Get documents about "Modeling a business ecosystem An agent-based simulation
Document Sample
ECCON 2005 Annual meeting
Modeling a business ecosystem
Modeling a business ecosystem: An agent-based simulation
Erik den Hartigh, Michiel Tol, Jie Wei, Wouter Visscher & Ming Zhao1
Delft University of Technology
Abstract
A business ecosystem is a network of suppliers and customers around a core technology, who depend on each
other for their success and survival. In this paper we present a simple agent-based model of a business
ecosystem. This includes a conceptualization of the model, a specification of the model and a small series of
experiments to investigate the influence of network structure on the diffusion of innovations. Based on those
experiments, propositions are derived that will be used in further research to conceptualize, specify and test a
more advanced version of the model.
1 Introduction
1.1 What is a business ecosystem?
A business ecosystem is a network of suppliers and customers around a core technology platform, who depend
on each other for their success and survival (Den Hartigh & Van Asseldonk, 2004). As a consequence of the
growing importance of technology networks, it is almost impossible for firms to engage the competitive battle on
their own. We therefore see patterns of competition emerge that do not match the economic models of perfect
competition or even of oligopolistic or monopolistic competition. Rather, competition takes place between a few
large coalitions, or networks, of firms around a common technological platform. Such networks, consisting of
multiple firms performing different roles, are not unlike biological ecosystems. For such networks therefore the
term ‘business ecosystems’ is increasingly used (Moore, 1996; Iansiti & Levien, 2004a, 2004b).
1.2 Overall goal of the research project
The generic research framework of the project of which this paper is a part, is a dynamic variant of the Structure-
Conduct-Performance paradigm (see figure 1).
Market structure
Market dynamics Firm
Market conditions Firm conduct
• pattern of performance
• interdependency • strategy
innovation • dominant
of decisions • network
diffusion (speed, technology
• network structure governance
extent, stability) • profitability
Figure 1: Generic research framework
1
Contact information: Erik den Hartigh, assistant professor, Delft University of Technology, Faculty of
Technology, Policy and Management, section Technology, Strategy and Entrepreneurship, Office c2.170, P.O.
Box 5015, 2600 GA Delft, The Netherlands, phone +31 15 278 3565, email e.denhartigh@tbm.tudelft.nl
1/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
For this research project there are three central research questions. The first research question is “What is the
influence of network structure on the diffusion of innovations?” To investigate this relation, the market structure
component of the model is split in market conditions and market dynamics. The market conditions reflect the art
and degree of the interdependency of decisions of the agents in the market as well as the structure of the network
of relations between them. The market dynamics reflect the pattern of innovation diffusion. The second research
question is “What are the consequences for business strategy and business performance?” and the third research
question is “How can a firm govern networked structure in such a way that the diffusion of innovations
optimally serves the strategic interests of the firm?” To address these three research questions, an agent-based
simulation model will be developed that represents a real-world business ecosystem as adequately as possible.
1.3 Contribution of this paper
In this paper we only address the first research question. We therefore develop a simple agent-based simulation
model to investigate the relation between market conditions, i.e., network structure and interdependency of
decisions, and market dynamics, i.e., the pattern of innovation diffusion. Our aim with this paper is to develop
propositions regarding the first research question that together with expert validations will be used in future
research to conceptualize, specify and test an advanced version of the simulation model.
1.4 Approach and method
Simulation is the approach we take to investigate the research questions stated in this project. By carefully
‘mimic’ the characteristics and dynamics of the system, simulation models can provide the means to solve
complex business problems. According to Shannon (1975) simulation is “the process of designing a model of a
concrete system and conducting experiments with this mode in order to understand the behavior of a concrete
system and / or to evaluate various strategies for the operation of the system.” And: “It also allows for a quick
implementation of adjustments in the modeled system, making it possible to analyze different alternative
solutions in a relatively short time. It makes possible the study, analysis and evaluation of situations that would
not be otherwise possible.”
An agent-based simulation model is built here to gain insight into a hypothetical business ecosystem. This model
will also act as a ‘proof of concept’ that simulation is the right approach to answer the research question. To
complete a simulation model that conforms as good as possible to reality, and with which we can address the
three research questions of our project, a lot more work needs to be done in extending an validating agent
decision rules and in obtaining real-world data to specify such a model.
2/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
Figure 2: Major paradigms in simulation modeling on abstraction level scale (Borschev et al., 2004)
Figure 2 shows the major paradigms in simulation and its application area in the abstraction level scales.
Dynamic System is used to model and to design “physical” systems. It is at the lowest abstraction level. System
Dynamics is mostly used on the highest abstraction level, dealing with aggregate values, global feedbacks or
trends (Borschev et al., 2004). While Agent-based modeling is used at all level of abstraction, because agent can
model “physical” objects at the lowest level, such as vehicles, pedestrians, or customer, warehouses, retailers at
the middle level, or as large as companies, organizations, societies at the highest level (Jennings, 2003).
For the overall research project, the object of study is a real-world business ecosystem. We need to model the
companies participating in this system, relations among companies, decision rules of companies and actions of
key companies in the system in order to obtain global view of system feedback. The extremely complex nature
of such a business ecosystem makes it very hard to define any dynamics on the system level; we have absolutely
lack of knowledge for that. But the experts within the companies do know about their own activities, and know
how an action they take can influence an individual partner or customer. We can make the best use of their
working knowledge, if we start with abstracting behavior of individual companies, and defining
relationships/interactions among them. Moreover, by modeling the individual companies as agents, we can
observe the emergent behavior taking place in the system.
Due to the above reasons, agent-based modeling is, at least from a theoretical perspective, considered to be an
appropriate method to address the research questions of our generic research model.
2 Conceptualization of the simple model
2.1 Aspects of the business ecosystem
The concept of a business ecosystem model includes: the agents (players) that make up the system, the decision
rules of these agents, the categorization of these agents into ‘species’, the technology platforms that form the
core of the ecosystem and the network relations between the agents. Each of these aspects should be explicitly
3/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
modeled at the micro level. Additionally, we would like to conceptualize at the macro level the model inputs and
outcomes. We discuss these aspects below.
2.2 Agents
In our simulation application, the agents are companies in the real world. They possess characteristics such as
size, risk aversity and company goals. These characteristics are agent-specific. Agents belong to a species and
agents have to recognize events in their environment, calculate implications of these events and take actions
upon these events. For this they have decision rules, the shape of which is dependent on the species the agent
belongs to.
2.3 Decision rules
Each agent has a set of decision rules. We assume that agents’ behavior can be described by rules with which
they decide whether or not they will adopt a product or a technology. In these rules, the positive or negative
network externalities they experience will have to be incorporated. This means that an agent will look at other
agents’ status in making its own decisions. Local and global network effects should therefore be included as two
important factors influencing an agent’s decision. Besides, we assume that an agent has its own subjective
valuation of a technology platform, the intrinsic value.
2.4 Species
We can categorize agents into clusters, within which they share some common characteristics. Like a biological
ecosystem a business ecosystem will be populated by a diversity of ‘species’, each performing their own unique
functions, having their own unique needs and wants and each delivering a unique contribution to the survival and
growth of the business ecosystem as a whole. Some examples provided by Iansiti & Levien (2004, p.71)
regarding Microsoft’s business ecosystem are system integrators, development service companies, independent
software vendors, trainers, small specialty firms, internet service providers business consultants, media stores,
applications integrators ad many others. In other words, all firms that provide products (goods or services) or
technologies that are complementary and compatible to Microsoft’s core software technology platform.
A conceptual problem here is which level of aggregation to choose for categorizing agents into species. In our
research, we made a first rough inventory to arrive at 148 different species. This number can be further split out
or further clustered at will. Such an ‘at random’ way of clustering is not a good basis for an abstract simulation
tool. After many discussions with experts, we decided on clustering agents based on the way they earn their
money. In this way, we arrived at three basic species:
• those that earn money with the production and sales of products; these products can be either physical
or virtual, e.g., a license for a software package
• those that earn their money by realizing margin on the reselling of products produced by others
• those that earn their money by providing services, e.g., consulting, solutions integration or
implementation
4/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
Individual agents can be a pure or a mixed form of those three species. Additionally, for our research, we also
explicitly include customers as a fourth basic species in the business ecosystem model.
2.5 Technologies
A business ecosystem is defined around a core technology platform encompassing a network of suppliers and
customers, who use this platform as a means of production. For example, Microsoft’s Windows platform
technology allows thousands of programmers to develop application software to serve the specific requirements
of customers. Technology value arises when the platform technology is improved, when another complementary
product is developed, and when more end customers use it.
2.6 Network relations
A basic thrust of the business ecosystem concept is that agents do not take their decisions in isolation, but they
are influenced by the decisions of others. In other words, their behavior is ‘embedded’ in a network of relations.
Abrahamson & Rosenkopf (1997) show that the structure and density of relational networks are important
determinants for the extent of innovation diffusion. Van Asseldonk, Den Hartigh & Berger (2003) hypothesized
that the structure and density of a network are important determinants of the network dynamics. To further
investigate these conclusions, the network of relations between the agents in a business ecosystem has to be
explicitly modeled.
An important aspect of the structure of networks is whether they are local or global (Bikchandani, Hirschleifer &
Welch, 1992; Redmond, 1991). A global network effect means that agents are influenced in their adoption
decision by the behavior of other agents in the entire market. As an example, they would base a decision to buy a
‘Wintel’ or an Apple computer system on the proportions of the total world market for ‘Wintel’, respectively
Apple. Besides the global network, agents are also known to be embedded in a social structure that can influence
their behavior to a large extent (Redmond, 1991; Abrahamson & Rosenkopf, 1997). Agents are more heavily
influenced by their direct social environment. For example, while the global network effect will make it more
efficient to work with a ‘Wintel’ computer, a consumer may choose an Apple computer if he is heavily
embedded in the graphical sector, where the majority uses Apple computers.
2.7 Model input (behavior)
Firms may pursue different strategies with respect to business ecosystems. Firms can choose to follow a ‘shaper’
strategy by sponsoring their own proprietary technology that will generate high returns when it becomes
dominant in the market or they can follow an ‘adapter’ strategy, joining the dominant technology by acquiring a
license for developing products based on this technology (Besen & Farrell, 1994; Shapiro & Varian, 1999). As
the input for our model, we are concerned mainly with firms that have a ‘shaper’ strategy. Such a firm tries to
develop or maintain its own business ecosystem, with itself and its technology in the core. Iansiti & Levien
(2004) point out that such a firm can pursue this shaper strategy in different ways. It can try to become a
‘dominator’ or a ‘keystone’. A ‘dominator’ is a firm that tries to manage a large proportion of the business
ecosystem relations directly and/or tries to internalize the larger part of the added value created in the business
ecosystem. The dominator, they state, will eventually become its own business ecosystem, absorbing the
5/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
network, extracting maximum value in the short term, but destroying the business ecosystem in the long term
(Iansti & Levien, 2004, p.74). Another way for a firm to pursue a shaper strategy is to become a ‘keystone’, i.e.,
by providing a common technology platform, by being an important hub in the network, performing the task of
connecting network participants and by continually trying to improve the business ecosystem as a whole.
Needless to say, according to Iansiti & Levien (2004), this keystone approach is the strategy that will enable the
business ecosystem and the keystone itself to grow and prosper.
2.7 Model output (performance)
Different output measures may be taken to review the behavior of an ecosystem at the macro level, among which
are the speed and extent of technology diffusion, the ecosystem health (Iansiti & Levien, 2002; 2004). For the
purpose of this paper, we only consider technology diffusion.
3 Specification of the simple model
As explained above, the scope of this paper is to provide a simplified version of a business ecosystem model. In
this model, there are 20 agents, there are two technology platforms, A and B, and each agent adopts either
technology platform A or technology platform B. The agents are of species 1, 2, 3 or species 4. They switch or
stay with the technology depending on comparing the value of the two technologies at that moment for them.
Network.xls Generate Build Display
Agents ecosystem ecosystem
Agent take
decisions Statistics
output
Ecosystem.xml
System
Parameters
Figure 3: Basic system design
3.1 Initial state
In the initial state, the model is specified as follows:
• Number of agents: 20
• Number of species: 4
• Number of technologies: 2 (A and B)
• Agents’ local network relations: given by a matrix defined in an Excel file
• Total time: 5 time units (e.g., years, months) every time step is 0,5 time unit
6/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
3.2 The agents
The initial adoption state of each agent is randomly given (either technology A or technolgy B). The species an
agent belongs to and the local network relations between the agents are also randomly given. Agent size factor is
a value between 0 and 1. All these values are fixed in all simulation runs.
Variable Value Note
Company size between (0, 1) Standard distribution over the agents
Company goal between (0, 1) Standard distribution over the agents (not used in simulation)
Risk attitude between (0, 1) Standard distribution over the agents (not used in simulation)
Table 1: The agent-determined variables
3.3 The decision rules
The agents decide on which technology (A or B) to adopt based on the revenue they hope to realize with this
technology and the adoption cost of the technology. Agents continually compare the revenue from technology A
with that from technology B. Depending on their current status, the comparison of revenues will lead them to
consider switching technologies or to stay with their current technology. In considering switching, agents will
include the adoption cost for the new technology. For example, if the agent is currently using technology A:
• If Revenue (A) ≥ Revenue (B), the agent stays with A;
• If Revenue (A) < Revenue (B), the agent will consider to switch; in this case:
o If Revenue (B) – Adoption cost (B) > 0, the agent will switch to B
o If Revenue (B) – Adoption cost (B) ≤ 0, the agent will remain with A
An analogous reasoning applies when the agent is currently using technology B.
To make these decisions, the agent calculates for every period the revenues of technologies A and B and the
adoption costs of technologies A and B:
• Revenue (A) = Revenue factor (A) * Value (A)
o Revenue factor (A) is a technology characteristic with a value between 0 and 1, that can be
influenced by the supplier (keystone or dominator) of technology A
o Value (A) = wak * Inherent value (A) + wbk * Global network value (A) + wck * Local network
value (A) + r
• Inherent value (A) is a technology characteristic with a value between 0 and 1, that
can be influenced by the supplier (keystone or dominator) of technology A
• Global network value (A) is a value between 0 and 1 that is calculated as: [the number
of agents that have adopted technology A] / [the total number of agents in the
ecosystem]
• Local network value (A) is a value between 0 and 1 that is calculated as: [the number
of agents who have relations with this agent and that have adopted technology A] /
[the total number of agents who have relations with this agent]
• r is a random number between 0 and 1, representing imperfect information and
irrational behavior
7/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
• wak, wbk and wck are weight factors with values between 0 and 1; these weight factors
are determined by the species (k, where k = 1 to 4) to which the agent belongs; the
rationale is that different species will have different decision rules
• Adoption cost (A) = Adoption cost factor (A) * Agent size adoption factor * wdk
o Adoption cost factor (A) is a technology characteristic with a value between 0 and 1, that can
be influenced by the supplier (keystone or dominator) of technology A
o Agent size factor is a number between 0 and 1 randomly determined by a normal distribution
of the agents in the ecosystem; the rationale is that a larger agent will have higher adoption
cost than a smaller one
o wdk is a species-dependent adoption factor with a value between 0 and 1; the rational is that
some species, e.g., resellers that only trade a technology, will have lower adoption costs than
others, e.g., producers that use a technology to create their own products
The same kind of formulas apply, of course, for technology B.
3.4 The species
The species an agents belongs to determine the weight parameters wa, wb, wc and wd. They are all numbers
between 0 and 1.
Species
Parameter 1 2 3 4
wa (weight factor for inherent value) (0,1) (0,1) (0,1) (0,1)
wb (weight factor for global network value) (0,1) (0,1) (0,1) (0,1)
wc (weight factor for local network value) (0,1) (0,1) (0,1) (0,1)
wd (species adoption factor) (0,1) (0,1) (0,1) (0,1)
Table 2: The species-determined weight factors
3.5 The technologies
The inherent value of the technology, revenue factor of the technology, and the adoption cost factor of
technology are all numbers between 0 and 1 that can be influenced by the supplier of the technology.
Technology
Variable A B
Inherent value (IV) (0,1) (0,1)
Revenue factor (RF) (0,1) (0,1)
Adoption cost factor (ACF) (0,1) (0,1)
Table 3: The technology-determined variables
3.6 The network
8/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
The local network relations between the different agents can be user-configured in an Excel file. In this way,
different network configurations can be created and tested. For this simple model, network relations are assumed
to be bi-directional and all of the same weight.
3.7 The input
The input to the model is the behavior of the agents following a shaper strategy, i.e., the keystones or dominators
in the ecosystem. We assume that those firms are able to influence:
• the inherent value of their technology platform, e.g., by improving the quality
• the revenue factor of their technology platform, e.g., by training and certifying their network partners
• the adoption cost factor of their technology platform, e.g. by reducing the price or by providing more or
better technology support
3.8 The output
As the output measure we choose technology diffusion, which is defined here as the number of adopters of
technology A or B over time. The variables measured are:
• the minimum of the number of adopters over the specified period
• the maximum of the number of adopters over the specified period
• the mean of the number of adopters over the specified period
• the variance of the number of adopters over the specified period
• the standard deviation of the number of adopters over the specified period
• the sum of the number of adopters over the specified period
4 Experiments with the simple model
For conducting an experiment we assign different values to the parameters in the Ecosystem.xml file. The user
can use sliding controls to give different inherent values, revenue factors and adoption cost factors for
technologies A and B. When running the model for some time, we compare the number of agents using
technology A and B over time.
4.1 Experiments with network layout 1
In this initial state, there are equal numbers of technology A adopters and technology B adopters. The red dots
represent adopters of technology A, the blue dots represent adopters of technology B. There are two suppliers of
technology platforms, V0 and V1. In this experiment the weight factors (wa, wb, wc, wd) are assumed to be the
same for each species. In intuition, the two technologies should have equal chance to dominate the ecosystem, or
stay unchanged. We now run some treatments to see if this is the case.
9/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
Figure 4: Network layout 1
wa=0.4 wb=0.25 wc=0.4 wd=0.5
Treatment 0 1 2 0-0 0-1
Tech A B A B A B A B A B
IV 0.50 0.50 0.50 0.50 0.50 0.50 0.55 0.50 0.55 0.50
ACF 0.35 0.35 0.35 0.95 0.35 0.95 0.35 0.35 0.35 0.95
RF 0.40 0.80 0.80 0.40 0.10 0.20 0.40 0.80 0.10 0.20
Adopters 0 20 0 20 12 8 20 0 20 0
Time to - 2.5 - 2.5 0.5 - 2.0 - 2.0 -
prevail
Table 4: Treatments for network layout 1
From treatment 0, the following extensions are made:
• treatment 0 to 1: increase adoption cost of technology B
• treatment 1 to 2: decrease revenue factor for both
• treatment 0 to 0-0: increase inherent value of technology A
• treatment 2 to 0-1: increase inherent value of technology A
10/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
Figure 5: Treatment 0
Figure 6: Treatment 1
Figure 7: Treatment 2
Figure 8: Treatment 0-0
11/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
Figure 9: Treatment 0-1
In treatment 0 and 1, the inherent values for both technologies are the same. Compare treatment 0 and 1, no
matter how costly it is to adopt technology B, and how much higher the revenue factor for technology A is, the
ecosystem will end with the same result: all adopt B. The above result means changes in Value of technology do
not change the result. This sheds some doubt on the correctness of the parameter settings. We tested whether
perhaps because revenue factor was too large in relation to the weight parameters of wa, wb and wc. Setting
revenue factor to 0.1 and 0.2 for technologies A and B, respectively, in treatment 2 makes adopters of
technology A increase to 12. This makes the previous explanation sound reasonable.
In treatment 0-0, when the inherent value of A is only slightly larger than B, the end result flipped around:
technology A prevails. In treatment 0-1, setting both revenue factors to a smaller value (in relation to wa, wb and
wc) results in technology A still prevailing but with less adopters. Comparing treatment 0-1 with treatment 2
shows that with the increase of inherent value of technology A, more agents will adopt A. This complies with
our speculation. Therefore, we should define revenue factor values around 0.1 to 0.2.
Now take a closer look at the initial network layout 1, V18 is an important agent. It has equal number of
connections to A adopters and B adopters. V18 has 4 connections (ranking the third in this network), and it has
both connections with A adopters and B adopters. If V18 is disconnected from V0, keeping all other parameters
unchanged, the simulation result will flip around: all agents end up adopting technology B. Agent V5 has a
similar position to agent V18, only it has fewer connections with B adopters than with A adopters. When it is
disconnected from an A adopter, the same simulation result holds. These observations show network
idiosyncrasies, i.e., “the location of agents in the network that form a boundary between the fully connected
network core and a not fully connected network periphery, can have a large influence on the extent of innovation
diffusion” (Abrahamson & Rosenkopf, 1997). So the hint is that for keystone V0, it is very important to keep the
loyalty of such partners.
4.2 Experiments with network layout 2
In this initial state of ecosystem, there are 9 technology A adopters and 11 technology B adopters. Again, the red
dots represent adopters of technology A and the blue dots represent adopters of technology B. There is only one
keystone agent, V0. In this experiment the weight factors (wa, wb, wc, wd) are assumed to be the same for each
species. It is to be expected that technology A will attract more agents. We now run some treatments to test this.
12/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
Figure 10: Network layout 2
wa=0.2 wb=0.25 wc=0.4 wd=0.5
Treatment 3 4 3-0 4-0
Tech A B A B A B A B
IV 0.5 0.5 0.5 0.5 0.5 0.53 0.5 0.53
ACF 0.35 0.35 0.65 0.35 0.35 0.35 0.65 0.35
RF 0.1 0.2 0.1 0.2 0.1 0.2 0.1 0.2
adopters 15 5 15 5 0 20 0 20
Time to 2.0 - 2.0 - - 2.0 - 2.5
prevail
Treatment 3-1 4-1 3-2 4-2
Tech A B A B A B A B
IV 0.5 0.53 0.5 0.53 0.5 0.6 0.5 0.6
ACF 0.35 0.35 0.65 0.35 0.35 0.35 0.35 0.65
RF 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2
adopters - - 0 20 - - - -
Time to 2.5
prevail
Table 5: Treatments for network layout 2
13/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
From treatment 3, the following extensions are made:
• treatment 3 to 4: increase the adoption cost of technology A
• treatment 3 to 3-0 and 4 to 4-0: increase inherent value of technology B
• treatment 3-0 to 3-1 and 4-0 to 4-1: increase the revenue factor of technology A
• treatment 3-1 tot 3-2: further increase inherent value of technology B
• treatment 4-1 to 4-2: increase the adoption cost of technology B
Figure 11: Treatments 3 & 4
Figure 11: Treatments 3-0 & 4-0
Figure 12: Treatments 4-1 & 3-1 (note the graph of 4-1 is on the left, that of 3-1 on the right)
14/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
Figure 13: Treatments 3-2 & 4-2
In treatments 3 and 4, technology A wins the majority, even when it has larger adoption cost. This result is
according to expectations.
In treatments 3-0 and 4-0 the inherent value of B is slightly larger than A. Technology B prevails, showing that
in some cases a small difference in inherent value may tilt the market.
In treatments 4-1 and 3-1 the revenue factor of technology A is raised, but technology B still prevails. Lowering
the adoption cost of technology A causes a kind of dynamic equilibrium, in which neither of the technologies
prevails.
In treatment 3-2 and 4-2, as B’s inherent value is further increased and we would expect that it becomes less easy
for A to prevail in the ecosystem. This is the case with the simulation result. When B’s adoption cost becomes
larger there are less agents adopting B. In both cases we see a kind of dynamic equilibrium, in which neither of
the technologies prevails.
4.3 Experiment with network layout 3
Treatment 5 has the same parameter settings as Treatment 0. The only difference lies in the network layout,
where the density is higher. According to Abrahamson & Rosenkopf (1997), a “network with a higher density
results in a higher extent of diffusion of an innovation, i.e., more agents within the network eventually adopt this
innovation.” To put it another way, in this model we should find that technology B will prevail within shorter
time period. It takes 1.5 years as shown in figure 15, while in treatment 0 it takes 2.5 years.
15/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
Figure 14: Network layout 3
wa=0.4 wb=0.25 wc=0.4 wd=0.5
Treatment 5
Tech A B
IV 0.5 0.50
ACF 0.35 0.35
RF 0.4 0.8
Adopters - 20
Time to 1.5
prevail
Table 6: Treatment for network layout 3
Figure 15: Treatment 5
16/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
5 Propositions from the simple model
A number of propositions can be formulated from the above experiments with the simple agent-based business
ecosystem model. These propositions can be used as hypotheses in testing an extended version of the model.
1. the keystone agent is influential in the sense that the system’s final state is very dependent on such
agent’s initial state
2. an agent will eventually adopt the technology that dominates its local network
3. network density has a very positive impact on the speed of technology diffusion
4. an ecosystem model will demonstrate network idiosyncrasies
5. agents that occupy a central position in the ecosystem are important in determining the technology
diffusion
Based on the simple model and the limited number of experimental treatments no definitive conclusions can be
drawn. However, it seems plausible to state that network structure is an important factor in the diffusion of
technology through a business ecosystem. Therefore, if a keystone agent wants to increase its technology
diffusion, it is important to analyze the network structure and the positions of individual agents (partners) in the
network. If a keystone knows which key agents (partners) it should target and if it can strengthen the relationship
with those agents, the diffusion of its technology platform will likely speed up. Network analysis will provide the
answer to identify such agents.
Besides the importance of network structure, our treatments show that small changes in inherent value can in
some cases be decisive in tilting the market. This is contrary to expectations. The default reasoning in literature
(e.g., Arthur, 1989) is that in markets with network effects, the ‘best’ technology, in terms of quality, i.e.,
inherent value, may not always prevail. In some cases this is still true, see treatments 3-1, 3-2 and 4-2. However,
form our simple model and experiments it seems plausible that technology quality, i.e., inherent value, does have
an important impact on diffusion. Therefore, looking at network structure alone is not sufficient to explain
technology diffusion in business ecosystems.
6 Literature
Abrahamson, Eric & Lori Rosenkopf (1997), Social network effects on the extent of innovation diffusion: a
computer simulation, Organization Science, vol.8 no.3, pp.289-309.
Arthur, W.B. (1989), Competing technologies, increasing returns, and lock-in by historical events, Economic
Journal, vol.99, pp.116-131.
Besen, S.M. & J. Farrell (1994), Choosing how to compete: strategies and tactics in standardization, Journal of
Economic Perspectives, vol.8, no.2, pp.117-131.
Bikhchandani, S., D. Hirschleifer & I. Welch (1992), A theory of fads, fashion, custom, and cultural changes as
informational cascades, Journal of Political Economy, vol.100, no.5, pp.992-1026.
17/18
ECCON 2005 Annual meeting
Modeling a business ecosystem
Borshchev, Andrei & Alexei Filippov (2004), From System Dynamics and Discrete Event to Practical Agent
Based Modeling: Reasons, Techniques, Tools.
Den Hartigh, Erik & Ton van Asseldonk (2004): Business ecosystems: A research framework for investigating
the relation between network structure, firm strategy, and the pattern of innovation diffusion; Contribution to the
European Chaos/Complexity in Organizations Network (ECCON) Conference, October 2004
Iansiti, Marco and Roy Levien (2002): Keystones and Dominators – Framing the operational dynamics of
business ecosystem; Working Paper.
Iansiti, Marco and Roy Levien (2004); Strategy as ecology; Harvard Business Review; March, pp.68-78.
Iansiti Marco and Roy Levien (2004); The Keystone Advantage: What the New Dynamics of Business
Ecosystems Mean for Strategy, Innovation, and Sustainability; Harvard Business School Press, Cambridge
(MA).
Jennings, Nicholas R. (2003), A Methodology for Agent-Oriented Analysis and Design.
Moore, James F. (1996), The Death of Competition: Leadership and Strategy in the Age of Business
Ecosystems, Wiley & Sons, Chichester.
Redmond, William H. (1991), When technologies compete: the role of externalities in nonlinear marker
response, Journal of Product Innovation Management, vol.8, pp.170-183.
Shannon, Robert E. (1975), Systems Simulation: the art and science. Prentice-Hall.
Shapiro, Carl & Hal R. Varian (1999), The art of standards wars, California Management Review, vol.41, no.2,
pp.8-32.
Van Asseldonk, Ton, Erik den Hartigh & Leon Berger (2003), Morphology, entropy, and stability in networked
structures, European Chaos/Complexity in Organisations Network (ECCON) annual meeting 2003, in: Putnik,
G.D. & A. Gunasekaran (eds.) Business Excellence I: Performance Measures, Benchmarking and Best Practices
in New Economy; University of Minho, Braga, pp.255-260.
18/18
