A Model based Realisation of Actor Model to Conceptualise an Aid
for Complex Dynamic Decision-making
Souvik Barat
1
, Vinay Kulkarni
1
, Tony Clark
2
and Balbir Barn
3
1
Tata Consultancy Services Research, Pune, India
2
Sheffield Hallam University, Sheffield, U.K.
3
Middlesex University, London, U.K.
Keywords: Organisational Decision Making, Simulation, Actor Model of Computation, Actor based Simulation.
Abstract: Effective decision-making of modern organisation requires deep understanding of various aspects of
organisation such as its goals, structure, business-as-usual operational processes etc. The large size and
complex structure of organisations, socio-technical characteristics, and fast business dynamics make this
decision-making a challenging endeavour. The state-of-practice of decision-making that relies heavily on
human experts is often reported as ineffective, imprecise and lacking in agility. This paper evaluates a set of
candidate technologies and makes a case for using actor based simulation techniques as an aid for complex
dynamic decision-making. The approach is justified by enumeration of basic requirements of complex
dynamic decision-making and the conducting a suitability of analysis of state-of-the-art enterprise modelling
techniques. The research contributes a conceptual meta-model that represents necessary aspects of
organisation for complex dynamic decision-making together with a realisation in terms of a meta model that
extends Actor model of computation. The proposed approach is illustrated using a real life case study from
business process outsourcing industry.
1 INTRODUCTION
Modern organisations constantly attempt to meet
organisational goals by adopting appropriate courses
of action (Shapira, 2002). Evaluation of the possible
courses of action and selection of the best option are
key challenges faced by organisations. It calls for the
precise understanding of various aspects such as
goals, organisation structure, operational processes,
historic data and the stakeholders (Daft, 2012). The
socio-technical characteristics (McDermott et al.,
2013), inherent uncertainty and non-linear causality
in business interactions (Conrath, 1967), and high
business dynamics (Sipp, 2012) further exacerbate
the complex dynamic decision making (CDDM)
endeavour.
The industrial practice of organisational decision-
making heavily relies on human experts who typically
use tools such as spreadsheets, word processors, and
diagram editors. Though adequate for capturing and
collating the required information, these tools offer
limited analysis support if at all (Locke, 2011). As a
result, CDDM tends to be a time-, effort- and
intellectually-intensive endeavour. Furthermore,
reports from leading consulting organisations such as
McKinsey and Harvard Business Review (Kahneman
et al., 2011, Meissner et al. 2015) often classify the
current state of practice as biased, based on short-term
emotion and imprecise for modern business context.
This perceived poor state-of-practice of decision-
making elicits a research question: What kinds of
technological aids are needed for decision makers to
arrive at precise, unbiased and effective decisions?
This paper argues that the success of decision-
making largely depends on two key factors: (i) the
ability to capture relevant information about
organisation and (ii) the ability to perform what-if and
if-what analyses of relevant information in a relatable
form. The former ensures completeness of
information and the latter ensures reduction of
analysis burden on human experts.
A wide variety of Enterprise Modelling (EM)
techniques have been proposed to capture the relevant
information about organisation in a formal manner
amenable to rigorous analysis (Kulkarni et al.,
2015a). However, they are found to be less effective
and insufficient for a class of decision-making
problems characterised by significant dynamism,
Barat, S., Kulkarni, V., Clark, T. and Barn, B.
A Model based Realisation of Actor Model to Conceptualise an Aid for Complex Dynamic Decision-making.
DOI: 10.5220/0006216306050616
In Proceedings of the 5th International Conference on Model-Driven Engineering and Software Development (MODELSWARD 2017), pages 605-616
ISBN: 978-989-758-210-3
Copyright © 2017 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
605

inherent uncertainty, and emergent behaviour (Barat
et al., 2016b). Being a socio-technical system, an
organisation can be best viewed as a set of interacting
units each having own goals and operating with the
intent of achieving them. Thus, behaviour of the
entire organisation is not known a priori (and hence
never specified as such) but emerges through the
interactions of units each having well-defined
behaviour (Hewitt, 2010). Behaviour of an
organisation unit can be specified in terms of the
many actor languages and frameworks available e.g.,
Erlang (Armstrong, 1996), SALSA (Varela and
Agha, 2001), AmbientTalk (Van Cutsem et al., 2007),
and Kilim (Srinivasan and Mycroft, 2008),
ActorFoundry (Astley, 1998), Scala Actors (Haller
and Odersky, 2009), Akka (Allen, 2013) etc. Though
capable of catering to the specification of
autonomous, intentional, and emergent behaviour,
these languages do not provide support for
uncertainty and temporal behaviour.
In this paper, we present an approach that extends
the actor model of computation with uncertainty and
temporal behaviour to serve as an effective aid for
CDDM. In particular, this paper makes two
contributions: i) a conceptual meta-model that
represents necessary aspects of the organisation along
with the inherent characteristics of CDDM, and ii) a
realisation meta-model that concretises conceptual
model by extending the core concepts of actor model
of computation. Also, we illustrate the proposed
approach demonstrating its efficacy with a case study
from Business Process Outsourcing (BPO) domain.
We envision an overarching research agenda
1
for
developing a business facing decision-making
framework to improve precision of decision-making,
reduce personal biases while considering decisions,
consider short term and long term effects before
arriving at decisions, and reduce the excessive
analysis burden on human experts in decision-making
process. We argue the presented contributions form a
basis of such a business facing decision-making
framework.
The paper proceeds as follows: Section 2 presents
research motivation by highlighting necessary tenets
of CDDM and reporting brief overview of state-of-
the-art of EM techniques and actor
language/frameworks. Section 2 concludes by
highlighting notable gaps that limit the adoption of
EM techniques and actor languages/frameworks for
CDDM. Section 3 presents the conceptual meta-
model that has a potential to address CDDM
1
http://www.tcs.com/research/Pages/Model-Driven-Organiza
tion.aspx
problems. A meta-model that realises conceptual
model by extending actor model of computation with
relevant concepts such as uncertainty, temporal
behaviour is described in section 4. The illustration of
the proposed approach using a case study from BPO
is highlighted in section 5 and a brief evaluation of
proposed approach is presented in section 6. The
paper concludes with a brief summary research
progress and future plan to realise the overarching
research agenda.
Figure 1: Overview of decision making.
2 MOTIVATION
An abstract representation of decision-making is
presented in Fig. 1. As shown in the figure, an
Organisation interacts with its Environment to
achieve its Goals. The Goals are typically assessed by
evaluating the key performance indicators or
Measures. The decision makers evaluate/predict
Measures with respect to Goals and decide
appropriate courses of action or Levers. In this
formulation, the decision-making is finding best
possible Levers for stated Goals in the context of
Organisation and Environment. Essentially it is an
iterative and refinement based method to explore
available Levers, evaluate them with respect to Goals,
Organisation and Environment, and finally decide
most effective options. We argue the efficacy of such
exploration depends on two key factors, i.e., ability to
specify information about relevant aspects of the
Organisation and ability to analyse available
information in the context of the Environment where
it operates.
The management literature advocates multiple
methods such as Incremental method (Mintzberg et
al., 1976) and Carnegie Method (Cyert et al., 1963)
for guided exploration of Levers for stated Goals.
IndTrackMODELSWARD 2017 - MODELSWARD - Industrial Track
606

Table 1: Requirements of CDDM.
Requirement Description
Why Intention
What Structural Specification
How Behavioural Specification
Who Stakeholders and Human actors
Modular Must encapsulate internal goal,
structure and behaviour.
Compositional Multiple parts should be
composed to a consistent whole.
Reactive Must respond appropriately to its
environment
Autonomous Possible to produce output
without any external stimulus.
Intentional Intent defines the behaviour
Adaptable Adapt itself based on context and
situation
Uncertain Precise intention and behaviour
are not known a-priori.
Temporal Indefinite time-delay between an
action and its response
Machine
Interpretable
Models that are interpretable by
machine (i.e., support for
simulation/execution)
However, they do not prescribe or recommend
any technological aid best suited for their proposed
methods as their focus in not pertaining to any
technological aspects. We conducted a series of
literature reviews and experiments to understand -
What kinds of modelling abstractions and analysis
techniques are available for specifying and analysing
different aspects of an organisation? Are they
capable of supporting expected characteristics of
CDDM? What are the gaps?
Our experiments such as (Kulkarni et al., 2015b)
and literature review such as (Barat et al., 2016a)
indicate inadequacy of state-of-the-art of relevant
enterprise wide modelling and analysis techniques
and tools. An overview of our explorations is
presented in this section. We first describe the key
tenets of CDDM that we use for evaluating the state-
of-the-art specification and analysis techniques.
Subsequently we discuss the findings and illustrate
the gaps by examining the state-of-the-art
specification and analysis techniques.
2.1 Tenets of CDDM
We argue that an Organisation can be understood well
by analysing what an enterprise is, how it operates,
why it is so, and who are the responsible stakeholders
(Barat et al., 2016b). This hypothesis is principally
aligned with the Zachman framework (Zachman et
al., 1987) and industry prevalent enterprise modelling
(EM) techniques such as ArchiMate (Iacob et al.,
2012).
CDDM puts some special demands on
specification in terms of desirable characteristics of
organisation that include reactive, adaptable,
modular, autonomous, intentional, compositional,
uncertainty and temporal. Essentially these
characteristics represent associated dynamism to
interact with Environment. Furthermore, industry
practice of decision-making desires precise what-if
and if-what analysis for a-priori indication of a
decision. As a result, a machine interpretable model
forms the basis of analysis requirements. Table 1
enumerates specification and analysis requirements
for CDDM.
2.2 Exploration of Specification and
Analysis Techniques
In (Barat et al., 2016a) we evaluated the suitability of
EM techniques in the context of CDDM using
Systematic Mapping Study methodology (Petersen et
al., 2008). The evaluation concluded with a critical
observation that the existing EM techniques are
capable of satisfying the expected requirements of
CDDM described in Table 1 in parts. In particular, we
found the EM techniques that support necessary
aspects of CDDM (such as Zachman Framework and
ArchiMate) are not machine interpretable and thus
not amenable for rigorous analyses. In contrast,
specification approaches such as BPMN (OMG,
2011), i* (Yu, 2006) and Stock-n-Flow (SnF)
(Meadows and Wright, 2008) are capable of
sophisticated analyses and simulation. For example,
the process aspect can be analysed and simulated
using BPMN based tool, the high level goals and
objectives can be evaluated using i*, and high level
system dynamics can be simulated using Stock-and-
Flow (SnF) tools such as iThink. However, they are
not capable of representing all necessary aspects.
Detailed review synthesis led us to explore multi-
modelling and co-simulation environments involving
multiple EM techniques to address CDDM. The
exploration was conducted using two activities: a) a
literature review on multi-modelling and co-
simulation environments such as DEVS (Camus et
al., 2015), AA4MM (Siebert et al, 2010), AnyLogic
(Borshchev, 2013), and b) an experiment on multi-
modelling and co-simulation approach by combining
i*, Stock-and-Flow and BPMN tools. The research
finding, experimental setup and experiences are
presented in (Kulkarni et at., 2015b). Both the
literature review and experiment on multi-modelling
and co-simulation approach have produced evidence
that indicate a multi-modelling and co-simulation
As
p
ect
Socio-technical Characteristics
A Model based Realisation of Actor Model to Conceptualise an Aid for Complex Dynamic Decision-making
607

Figure 2: Meta-model for representing Actor Language/Framework (AMModel).
based approach using multiple EM techniques are
capable of representing necessary aspects and they
collectively support the analyses needs. However,
they are largely prone to intrinsic complexity (as
discussed in (Kulkarni et al., 2015c) and accidental
complexity (as discussed in (Kulkarni et al., 2015b).
Moreover, they are not capable of expressing many
socio-technical characteristics such as autonomy,
uncertainty, temporal behaviour and adaptability.
The inadequate support for socio-technical
characteristics in EM techniques (as in individual or
within multi-modelling setup) opens up a scope for
exploring the languages and frameworks that are
based on the actor model of computation. A literature
review on actor language and frameworks discloses
their suitability in the context of CDDM. Essentially
actor languages and frameworks are capable of
specifying and analysing a range of socio-technical
characteristics.
The key concepts and core capabilities of
prevalent actor languages and frameworks (such as
Scala Actor and Akka) are represented using a meta-
model (termed as AMModel
2
) as depicted in Fig. 2.
The concept Actor, a named entity, encapsulates
State, Behaviour and internal Actors. A State can be
specified using attributes and the Behaviour is
defined using set of Event specifications. The
behaviouralSpec of Behaviour influences State of an
Actor and is capable of representing four kinds of
behaviour namely ReactiveBehaviour,
AutonomousBehaviour, AdaptableBehaviour and
EmergentBehaviour. The ReactiveBehaviour reacts
by responding to Events, AutonomousBehaviour
triggers internal Events, AdaptableBehaviour
describes the adaptability of an Actor using set of
rules and EmergentBehaviour specifies the emergent
2
Meta-models are drawn using xModeler tool
(http://www.eis.mdx.ac.uk/staffpages/tonyclark/Software/XM
odeler.html)
behaviour of an actor model (Agha, 1985).
The proposed AMModel is capable of
representing the what aspect using the structure and
composition, how aspect using behavioural
specification and the who aspects using the Actor
itself. It is also capable of representing characteristics
such as modular using the notion of Actor,
compositional using contents association (See Fig. 2),
reactive using ReactiveBehaviour, adaptable using
AdaptableBehavior and EmergentBehaviour and
autonomous using AutonomousBehaviour. However,
it is not capable of describing the why aspect and other
characteristics such as intentional, uncertainty and
temporal behaviour.
In the next section, we describe the necessary
concepts of CDDM that satisfy the tenets described in
table 1 in terms of a meta-model. Then we discuss an
extended form of actor model that realises this meta-
model.
3 CONCEPTUAL MODEL
We define a meta-model to represent the relevant
aspects of an organisation along with the
characteristics described in Table 1. The proposed
meta-model, termed as CMModel, is depicted in Fig.
3. In the figure, an OrgUnit represents an
Organisation which is an autonomous self-contained
functional unit having high coherence and low
external coupling. It has a set of Goals that represent
its intention, Measures that describe the key
performance indicators of OrgUnit, and Levers that
represent possible courses of action with a potential
to change an OrgUnit in terms of structure, behaviour
and/or intentions. The goalSpec of Goal element uses
Measure (and thus historical and current Data of an
IndTrackMODELSWARD 2017 - MODELSWARD - Industrial Track
608

Figure 3: Meta-model for describing Organisation in the context of CDDM (CMModel).
Figure 4: Extended Actor meta-model (EAMModel).
OrgUnit). The Measures are expression over Data and
the leverSpec of Lever element describes the change
specification or defines configuration parameters. An
OrgUnit interacts with environment through a set of
Events.
Internally an OrgUnit contains Data, Behaviour,
Structure and Participants. Data represents the
current and historical States of OrgUnit, i.e. current
state and traces. The Structure of an OrgUnit is
described using multiple contained OrgUnits and
Participants. The contained units can interact with
each other to delegate responsibilities or can
participate in a hierarchical composition structure to
accomplish higher level goals. The Participant, a
specialised OrgUnit, represents the resources of the
OrgUnit. Proposed meta-model advocates five kinds
of Behaviour namely ReactiveBehaviour,
AutonomousBehaviour, StochasticBehaviour,
TemporalBehaviour and AdaptiveBehaviour. The
ReactiveBehaviour represents external interactions
using (external) Events and AutonomousBehaviour
represents the internal behaviour using (internal)
Events. The behaviouralSpec of StochasticBehaviour
describes associated uncertainty of raising an Events
and responding to an Event, and the behaviouralSpec
of TemporalBehaviour describes the temporality of
Event specification. The behaviouralSpec of
AdaptableBehaviour describes adaptation rules. We
introduce a concept ‘Time’ to represent time aspect
of an OrgUnit. There could be a central ‘Time’
element for an Organisation or each OrgUnit may
own a ‘Time’ element.
Conceptually, the elements OrgUnit, Event, Data,
and nesting capability of OrgUnit specifies the what
aspect, Goal specifies the why aspect, Behaviour
specifies the how aspect and Participant specifies the
who aspect of an Organisation or OrgUnit. Event
helps to capture reactive nature, the intent is captured
using Goal, the modularity is achieved through the
concept of OrgUnit, autonomy is possible due to the
concept of AutonomousBehaviour, Events and Time,
and composition can be specified using nesting
relation. Also, OrgUnit is adaptable as it can construct
and reconstruct its structure using
AdaptableBehaviour; modular as it encapsulates
Data, Structure and Behaviour; intentional as it has its
own goals; and compositional as it can be an
assembly of OrgUnits. The StochasticBehaviour and
A Model based Realisation of Actor Model to Conceptualise an Aid for Complex Dynamic Decision-making
609

Figure 5: Overview of Business Process Outsourcing Scenario. Figure 6: Interactions and Behaviours.
TemporalBehaviour are capable of representing
associated uncertainty and temporality.
Thus we argue this meta-model conceptually
covers all the specification needs stated in table 1 and
a machine interpretable specification realising this
meta-model can serve the analysis need.
The proposed meta-model is grounded with a set
of existing concepts. The modularisation and reactive
unit hierarchy are taken from component model
concepts. Goal-directed reactive and autonomous
behaviour can be traced to actor behaviour (Agha,
1985). Defining states in terms of a type model is
borrowed from UML. An event driven architecture is
introduced for reactive behaviour and the concept of
intentional modelling (Yu, 2006) is adopted to enable
specification of goals.
4 REALISATION MODEL
In this section we propose extensions to the actor
meta-model (AMModel) of Fig. 2 to realise the
proposed conceptual model (CMModel) of Fig. 3.
The extensions are presented using a meta-model
(termed as EAMModel) in Fig. 4. As shown in the
figure, the concept of Actor (described in AMModel)
is primarily extended with the concepts of Goal,
Measure, Lever and Time. The extended Actor is
represented as ExtendedActor in EAMModel. The
concepts Goal, Measure, Lever and Time associated
with ExtendedActor of EAMModel conforms to
Goal, Measure, Lever and Time concepts introduced
in CMModel.
The Behaviour of an Actor in AMModel is
extended with two additional behavioural types
namely StochasticBehaviour and
TemporalBehaviour wherein the
StochasticBehaviour and TemporalBehaviour of
EAMModel conform to the definitions of
StochasticBehaviour and TemporalBehaviour of
CMModel respectively.
We argue, the extended actor model (i.e., unified
version of AMModel and EAMModel) is capable of
realising the conceptual model represented in
CMModel. The concept ExtendedActor of AMModel
is capable of representing OrgUnit, Organisation and
Participants as the concept ExtendedActor is an
encapsulated, modular, autonomous, composable
entity.
Actor of AMModel (and thus ExtendedActor of
EAMModel) is capable of representing its current
states using attributes of State entity. ExtendedActor
definition is further capable of representing traces
using historicalTraces association to State entity.
Thus the unified meta-model (of AMModel and
EMModel) is capable of representing Data of
CMModel.
Similarly, the Actor described in AMModel is
capable of representing reactive, autonomous and
adaptable behaviour using ReactiveBehaviour,
AutonomousBehaviour and AdaptableBehaviour.
ExtendedActor in EAMModel further introduces the
StochasticBehaviour and TemporalBehaviour. Thus
we argue they collectively represent all necessary
behavioural types described in CMModel.
Finally, the extensions Goal, Measure, Lever and
Time to the conventional actor meta-model help in
realising conceptual model that is necessary for
CDDM.
We have conceptualised a language termed as
Enterprise Simulation Language (ESL) that
implements the concepts of conventional actor model
as depicted in Fig. 2 along with the extension
proposed in Fig. 4. We have developed a prototype
IndTrackMODELSWARD 2017 - MODELSWARD - Industrial Track
610

Figure 7: Realisation of BPO scenario using extended actor meta-model.
implementation of ESL using DrRacket
3
. We have
also developed a prototypical simulation engine to
iterate over the “apply Lever – Observe Measure –
Analyse the feasibility of Goals” loop as depicted in
Fig. 1. The simulation machinery comprises of
Spreadsheet, DrRacket based ESL execution engine
and Python wherein Spreadsheet is used for
specifying Lever configuration, ESL engine for
simulation, and Python for visualising Measures and
Goals.
We illustrate the proposed realisation model and
simulation capability of its implementation using an
industrial case from BPO domain. The next section
presents the case study and observed results.
5 ILLUSTRATION
Consider the business process outsourcing (BPO)
domain. Customers outsource business processes for
a variety of reasons such as reducing Cost (C),
increasing Efficiency (E), bringing about a major
transformation, i.e., Delight (D). The outsourced
processes can be classified into three buckets based
on maturity of BPO verticals. For instance, Transcript
Entry process of Healthcare vertical is one of the early
adopter of BPO and has derived almost all potential
benefits accruable from outsourcing (termed as
Sunset or SS). On the other hand, IT Infrastructure
3
https://racket-lang.org
Management process being a late adopter of BPO has
a large unrealized potential to be tapped (termed as
Sunrise or SR). And there are processes such as Help
Desk, Account Opening, Monthly Alerts etc., that fall
somewhere in between the two extremes as regards
benefits accrued from BPO (i.e. Steady or ST). Thus,
BPO demand space can be viewed as set of customers
where each customer is characterised by one of the
type described using a 3 x 3 matrix of Fig 5.
A customer invites bids from the vendors for a
specific BPO outsourcing project. Typically, factors
such as Quadrant (i.e. ranking as per independent
agency such as analysts), FTE Count Range (i.e. full
time employees to be deployed on the outsourced
process), Billing Rate Range (i.e. per hour rate of full
time employee), Organisation Size(the number of
employee) and Track Record (i.e., familiarity with the
processes being outsourced), influence who wins the
bid. Other soft issues such as Market Influence (i.e.
perception of the market as regards delivery certainty
with acceptable quality), the rapport with the vendor
etc., also play a part in bid evaluation. In addition to
these known factors there could be some uncertainty
in bid evaluation criteria (in other words, bid
evaluation criteria can’t be fully known a-priori).
It is common observation that BPO outsourcing
projects come up for renewal after few years
(typically 3 to 5 years). Customer may renew the
contract with the existing vendor on modified terms
A Model based Realisation of Actor Model to Conceptualise an Aid for Complex Dynamic Decision-making
611

(typically advantageous to the customer) or may opt
for rebidding. Factors influencing the renewal
decision are reduction offered in FTE Count, Billing
Rate, number and degree of escalations, perception
that the external agent has as regards ability to meet
the project requirements, inherence uncertainty, etc.
Contracts that fail to get renewed become candidates
for later bidding. Fig 5 shows a high level schematic
of BPO space. The events of interest illustrating the
interactions between customer and vendor, and the
state transition of outsourcing project are depicted in
Fig 6.
The demand space exhibits temporal dynamism.
For instance, new processes emerge as candidates for
outsourcing and some of the existing processes no
longer need to be outsourced as, say, technology
advance eliminates the need for human intervention
in the process thus making it straight-through. Thus
the BPO space can be viewed as an event-driven
system where events have a certain frequency and are
stochastic in nature. The frequency and stochastic
characteristics typically vary from process to process.
While operating in this uncertain space, a BPO
vendor needs to make decisions of the following kind:
Will continuation with the current strategy keep me
viable ‘n’ years hence? What alternative strategies are
available? How effective will a given strategy be? By
when will a given strategy start showing positive
impact? Will I be growing at the expense of
competition or vice versa? and so on.
Answers to the above questions are essentially
linked to the evaluation of portfolio basket i.e., 3 x 3
matrix of Fig 5, of the organisation in terms of
revenue accruable and expenses in terms of FTEs.
The ability to predict the portfolio basket of the
organisation and its competitors after a given time
period becomes critical to support decision-making.
5.1 Realisation using Actor Model
We model BPO scenario using our extended actor
model (i.e., the unified meta-model of AMModel
depicted in Fig. 2 and the EAMModel depicted in Fig.
4). The key elements demand, vendor and outsourced
projects are modelled as ExtendedActor as shown in
Fig. 7. The demand ExtendedActor comprises of nine
attributes where each attribute represents a bag of
outsourced projects of specific type from the demand
classification i.e., {SR, ST, SS} X {C, E, D}. Each
outsourced project ExtendedActor represents a
specific demand classification using its State
attributes and implements the state-machine depicted
in Fig. 6. The increase (or decrease) of specific type
of outsourced project in demand space is specified by
instantiating (or destructing) specific outsourced
project ExtendedActor.
Vendors ExtendedActors has a set of State
attributes to represent portfolio baskets (i.e., flattened
out 3 x 3 matrix), Resources, Traces and other
attributes such as Revenue. The portfolio basket
represents the bags of outsourced project
ExtendedActors. The State attribute Resource
contains a bag of ExtendedActors that represent the
FTEs and the State attribute Traces is historical data
about the values of State attributes of vendor such as
Revenue at specific time. The characteristics of a
vendor such as Quadrant, Billing Rate, FTE count,
Market Influence and Delivery Excellence are also
represented using State attribute of vendor
ExtendedActor. In this formulation, one vendor is
marked as ‘We’ and rest are classified as competitors.
This example considers two competitors –
Competitor 1 and Competitor 2 as depicted in Fig. 7.
The table in Fig 7 shows the initial characteristics
of ‘We’ ExtendedActor. We make these
characteristics configurable to attenuate their values,
thus these State attributes also act as Lever in this
example. As shown in the figure, a vendor is equipped
with a set of negotiation levers namely, the range of
Billing Rate (employees billed against the outsourced
process), range of FTE Productivity (percent
reduction possible in number of full time employees),
range of FTE Reduction (reduction possible during
renewal of a contract), range of Billing Rate
Reduction (reduction possible in billing rate during
renewal of a contract), Influence Relation
and
Delivery Excellence. The Influence Relation is a
qualitative characteristic that is quantified using four
weighted labels namely ‘Excellent’, ‘Good’,
‘Normal’ and ‘Not Good’. Value of Delivery
Excellence attribute is a probability distribution. For
instance, ‘We’ ExtendedActor is confident of
delivering ‘Excellent’ quality on 60% of Cost kind of
BPO projects won. The values for ‘Good’, ‘Normal’
and ‘Below Normal’ quality for this kind of BPO
projects are 30%, 10% and 0% respectively.
There are many Measures that are of interest to
the stakeholders of ‘We’ vendor. Fig. 7 depicts three
key Measures namely Revenue (i.e., the State
attribute Revenue), Customer Count (i.e., total
number of outsourced projects contain by ‘We’
vendor) and Realisation (i.e., average Revenue earned
by each Resource per hour) for illustration purpose.
One can model different kinds of vendor by
setting appropriate values to the initial setting. The
‘Competitor’ ExtendedActors are modelled on the
same lines as ‘We’ ExtendedActor.
IndTrackMODELSWARD 2017 - MODELSWARD - Industrial Track
612

Figure 8: Simulation Results 1.
This example considers a central Time to synchronise
the entire business-as-usual (i.e., Behaviour). Time
triggers two events namely ‘Month’ and “Year” to the
demand and vendor ExtendedActors. Demand
ExtendedActor raises RFP events for outsourcing
project. Each RFP event is characterized by the kind
of process being outsourced (i.e. SR or ST or SS), the
objective for outsourcing (i.e. C or E or D), size of the
process in terms of FTE count, and the desired billing
rate. Interested vendors respond to the RFP event by
picking suitable values from their characteristics at
random. Bid evaluation function is a weighted
aggregate of the various elements of RFP response
and a random value to capture effect of inherent
uncertainty. The vendor with the best evaluated value
wins the outsourcing project which gets executed as
defined by the characteristics of the particular vendor.
Essentially, an outsourcing project ExtendedActor
moves from Demand ExtendedActor to a vendor
ExtendedActor (i.e., from demand basket to vendor
portfolio basket) as shown in Fig. 7. The existence of
an outsourcing project within a vendor impact
vendor’s State attribute (and thus Measures) as
outsourcing project consumes the resources and
contribute the revenue, the customer count and other
measures. It also impacts the track record and market
influences over the time.
The decision to renew existing contract is
modelled on similar lines but with a different set of
characteristic attributes influencing the decision.
Essentially the autonomous outsourcing project
ExtendedActor raises Renew event after 3 to 5 years
timeframe. Here too, the evaluation is cognizance of
incomplete and uncertain knowledge renewability
criteria.
5.2 Simulation and Results
We use the simulation environment developed using
DrRacket and Python to run the system for 10 years.
In the interest of space, two decision questions from
many scenario playing are discussed in this paper.
Decision questions are – i) Will continuation with the
current strategy keep me viable ‘n’ years hence with
respect to the competition? And ii) What will be my
position if we decide to change my characteristics?
An overview results of simulation run is shown in
Fig 8. As can be seen, the current revenue of ‘We’
(represented using shades of ‘blue’ disks) is 438.98
MUSD from 90 customers with a realization of nearly
15.5 USD per hour per FTE. Corresponding numbers
for competitor 1 and competitor 2 respectively are <
319.97, 78, 13.33 > (depicted using shades of ‘violet’
disks) and < 352.32, 79, 15.1 > (depicted using shades
of brown disks). In short, at present ‘We’ vendor is
doing much better than competition.
‘We’ vendor set a goal to deliver <750, 200, 17>
after 5 years and <1000, 290, 18> after 10 years
(depicted using green disk). As can be seen, by
continuing to operate the same way the ‘We’ vendor
will be delivering <587.58, 160, 13.5> after 5 years
and <857.51, 215, 14> after 10 years (as directed by
red line in Fig. 10) thus missing both the targets by a
considerable margin. More importantly, competitor 2
will be overtaking ‘We’ vendor after 5 years and both
the competitors will be significantly ahead of ‘We’
vendor after 10 years.
Clearly, ‘We’ vendor cannot afford to continue
with its current way of operation. Further detailed
analysis, involving model elements not described in
A Model based Realisation of Actor Model to Conceptualise an Aid for Complex Dynamic Decision-making
613

Figure 9: Lever Specification and Simulation Results 2.
this paper for want of space, shows that much of
current revenue of ‘We’ vendor is from sunset kinds
of outsourced processes for cost reasons. Over time
this market is going to shrink considerably with
demand for steady as well as sunrise processes (for
objectives other than pure cost reduction) increasing
significantly. Thus ‘We’ vendor needs to bring about
a change in its characteristics so as to be able to win
more bids in this demand situation. Fig 9 shows the
modified characteristics of ‘We’ vendor leading to the
improved performance as shown in Fig 9. With the
changed parameters, the ‘We’ vendor is able to beat
both revenue and customer targets while failing to
meet the realization target narrowly.
6 EVALUATION
For the kind of decision-making problem illustrated
in this paper, industry practice relies extensively on
Excel spreadsheets. Such an approach typically
represents the influence of lever onto measures in
terms of static algebraic equations. However, value of
a lever and influence of a lever onto a set of measures
can vary over time. This behaviour cannot be
captured using excel sheet. There is no support for
encoding stochastic behaviour either.
Stock-and-Flow models are also used for a class
of decision-making. In this approach, the system
behaviour is expected to be known a priori.
Essentially the system is specified in terms of stocks,
flows of stocks, and a fixed set of equations over
system variables that control the flows. Value of a
stock or a flow or a variable is a discrete number or a
range or a distribution. The quantitative nature of
Stock-and-Flow models and sophisticated simulation
support enables decision-making through what-if
scenario playing. It is possible for a stock or an
individual variable to have a value that is a probability
distribution, however, the structure of the stock-n-
flow model must remain unchanged. Thus, systems
dynamics modelling provides only a partial support
for specifying and processing the inherent uncertainty
within a system. Moreover, it is best suited for an
aggregated and generalized view of a system where
individual details get eliminated through averaging,
and sequences of events are grouped as continuous
flows. This generalized approach and ignorance of
individual characteristics that significantly influence
the system over time often leads to a model that is
somewhat removed from reality. Though not
designed to specify specialized behaviour, it can be
done using systems dynamics modelling. But this is
an effort-intensive endeavour, and more importantly
leads to model size explosion (Kulkarni et al., 2015b).
For example, modelling of 4 competitors each having
special characteristics leads to roughly 4 times
increase in the size of systems dynamics model.
The proposed approach enables modelling of a
system as a set of units each
listening/responding/raising events of interest and
interacting with other units by sending messages. A
IndTrackMODELSWARD 2017 - MODELSWARD - Industrial Track
614

unit encapsulates state (i.e. a set of State attributes),
trace (i.e. events it has responded to and raised till
now) and behaviour (i.e. encoding of reactions). As
the modelling abstraction supports ‘time’ concept,
value of a variable and relationships between
variables can change with respect to time. Consider
the example of determining the impact of track record
on bid win of organisation where the value of track-
record variable changes over time thus affecting bid
win factor. Since a process is an individual actor,
simulator can determine the impact of successful
contract completion, renewal with/without
negotiation etc., for that specific process – systems
dynamics model falls short here. A trace of events
serves as a memory that can be queried to establish
more complex relationships between levers. For
example, successful completion of contract leads to
improved track record as well as better rapport with
the customer thus improving the bid win factor of
future outsourcing bids everything else remaining the
same. Thus, the abstraction provides primitives for
creating models that closely mimic reality.
7 CONCLUSIONS
Effective decision-making is a challenge that all
modern enterprises face. It requires deep
understanding of aspects such as organisational goals,
structure, operational processes. Large size, socio-
technical characteristics, and increasingly fast
business dynamics make this activity much more
difficult task for decision makers. Inadequate support
for representing necessary aspects of an organisation
in a relatable form and inability to handle inherent
uncertainty and temporal characteristics are the
present lacuna in state-of-the-art technological aids
that are used in decision-making.
This paper shows the gaps by evaluating
technological aids with respect to the needs of
complex dynamic decision-making. We began by
outlining a conceptual model (i.e. CMModel) that has
potential to mitigate the identified gaps between the
available technical capabilities and expected
characteristics. We then argued that an extended form
of actor model (i.e., AMModel and EAMModel) can
address these needs. We validated the hypothesis
through an industry scale case study from BPO
domain. We have shown how the case study can be
modelled in terms of the proposed realisation model
that is an extension of actor model of computation for
complex dynamic decision making. We have shown
how simulation of this model helped in identifying
the most appropriate of the available alternatives at
each decision point. Thus, it can be said that the
proposed approach can be used to define purpose-
specific strategy and/or evaluate the most appropriate
from a set of candidate strategies.
We acknowledge this paper does not discuss the
language constructs of ESL, but, principal objectives
of paper were: establish the core concepts of CDDM,
correlate the core concepts with actor model of
computation, and propose the necessary extensions to
actor model for supporting complex dynamic
decision-making.
Our next step is to use the proposed extended
actor meta-model and its implementation in the form
of ESL for developing a business-facing decision-
making framework that will improve the precision of
decision-making, reduce personal biases while
considering decisions, consider short term and long
term effects before arriving at decisions, and reduce
the excessive analysis burden on human experts in
decision-making process.
In addition, further exploration of behavioural
adaptability, understanding of emergent behaviour in
an organisation, and the introduction of game
theoretic approach in a simulation are part of our
research agenda.
REFERENCES
Agha, G.A., 1985. Actors: A model of concurrent
computation in distributed systems. Tech. rep., DTIC
Document.
Allen, J., 2013. Effective akka. O'Reilly Media, Inc.
Armstrong, J., 1996. Erlang - a survey of the language and
its industrial applications. In: Proc. INAP. vol. 96.
Astley, M., 1998. The actor foundry: A java-based actor
programming environment. University of Illinois at
Urbana-Champaign: Open Systems Laboratory.
Barat, S., Kulkarni, V., Clark, T., Barn, B., 2016a:
Enterprise Modeling as a Decision Making Aid: A
Systematic Mapping Study. PoEM 2016: 289-298.
Barat, S., Kulkarni, V., Clark, T., Barn, B., 2016b: A
Simulation-based Aid for Organisational Decision-
making. ICSOFT-PT 2016: 109-116.
Borshchev, A., 2013. The big book of simulation modeling:
multimethod modeling with AnyLogic 6. AnyLogic
North America Chicago.
Camus, B., Bourjot, C., Chevrier, V., 2015. Combining
devs with multi-agent concepts to design and simulate
multi-models of complex systems. In: Proceedings of
the Symposium on Theory of Modeling & Simulation:
DEVS Integrative M&S Symposium. pp. 85-90.
Conrath, D.W., 1967. Organizational decision making
behavior under varying conditions of uncertainty.
Management Science 13(8), B-487.
Cyert, R.M., March, J.G., et al., 1963. A behavioral theory
of the firm. Englewood Cliffs, NJ 2).
A Model based Realisation of Actor Model to Conceptualise an Aid for Complex Dynamic Decision-making
615

Daft, R., 2012. Organization theory and design. Nelson
Education.
Haller, P., Odersky, M., 2009. Scala actors: Unifying
thread-based and event-based programming.
Theoretical Computer Science 410(2), 202-220.
Hewitt, C., 2010. Actor model of computation: scalable
robust information systems. arXiv:1008.1459.
Iacob, M., Jonkers, D.H., Lankhorst, M., Proper, E.,
Quartel, D.D., 2012. Archimate 2.0 specification: The
open group, Van Haren Publishing.
Kahneman, D., Lovallo, D., Sibony, O., 2011. Before you
make that big decision. Harvard business review 89(6).
Kulkarni, V., Barat, S., Clark, T., Barn, B., 2015a: A Wide-
Spectrum Approach to Modelling and Analysis of
Organisation for Machine-Assisted Decision-Making.
EOMAS@CAiSE 2015: 87-101.
Kulkarni, V., Barat, S., Clark, T., Barn, B., 2015b: Toward
overcoming accidental complexity in organisational
decision-making. MoDELS 2015: 368-377.
Kulkarni, V., Barat, S., Clark, T., Barn, B., 2015c: Using
simulation to address intrinsic complexity in multi-
modelling of enterprises for decision making.
SummerSim 2015: 9:1-9:11.
Locke, E., 2011. Handbook of principles of organizational
behavior: Indispensable knowledge for evidence-based
management. John Wiley & Sons.
McDermott, T., Rouse, W., Goodman, S., Loper, M., 2013.
Multi-level modeling of complex socio-technical
systems. Procedia Computer Science 16, 1132-1141.
Meadows, D.H., Wright, D., 2008. Thinking in systems: A
primer. Chelsea Green Publishing.
Meissner, P., Sibony, O., Wulf, T., 2016. Are you ready to
decide? McKinsey Quarterly, April. 2016.
Mintzberg, H., Raisinghani, D., Theoret, A., 1976. The
structure of unstructured decision processes.
Administrative science quarterly pp. 246-275.
OMG., 2011. Business Process Model and Notation
(BPMN), Version 2.0 (January 2011).
Petersen, K., Feldt, R., Mujtaba, S., Mattsson, M., 2008.
Systematic mapping studies in software engineering.
In: 12th international conference on evaluation and
assessment in software engineering. vol. 17, pp. 1-10.
Shapira, Z., 2002. Organizational decision making.
Cambridge University Press.
Siebert, J., Ciarletta, L., Chevrier, V., 2010. Agents and
artefacts for multiple models co-evolution: building
complex system simulation as a set of interacting
models. In: 9th International Conference on
Autonomous Agents and Multiagent Systems: 509-516.
Sipp, C.M., Elias, C., 2012. Real Options and Strategic
Technology Venturing: A New Paradigm in Decision
Making, vol. 31. Springer Science & Business Media.
Srinivasan, S., Mycroft, A., 2008. Kilim: Isolation-typed
actors for java. In: European Conference on Object-
Oriented Programming. pp. 104-128.
Varela, C., Agha, G., 2001 Programming dynamically
reconfigurable open systems with salsa. ACM
SIGPLAN Notices 36(12), 20-34.
Van Cutsem, T., Mostinckx, S., Boix, E.G., Dedecker, J.,
De Meuter, W., 2007. Ambienttalk: object-oriented
event-driven programming in mobile ad hoc networks.
In: Chilean Society of Computer Science, XXVI
International Conference of the. pp. 3-12.
Yu, E., Strohmaier, M., Deng, X., 2006. Exploring
intentional modeling and analysis for enterprise
architecture. 10th IEEE International Enterprise
Distributed Object Computing Conference Workshops.
Zachman, J., et al., 1987. A framework for information
systems architecture. IBM systems journal 26(3), 276-
292.
IndTrackMODELSWARD 2017 - MODELSWARD - Industrial Track
616