of this research: the integration of knowledge from information systems and operations research.
Within an organization, business processes play a dual role of: (i) being the result of applying design constraints for a particular organization reality (Hoogervorst, 2009), and are valid over a given period of time, and (ii) support for implementation of systems operating actions performed by organizational actors. However, the actors have an active and independent role in the execution of business processes. Therefore, it does not guarantee that the requirements of business processes are met properly on the daily routines. For example, if a company's recommendation is to always obtain a written record when contacts are made with clients, nothing limits the ability of an actor to contact a client directly, by phone, without leaving any trace of that communication. The same example can be applied to the contemporary financial markets, with a huge adverse impact to the organization and to its environment. Alter (2014) describes this phenomenon as workarounds found by the actors.
Accordingly, with this presented problem, the authors narrow the decision-making management field to the business processes operation optimization. This well-defined domain sets the research within the scope of unorganized complexity offering solutions for decision-making (Mezghiche et al., 2015; Quttineh et al., 2017; Öner-Közen & Minner, 2017). So forth, this paper proposes and evaluates an innovative approach combining the DEMO enterprise engineering (Dietz, 2006b) with a stochastic approach based in partial observable Markov decision process (POMDP) theory (Russell & Norvig, 2010).