A supply chain or logistics network is the system of organizations, people, technology, activities,
information and resources involved in moving a product or service from supplier to customer.
Supply chain activities transform natural resources, raw materials and components into a finished
product that is delivered to the end customer. In sophisticated supply chain systems, used
products may re-enter the supply chain at any point where residual value is recyclable. Supply
chains link value chains.
A typical supply chain begins with ecological and biological regulation of natural resources,
followed by the human extraction of raw material and includes several production links, for
instance; component construction, assembly and merging before moving onto several layers of
storage facilities of ever decreasing size and ever more remote geographical locations, and
finally reaching the consumer.
Many of the exchanges encountered in the supply chain will therefore be between different
companies who will seek to maximize their revenue within their sphere of interest, but may have
little or no knowledge or interest in the remaining players in the supply chain. More recently, the
loosely coupled, self-organizing network of businesses that cooperates to provide product and
service offerings has been called the Extended Enterprise.
Supply Chain Management
In the 1980s the term Supply Chain Management (SCM) was developed to express the need to
integrate the key business processes, from end user through original suppliers - Original
suppliers being those that provide products, services and information that add value for
customers and other stakeholders. The basic idea behind the SCM is that companies and
corporations involve themselves in a supply chain by exchanging information regarding market
fluctuations, production capabilities.
If all relevant information is accessible to any relevant company, every company in the supply
chain has the possibility to and can seek to help optimizing the entire supply chain rather than
sub optimize based on a local interest. This will lead to better planned overall production and
distribution which can cut costs and give a more attractive final product leading to better sales
and better overall results for the companies involved.
Incorporating SCM successfully leads to a new kind of competition on the global market where
competition is no longer of the company versus company form but rather takes on a supply chain
versus supply chain form.
The primary objective of supply chain management is to fulfill customer demands through the
most efficient use of resources, including distribution capacity, inventory and labor. In theory, a
supply chain seeks to match demand with supply and do so with the minimal inventory. Various
aspects of optimizing the supply chain include liaising with suppliers to eliminate bottlenecks;
sourcing strategically to strike a balance between lowest material cost and transportation,
implementing JIT (Just In Time) techniques to optimize manufacturing flow; maintaining the
right mix and location of factories and warehouses to serve customer markets, and using
location/allocation, vehicle routing analysis, dynamic programming and, of course, traditional
logistics optimization to maximize the efficiency of the distribution side.
There is often confusion over the terms supply chain and logistics. It is now generally accepted
that the term Logistics applies to activities within one company/organization involving
distribution of product whereas the term Supply chain also encompasses manufacturing and
procurement and therefore has a much broader focus as it involves multiple enterprises,
including suppliers, manufacturers and retailers, working together to meet a customer need for a
product or service.
Starting in the 1990s several companies choose to outsource the logistics aspect of supply chain
management by partnering with a 3PL, Third-party logistics provider. Companies also outsource
production to contract manufacturers.
There are actually four common Supply Chain Models. Besides the two mentioned above, there
are the American Productivity & Quality Center's (APQC) Process Classification Framework
and the Supply Chain Best Practices Framework. Critics have questioned the validity of all these
Supply chain activities can be grouped into strategic, tactical, and operational levels of activities.
Supply Chain Activities
Strategic network optimization, including the number, location, and size of warehouses,
distribution centers and facilities
Strategic partnership with suppliers, distributors, and customers, creating communication
channels for critical information and operational improvements such as cross docking, direct
shipping, and third-party logistics
Product design coordination, so that new and existing products can be optimally integrated into
the supply chain, load management
Information Technology infrastructure, to support supply chain operations
Where-to-make and what-to-make-or-buy decisions
Aligning overall organizational strategy with supply strategy
Sourcing contracts and other purchasing decisions
Production decisions, including contracting, scheduling, and planning process definition
Inventory decisions, including quantity, location, and quality of inventory
Transportation strategy, including frequency, routes, and contracting
Benchmarking of all operations against competitors and implementation of best practices
throughout the enterprise
Focus on customer demand
Daily production and distribution planning, including all nodes in the supply chain
Production scheduling for each manufacturing facility in the supply chain (minute by minute)
Demand planning and forecasting, coordinating the demand forecast of all customers and
sharing the forecast with all suppliers
Sourcing planning, including current inventory and forecast demand, in collaboration with all
Inbound operations, including transportation from suppliers and receiving inventory
Production operations, including the consumption of materials and flow of finished goods
Outbound operations, including all fulfillment activities and transportation to customers
Order promising, accounting for all constraints in the supply chain, including all suppliers,
manufacturing facilities, distribution centers, and other customers.
Developments in Supply Chain Management
Six major movements can be observed in the evolution of supply chain management studies:
Creation, Integration, and Globalization, Specialization Phases One and Two, and SCM 2.0.
The term supply chain management was first coined by an American industry consultant in the
early 1980s. However the concept of supply chain in management, was of great importance long
before in the early 20th century, especially by the creation of the assembly line. The
characteristics of this era of supply chain management include the need for large scale changes,
reengineering, downsizing driven by cost reduction programs, and widespread attention to the
Japanese practice of management.
This era of supply chain management studies was highlighted with the development of
Electronic Data Interchange (EDI) systems in the 1960s and developed through the 1990s by the
introduction of Enterprise Resource Planning (ERP) systems. This era has continued to develop
into the 21st century with the expansion of internet-based collaborative systems. This era of SC
evolution is characterized by both increasing value-added and cost reduction through integration.
The third movement of supply chain management development, globalization era, can be
characterized by the attention towards global systems of supplier relations and the expansion of
supply chain over national boundaries and into other continents. Although the use of global
sources in the supply chain of organizations can be traced back to several decades ago (e.g. the
oil industry), it was not until the late 1980s that a considerable number of organizations started to
integrate global sources into their core business. This era is characterized by the globalization of
supply chain management in organizations with the goal of increasing competitive advantage,
creating more value-added, and reducing costs through global sourcing.
Specialization Era -- Phase One -- Outsourced Manufacturing and Distribution
In the 1990s industries began to focus on “core competencies” and adopted a specialization
model. Companies abandoned vertical integration, sold off non-core operations, and outsourced
those functions to other companies. This changed management requirements by extending the
supply chain well beyond the four walls and distributing management across specialized supply
This transition also refocused the fundamental perspectives of each respective organization.
OEMs became brand owners that needed deep visibility into their supply base. They had to
control the entire supply chain from above instead of from within. Contract manufacturers had to
manage bills of material with different part numbering schemes from multiple OEMs and
support customer requests for work -in-process visibility and vendor-managed inventory (VMI).
The specialization model creates manufacturing and distribution networks composed of multiple,
individual supply chains specific to products, suppliers, and customers, who work together to
design, manufacture, distribute, market, sell, and service a product. The set of partners may
change according to a given market, region, or channel, resulting in a proliferation of trading
partner environments, each with its own unique characteristics and demands.
Specialization Era -- Phase Two -- Supply Chain Management as a Service
Specialization within the supply chain began in the 1980s with the inception of transportation
brokerages, warehouse management, and non asset based carriers and has matured beyond
transportation and logistics into aspects of supply planning, collaboration, execution and
At any given moment, market forces could demand changes within suppliers, logistics providers,
locations, customers and any number of these specialized participants within supply chain
networks. This variability has significant effect on the supply chain infrastructure, from the
foundation layers of establishing and managing the electronic communication between the
trading partners to the more-complex requirements, including the configuration of the processes
and work flows that are essential to the management of the network itself.
Supply chain specialization enables companies to improve their overall competencies in the same
way that outsourced manufacturing and distribution has done; it allows them to focus on their
core competencies and assemble networks of best in class domain specific partners to contribute
to the overall value chain itself – thus increasing overall performance and efficiency. The ability
to quickly obtain and deploy this domain specific supply chain expertise without developing and
maintaining an entirely unique and complex competency in house is the leading reason why
supply chain specialization is gaining popularity.
Outsourced technology hosting for supply chain solutions debuted in the late 1990s and has
taken root in transportation and collaboration categories most dominantly. This has progressed
from the Application Service Provider (ASP) model from approximately 1998 through 2003 to
the On-Demand model from approximately 2003-2006 to the Software as a Service (SaaS)
model we are currently focused on today.
Supply Chain Management 2.0 (SCM 2.0)
Building off of globalization and specialization, SCM 2.0 has been coined to describe both the
changes within the supply chain itself as well as the evolution of the processes, methods and
tools that manage it in this new "era".
Web 2.0 is defined as a trend in the use of the World Wide Web that is meant to increase
creativity, information sharing, and collaboration among users. At its core, the common attribute
that Web 2.0 brings is it helps us navigate the vast amount of information available on the web to
find what we are looking for. It is the notion of a usable pathway. SCM 2.0 follows this notion
into supply chain operations. It is the pathway to SCM results – the combination of the
processes, methodologies, tools and delivery options to guide companies to their results quickly
as the complexity and speed of the supply chain increase due to the effects of global competition,
rapid price commoditization, surging oil prices, short product life cycles, expanded
specialization, near/far and off shoring, and talent scarcity.
SCM 2.0 leverages proven solutions designed to rapidly deliver results with the agility to quickly
manage future change for continuous flexibility, value and success. This is delivered through
competency networks composed of best of breed supply chain domain expertise to understand
which elements, both operationally and organizationally, are the critical few that deliver the
results as well as the intimate understanding of how to manage these elements to achieve desired
results, finally the solutions are delivered in a variety of options as no-touch via business process
outsourcing, mid-touch via managed services and software as a service (SaaS), or high touch in
the traditional software deployment model.
Supply Chain Management Problems
Supply chain management must address the following problems:
Distribution Network Configuration
Number, location and network missions of suppliers, production facilities, distribution centers,
warehouses, cross-docks and customers
Including questions of operating control (centralized, decentralized or shared); delivery scheme
(e.g., direct shipment, pool point shipping, Cross docking, DSD (direct store delivery), closed
loop shipping); mode of transportation (e.g., motor carrier, including truckload, LTL, parcel;
railroad; intermodal, including TOFC and COFC; ocean freight; airfreight); replenishment
strategy (e.g., pull, push or hybrid); and transportation control (e.g., owner-operated, private
carrier, common carrier, contract carrier, or 3PL). Trade-Offs in Logistical Activities
Integration of all processes through the supply chain to share valuable information, including
demand signals, forecasts, inventory, transportation, and potential collaboration etc
Quantity and location of inventory including raw materials, work-in-process and finished goods
Arranging the payment terms and the methodologies for exchanging funds across entities within
the supply chain
The above activities must be coordinated well together in order to achieve the least total logistics
cost. Trade-offs exist that increase the total cost if only one of the activities is optimized. For
example, full truckload (FTL) rates are more economical on a cost per pallet basis than less than
truckload (LTL) shipments. If, however, a full truckload of a product is ordered to reduce
transportation costs there will be an increase in inventory holding costs which may increase total
logistics costs. It is therefore imperative to take a systems approach when planning logistical
activities. These trade-offs are key to developing the most efficient and effective Logistics and
An intelligent agent (IA) is an entity which observes and acts upon an environment (i.e. it is an
agent) and directs its activity towards achieving goals (i.e. it is rational). Intelligent agents may
also learn or use knowledge to achieve their goals. They may be very simple or very complex: a
reflex machine is an intelligent agent, as is a human being, as is a community of human beings
working together towards a goal.
Simple reflex agent
Intelligent agents are often described schematically as an abstract functional system similar to a
computer program. For this reason, intelligent agents are sometimes called abstract intelligent
agents (AIA) to distinguish them from their real world implementations as computer systems,
- 10 -
biological systems, or organizations. AIA is an entity which exhibits an essence of human-like
intelligence and, as an IA, may have numerous other properties resulting from the properties of
its carrier physical or software system. For this reason IA can be either rational or
emotive/irrational or, according to Herbert Simon, it represents bounded rationality.
Some definitions of intelligent agents emphasize their autonomy, and so prefer the term
autonomous intelligent agents. Still others (notably Russell & Norvig) considered goal-directed
behavior as the essence of rationality and so preferred the term rational agent.
In order to separate necessary and not necessary properties of IA, in the computational TOGA
meta-theory, every cognitive AIA acts on the base of its/his/her available information, possessed
preferences and knowledge (IPK model) with a different range, on various abstraction levels, and
in different domains of activity. Such agent is called personoid. The quality of application and
processing of its information, knowledge and preferences depends on the characteristics of AIA's
carrier system, i.e. memory available, velocity and other its structural properties. According to
different I, P,K bases, IA may be specialized for numerous roles.
Intelligent agents are closely related to agents in economics, and versions of the intelligent agent
paradigm are studied in cognitive science, ethics, the philosophy of practical reason, as well as in
many interdisciplinary socio-cognitive modeling and computer social simulations.
Intelligent agents are also closely related to software agents (an autonomous software program
that assists users). In computer science, the term intelligent agent may be used to refer to a
software agent that has some intelligence, regardless if it is not a rational agent by Russell and
Norvig's definition. For example, autonomous programs used for operator assistance or data
mining (sometimes referred to as bots) are also called "intelligent agents".
Classes of Intelligent Agents
Russell & Norvig describe multiple types of agents and sub-agents. For example:
Physical Agents - A physical agent is an entity which percepts through sensors and acts through
Temporal Agents - A temporal agent may use time based stored information to offer instructions
or data acts to a computer program or human being and takes program inputs percepts to adjust
its next behaviors.
- 11 -
Believable agents - An agent exhibiting a personality via the use of an artificial character (the
agent is embedded) for the interaction.
A simple agent program can be defined mathematically as an agent function which maps every
possible percepts sequence to a possible action the agent can perform or to a coefficient,
feedback element, function or constant that affects eventual actions:
f:P * - > A
The program agent, instead, maps every possible percept to an action.
It is possible to group agents into five classes based on their degree of perceived intelligence and
Simple reflex agents
Model-based reflex agents
Simple reflex agents act only on the basis of the current percept. The agent function is based on
the condition-action rule: if condition then action
This agent function only succeeds when the environment is fully observable. Some reflex agents
can also contain information on their current state which allows them to disregard conditions
whose actuators are already triggered.
Model-based reflex agents can handle partially observable environments. Its current state is
stored inside the agent maintaining some kind of structure which describes the part of the world
which cannot be seen. This behavior requires information on how the world behaves and works.
This additional information completes the “World View” model.
Goal-based agents are model-based agents which store information regarding situations that are
desirable. This allows the agent a way to choose among multiple possibilities, selecting the one
which reaches a goal state.
Utility-based agents - Goal-based agents only distinguish between goal states and non-goal
- 12 -
states. It is possible to define a measure of how desirable a particular state is. This measure can
be obtained through the use of a utility function which maps a state to a measure of the utility of
Learning agents learn from experience by evaluating performance of experiments against new
Other classes of intelligent agents - Some of the sub-agents that may be a part of an intelligent
agent or a complete intelligent agent in themselves are:
Temporal Agents (for time-based decisions);
Spatial Agents (that relate to the physical real-world);
Input Agents (that process and make sense of sensor inputs - example neural network based
agents neural network);
Processing Agents (that solve a problem like speech recognition);
Decision Agents (that are geared to decision making);
Learning Agents (for building up the data structures and database of other intelligent agents);
World Agents (that incorporate a combination of all the other classes of agents to allow
Employment of Decision Agents in E-commerce
A supply chain is a network of suppliers, factories, warehouses, distribution centers and retailers,
- 13 -
through which raw materials are acquired, transformed, produced and delivered to the customer.
A supply chain management system (SCMS) manages the cooperation of these system
components. The roles of individual entities in a supply chain can be implemented as distinct
agents. The functions and procedures of a company in the real market are complicated and
include information collection, policy making and actions. It is impossible to describe software
agent behaviors for an uncertain e-commerce environment such as supply chain management in
the traditional single threaded model. It is therefore required that the concept of negotiation is
introduced into software agent design for supply chain management to solve the problem of
communication and decision-making for negotiating agents.
Required Decision Agent Characteristics
A decision agent is a self-contained, callable service with a view of all the conditions and actions
that need to be considered to make an operational business decision. Or, more simply it is a
service that answers a business question for other services.
A decision agent must conform to the standard characteristics for a well defined service plus, in
Have a Behavior understandable to the business - After all we are talking about a "business
decision" here so the business had better be able to verify exactly what is going on inside.
Support rapid iteration without disruption - Business decisions change all the time so a decision
agent has to be both flexible and designed for this change.
Integrate historical data - Business decisions are increasingly made "by the numbers" with much
reference to historical data. Decision agents need a similar ability to use historical data, and
trends/insight extrapolated from it.
Expect multi-channel use - While this is largely covered by the standard items it is worth noting
as it means that VERY different kinds of applications will use the service - everything from other
agents in the enterprise to other applications in the supply chain belonging to other enterprises.
Manage exceptions well - Not only should it respond sensibly when it cannot decide, it should
ensure that enough context is returned as to why it could not decide to assist a manual process.
Must explain its execution - Many decisions must demonstrate compliance or conformance with
policy. Any decision agent must be able to log exactly how it decided and that information must
- 14 -
be accessible to non-technical users.
Data Flow options in Decision Agents
Decision agents should be fundamentally synchronous. There are various options to implement
how the data management for decision agents:
1. A decision agent can be passed all the data available and forced to either decide or to pass
back some reason why it could not decide (to do with a lack of data, say) so that the calling
application can assemble the additional data required and try again.
2. A decision agent can be passed the data available but allowed to call external services and/or
databases to gather the data it needs to complete the decision. Only synchronous calls should
be allowed as the service should remain synchronous.
3. A decision agent can be passed the data available and allowed to call external services and to
request additional data from a user interface. The decision is still synchronous in that it
continues to run/use a thread until the data is provided through the user interface or the
request for a decision is cancelled.
4. A decision agent can be passed the data available and allow it to gather the data it needs in
any way. While the decision remains synchronous, in that the calling service is waiting for an
answer, the decision thread may be put “on ice” while waiting for the necessary data.
If synchronous behavior is not required then clearly a decision agent can be invoked
asynchronously, allowed to gather the data it needs to make a decision and then return its result,
typically through transmission of an event. This kind of service is common in event-based
Multi-agent Supply Chain Management Systems
An agent-based model (ABM) is a computational model for simulating the actions and
interactions of autonomous individuals in a network, with a view to assessing their effects on the
system as a whole. It combines elements of game theory, complex systems, emergence,
computational sociology, multi agent systems, and evolutionary programming. Monte Carlo
Methods are used to introduce randomness.
The models simulate the simultaneous operations of multiple agents, in an attempt to re-create
- 15 -
and predict the actions of complex phenomena. The process is one of emergence from the lower
(micro) level of systems to a higher (macro) level. The individual agents are presumed to be
acting in what they perceive as their own interests, such as reproduction, economic benefit, or
social status, and their knowledge is limited. ABM agents may experience "learning", adaptation,
A multi-agent system (MAS) is a system composed of multiple interacting intelligent agents.
Multi-agent systems can be used to solve problems which are difficult or impossible for an
individual agent or monolithic system to solve. Examples of problems which are appropriate to
multi-agent systems research include online trading, disaster response, and modeling social
The agents in a multi-agent system have several important characteristics:
Autonomy: the agents are at least partially autonomous
Local views: no agent has a full global view of the system, or the system is too complex for an
agent to make practical use of such knowledge
Decentralization: there is no one controlling agent (or the system is effectively reduced to a
Multi-agent systems can manifest self-organization and complex behaviors even when the
individual strategies of all their agents are simple. Agents can share knowledge using any agreed
language, within the constraints of the system's communication protocol. Example languages are
Knowledge Query Manipulation Language (KQML) or FIPA's Agent Communication Language
A Supply Chain Management System is transformed into a multi-agent system when the
software agents enter into the market. Since software agents might belong to different companies
and are self-interested, a pure scheduling scheme can not help. In addition, software agents tend
to cooperate in a relatively dynamic way. To address these problems, a Multi-agent system of
negotiating agents for supply chain management is required. Where there is no preset
relationship between agents. When an order comes, a virtual supply chain may emerge through
negotiation processes. The components of the chain may change according to the external
situation even after the order has been accepted.
- 16 -
Four general issues need to be addressed by software agents in a multi-agent system for supply
1. Communication. In order to support meaningful communication among the negotiating
parties there need to be a common language for expressing primitive communicative acts that
make up a negotiation (e.g., a call for proposals, a rejection of a proposal, etc.) as well as a
way to specify different protocols that can be used (e.g., English auction, contract net,
2. Representation. Most negotiation is about complex objects (physical or abstract) that may
require the support of a sophisticated representation scheme. Examples can range from orders
for goods to contracts for services.
3. Problem solving. Many aspects of negotiation can be modeled as an exercise in distributed
constraint solving. There is a large body of work on algorithms and techniques for constraint
solving that can be applied to negotiation problems.
4. Human interaction. Negotiation has to be carried out in the context of existing human
organizations. Whatever automated negotiation processes have to be coupled with humans in
appropriate ways, either for authorization and modification or as part of a larger workflow
The behaviors of a software agent, which correspond to the above four aspects of negotiation, are
called negotiating actions. A software agent is a negotiating agent if it can at least take
communicating and problem solving actions in a specific domain.
A Multi-agent system for supply chain management can consist of two types of negotiating
agents – functional agents and information agents. Functional agents implement some supply
chain management functionality. These agents are usually owned by different companies and are
therefore assumed to be self-interested and thus free to join, remain in or leave the supply chain
system. Information agents are predefined in the system and help functional agents to find
potential negotiation partners or provide other altruistic service such as accepting the registration
from a functional agent. All of the negotiating agents have some understanding of system
ontology and use a certain Agent Communication Language (ACL) to make conversation.
- 17 -
The interaction between them, the negotiation process, is modeled as a process of cooperatively
assigning values to a set of variables. There are no centralized super-agents or distributed
mediators to handle the agent cooperation. All these activities occur through negotiation
processes, regardless of whether two sides are involved in bargaining for some goods
intentionally or de-committing a contract caused by the outside events.
In a supply chain negotiation process, negotiating agents use an Agent Communication
Language (ACL) to bargain with each other. The table below presents the peformatives designed
for the negotiating agents based on FIPA ACL. A negotiation protocol, formally described using
Color Petri Net (CPN) is also given.
Implementing Decision Agents
- 18 -
A simple and accessible program for creating agent-based models is NetLogo. NetLogo was
originally designed for educational purposes but now numbers many thousands of research users
as well. Many colleges have used this as a tool to teach their students about agent-based
modeling. A similar program, StarLogo, has also been released with similar functionality. Swarm
was one of the first general purpose ABM systems. Swarm, developed by the Swarm
Development Group, uses the Objective C programming language, and is recommended for C
programmers with little object-oriented programming experience. Swarm can also be
implemented by Java programmers, as can Ascape. Both MASON and Repast are widely used,
and EcoLab is suitable for C++ programmers. Cormas is another platform, focusing on natural
resources management, rural development or ecology research, based on the Smalltalk language.
All the toolkits described previously are based on serial von-Neumann computer architectures.
This limits the speed and scalability of these systems. A recent development is the use of data-
parallel algorithms on Graphics Processing Units GPUs for ABM simulation. The extreme
memory bandwidth combined with the sheer number crunching power of multi-processor GPUs
has enabled simulation of millions of agents at tens of frames per second
Agent Based Model vs. Equation Based Model Approaches to Supply Chain Management
The discrete, dynamic and distributed nature of data and applications require that supply chain
solutions not merely respond to requests for information but intelligently anticipate, adapt and
actively support users. Agents can support a clearly discernible task or process, interact with
each other in a specified environment (say, inventory management), interact with other Agents
directly or via a message bus, continuously harness real-time data (for example, from RFID tags,
GPS, sensors) and share this data with other Agents to offer true real-time adaptability in supply
This concept is at the heart of Multi-Agent System. Real-time adaptability may affect a vast
array of static or pre-set business processes. It is likely that many processes may change to
evolve into the paradigm shift that is leading toward the adaptable business network (ABN). In
particular, real-time adaptability may revolutionize supply chain management, fostering supply
chain innovation through deployment of Multi-Agent Systems. Agent-based modeling draws
clues from natural behavior of biological communities of ants, wasps, termites, birds, fishes and
wolves, to name a few.
- 19 -
In commercial supply chain software (i2, SAP, Oracle, Manugistics) processes are defined in
terms of rates and flows (consumption, production). System variables (cost, rebates,
transportation time, out-of-stock) evaluate or integrate sets of algebraic equations (ODE,
ordinary differential equations or PDE, partial differential equations) relating these variables to
optimize for best results (best price, shortest lead time, minimal inventory). The process (EBM
or equation-based modeling) assumes that these parameters are linear in nature and relevant data
are available. In the real world, events are non-linear, actions are discrete and information about
data is distributed.
Agents-based supply chain software may function continuously and autonomously in a particular
environment, often inhabited by other Agents (Multi-Agent Systems) and processes. Continuity
and autonomy indicates that Agents are able to execute processes or carry out activities in a
flexible and intelligent manner that is both adaptive and responsive to changes in the
environment without requiring constant human guidance, intervention or top-down control from
a system operator. An Agent that functions continuously in an environment over a period of time
would be able to learn from experience (patterns). In addition, Agents that inhabit an
environment with other Agents (Multi-Agent Systems) are able to communicate, cooperate and
are mobile (from one environment to another). The mobile, networked, autonomous, self-
learning, adaptive Agent may have radically different principles compared to those that were
developed for monolithic systems. Examination of naturally occurring Agent-based systems
suggests design principles for the next generation of Agents. While particular circumstances may
warrant deliberate exceptions, in general, the research aligns with these concepts:
1. Agents should correspond to “things” in the problem domain rather than to abstract
2. Agents should be small in mass, time (able to forget) and scope (avoid global knowledge
3. Multi-Agent Systems should be decentralized (no single point of control/failure).
4. Agents should be neither homogeneous nor incompatible but diverse.
5. Agent communities should include a dissipative mechanism (entropy leak).
6. Agents should have ways of caching and sharing what they learn about their environment.
- 20 -
7. Agents should plan and execute concurrently rather than sequentially.
Computer-based modeling has largely used system dynamics based on ODE (ordinary
differential equations). However, a multitude of industrial and businesses, including supply chain
management, are struggling to respond in real-time. Eventually this transition may emerge as
real-time adaptable business network. This paradigm shift will make it imperative to model
software based both with agents and equations. The question is no longer whether to select one
or the other approach but to establish a mix of both and develop criteria for selecting one or other
approach, that can offer solutions. The “balance” is itself subject to change. For experts supply
chain management situation is analogous to “push-pull” strategy where the push-pull boundary
may shift with changing demand.
Difference in representational focus between ABM (Agent Based Model) vs. EBM (Equation
Based Model) has consequences for how models are modularized. EBMs represent the system as
a set of equations that relate observables to one another. The basic unit of the model, the
equation, typically relates observables whose values are affected by the actions of multiple
individuals, so the natural modularization often crosses boundaries among individuals. ABM
represents the internal behavior of each individual. An Agent’s behavior may depend on
observables generated by other individuals, but does not directly access the representation of
those individuals’ behaviors, so the natural modularization follows boundaries among
individuals. This fundamental difference in model structure gives ABM a key advantage in
commercial applications such as adaptable supply chain management, in two ways:
First, in an ABM, each firm has its own Agents. An Agent’s internal behaviors are not required
to be visible to the rest of the system, so firms can maintain proprietary information about their
internal operations. Groups of firms can conduct joint modeling exercises (Public Marketplace)
while keeping their individual Agents on their own computers, maintaining whatever controls are
needed. Construction of EBMs requires disclosure of relationships that each firm maintains on
observables so that equations can be formulated and evaluated. Distributed execution of EBM is
not impossible, but does not naturally respect boundaries among the individuals (why public e-
Marketplaces failed to take-off).
Second, in many cases, simulation of a system is part of a larger project whose desired outcome
is a control scheme that more or less automatically regulates the behavior of the entire system.
- 21 -
Agents correspond one-to-one with the individuals (firms or divisions of firms) in the system
being modeled, and their behaviors are analogs of the real behaviors. These two characteristics
make Agents a natural locus for the application of adaptive techniques that can modify their
behaviors as Agents execute, so as to control emergent behavior of the overall system. Migration
from simulation model to adaptive control model is much straightforward in ABM than EBM.
One can imagine a member of adaptable business network or supply chain using its simulation
Agent as the basis for an automated control Agent that handles routine interactions with trading
partners. It is much less likely that such a firm would submit aspects of its operation to an
external “equation manager” that maintains specified relationships among observables from
ABMs support more direct experimentation. Managers playing “what-if” games with the model
can think directly in terms of familiar business processes, rather than having to translate them
into equations relating observables. ABMs are easier to translate back into practice. One purpose
of “what-if” experiments is to identify improved business practices that ca be implemented. If
the model is expressed and modified directly in terms of behaviors, implementation of its
recommendations is a matter of transcribing the modified behaviors of Agents into task
descriptions for the underlying physical entities in the real world.
The disadvantages of EBM result largely from the use of averages of critical system variables
over time and space. EBM assumes homogeneity among individuals but individuals in real
systems are often heterogeneous. When the dynamics are non-linear, local variations from the
averages can lead to significant deviations in overall system behavior. In business applications
driven by ‘if-then’ decisions, non-linearity is the rule. Because ABMs are inherently local, it is
natural to let each Agent monitor the value of system variables locally, without averaging over
time and space and thus without losing the local idiosyncrasies that can determine overall system
The approach to system design and supply chain management with Agents in the software
landscape is at odds with the centralized top-down tradition in current systems. The question
usually arises in terms of the contrast between local and global optimization. Decision-makers
fear that by turning control of a system over to locally autonomous Agents without a central
- 22 -
decision-making body, they will lose value that could have been captured by an integrated
(enterprise) global approach. The benefits of Agent-based architecture approaches vs. centralized
ones are conditional, not absolute. In a stable environment, a centralized approach can be
optimized to out-perform the initial efforts of an opportunistic distributed system of Agents. If
the distributed system has appropriate learning capabilities, it will eventually become as
efficient. Market conditions are marked by rapid and unpredictable change, not stability. Change
and contingency are inescapable features of the real world. The appropriate comparison is not
between local and global optima but between static versus adaptable systems. Real-time
adaptability is crucial to supply chain management.
- 23 -
Supply chain management
Negotiating Agents for Supply Chain Management
Ye Chen, Yun Peng, Tim Finin, Yannis Labrou, and Scott Cost
Artificial Intelligence: A Modern Approach
Is it an Agent, or just a Program?: A Taxonomy for Autonomous Agents
Botticelli: A Supply Chain Management Agent Designed to Optimize under Uncertainty
Supply chain optimization
M I T - Forum for Supply Chain Innovation