Journal of Industrial Information Integration 42 (2024) 100698

Contents lists available at ScienceDirect

Journal of Industrial Information Integration

journal homepage: www.elsevier.com/locate/jii

Full length article

Towards autonomous supply chains: Definition, characteristics, conceptual

framework, and autonomy levels
Liming Xu ∗, Stephen Mak, Yaniv Proselkov, Alexandra Brintrup
Supply Chain AI Lab, Institute for Manufacturing, Department of Engineering, University of Cambridge, Cambridge CB3 0FS, UK

ARTICLE INFO ABSTRACT

Keywords: Recent global disruptions, such as the COVID-19 pandemic and the ongoing geopolitical conflicts, have pro-
Supply chain management foundly exposed vulnerabilities in traditional supply chains, requiring exploration of more resilient alternatives.
Autonomous supply chain Among various solution offerings, Autonomous supply chains (ASCs) have emerged as key enablers of increased
Autonomy levels
integration and visibility, enhancing flexibility and resilience in turbulent trade environments through the
Conceptual framework
widespread automation of low level decision making. Although ASC solutions have been discussed and trialled
Multi-agent system
over several years, they still lack well-established theoretical foundations. This paper addresses this research
gap by presenting a formal definition of ASC along with its defining characteristics and auxiliary concepts. We
propose a layered conceptual framework, called the MIISI model. An illustrative case study focusing on the
meat supply chain demonstrates an initial ASC implementation based on this conceptual model. Furthermore,
we introduce a seven-level supply chain autonomy reference model, delineating a trajectory towards achieving
full supply chain autonomy. Recognising that this work represents an initial endeavour, we emphasise the need
for continued exploration in this emerging domain. This work is designed to stimulate further research, both
theoretical and technical, and contribute to the continual evolution of ASCs.

1. Introduction chains must evolve to be smarter and integrated [6], digitalised [4],
more automated [8], resilient and agile [9], and structurally adaptable
As the backbone of any business, the supply chain is undoubtedly and flexible [10]. This urgency is highlighted in a recent Nelson-
one of the most important components of industry. However, in recent Hall’s report for Capgemini [9], where over 70% of surveyed supply
years, global supply chains have been severely disrupted by events such chain leaders emphasised the need to enhance agility and operational
as the COVID-19 pandemic, trade wars, and the ongoing geopolitical resilience.
tensions [1–3]. These disruptions have profoundly exposed the vulner- To achieve structural flexibility and resilience, companies must
abilities of traditional supply chain models. The ongoing crises, rapidly adopt a collaborative approach, working stakeholders across the ex-
evolving markets, and the proliferation of distributed information have tended enterprise and integrating information and operations across
collectively driven business leaders to accelerate the transformation of parties in their supply chains [10]. Supply chain executives recognise
their supply chains on enhancing agility and resilience with significant the importance of visibility and information sharing [6,9], but achiev-
digitalisation and reconfiguration initiatives at most multinational cor-
ing effective visibility, especially external visibility, remains challeng-
porations [4]. Moreover, information integration between companies
ing despite increased connectivity and abundant information. Among
across supply chains has becoming increasingly vital in facilitating
others, technological and cultural barriers hinder visibility attainment.
transparency and visibility — key enablers of an ethical and resilient
While inadequate IT infrastructure impacts visibility and collabora-
supply chains [5].
tion, cultural obstacles such as organisational silos, lack of incentives,
Although the old mantra of ‘‘cheaper, faster, better’’ remains rel-
busy schedules, and intellectual property concerns have a significant
evant, it lacks the sophistication needed to address today’s complex
influence [6]. Addressing these human-related barriers is crucial for
business challenges [6–8]. Modern supply chains have thus become
increasingly interconnected, uncertain, and complex. As a result, crises information exchange and integration throughout the supply chain.
in distant regions can now rapidly ripple through supply chains, causing Various research streams have identified that mitigating some of these
significant turbulence and disrupting all entities in the interconnected barriers need the automation of inefficient processes by deploying
supply network. In response, researchers have asserted that supply digital technology within and across organisational boundaries.

∗ Corresponding author.
E-mail address: lx249@cam.ac.uk (L. Xu).

https://doi.org/10.1016/j.jii.2024.100698
Received 1 March 2024; Received in revised form 23 September 2024; Accepted 23 September 2024
Available online 5 October 2024
2452-414X/Crown Copyright © 2024 Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
L. Xu et al. Journal of Industrial Information Integration 42 (2024) 100698

Digital transformation has been widely proposed to enhance au- This paper introduces an ASC conceptual model to frame future
tomation and even autonomy in supply chain management (SCM) [7– research into supply chain autonomy by providing clear definitions
9,11–13]. The concept of automation has continuously evolved in the of each stage of autonomy, By establishing a shared language, we
supply chain and logistics domain. Industry 4.0 technologies such as hope to standardise discussions and methodologies in both academic
AI, IoT, and advanced robotics, are increasingly permeating supply research and practical implementation. The MIISI model offers a path-
chains [14], automating a wide range of processes. This include en- way directing research towards a systems architecture that is expected
terprise resource planning systems that streamline back-office business to approach autonomy. Alongside, the seven-level reference model
functions, material requirements planning systems for manufacturing permits a means to assess and measure the level of autonomy achieved
resources planning, and robotic process automation for handling rou- by such methods. Together, this paper seeks to systematise the ap-
tine, error-prone tasks. However, according to NelsonHall [9], most proach to supply chain autonomy research and demonstrate how to
surveyed managers still consider their supply chain processes (RPA) to implement them with a case study, providing both researchers and
be largely manual rather than automatic, let alone autonomous. industry professionals with a reference framework for implementing
While automation has been a significant trend in supply chains autonomous supply chains.
for years, it has mainly focused on individual functions or processes, While this paper focuses on integration in supply chains, often re-
with less attention given to cross-functional and cross-organisational garded as the backbone of any successful business, it also contributes to
integration and stakeholder involvement. In this paper, we address the broader theme of industrial integration, particularly in information
the integration of supply chain functions and processes, proposing
integration, by facilitating information sharing between distributed and
conceptual frameworks to facilitate this integration within supply chain
decentralised parties across supply chains.
domain.
The rest of this paper is structured as follows. Section 2 presents
The benefits of information and decision-making cross-functions and
the related work on ASC developments. Section 3 presents the formal
organisations have been well recognised in literature, with much of
definition of ASC and its defining characteristics. Section 5 introduces
SCM research focusing on the discourse and mechanisms of supply
the ASC maturity model. Section 6 outlines a case study, implementing
chain integration [10,15–17]. These benefits include reduced wastage
an autonomous meat supply chain. Section 7 discusses the limitations
across the supply chain (e.g., excess supply and under supply), closer
and implications of this work. Finally, Section 8 concludes this paper
and more stable relationships with partners that can lead to more
efficient collaboration, fairer allocation of profits and losses, smoother and discusses future work.
cash flow, and increased resilience to disruptions. However, various
barriers hinder the achievement of such integration, including the costs 2. Related work
of integration, relationship lock-in, and the lack of a technological
framework that facilitate integration. Our work fits into the last cat- In this section, we review the existing literature on ASC develop-
egory, exploring how, and which, integration benefits can be achieved ments. Following from our focus on the theoretical aspects of ASCs, we
through autonomous decision-making processes across organisational primarily cover conceptual advancements, and mention the technical
boundaries. efforts aimed at achieving it.
Automation is crucial for achieving supply chain efficiency [18].
To achieve efficiency gains and build resilience into low-level sup- 2.1. ASC conceptual developments
ply chain operations, automation must extend across the entire sup-
ply chain. We define supply chain automation as the replacement of The initial concept of an ASC can be traced back to the idea of
manual processes with computing systems that establish integrated, ‘‘intelligent products’’ in early 2000s [20,21], which explores estab-
automated processes, facilitating the flow of material, information, and lishing the connectivity of products with their real-time information
finance. For instance, an ASC could be characterised as a connected and through the Auto ID technology. As described in [20], the so-called
self-orchestrating supply chain, capable of forecasting disruptions and intelligent products would then enhance supply chain effectiveness
responding to changes through automated reconfiguration and adjust- thorough new possible functionalities such as self-organised inven-
ments. In an ASC, it is necessary to endow its component systems with tory, real-time routing planning and life-cycle information. While this
advanced computational decision-making capabilities [9,19]. Despite intelligent product-driven supply chain highlights the importance of
the widespread adoption of automation in industry and the existence information and connection, it mainly focuses on product-centric infor-
of the ASC concepts for many years, the development of ASC is still in mation and connectivity, overlooking the broader connectivity between
its nascent stage, both conceptually and technically. supply chain entities and their decision-making capabilities. Butner
This paper thus aims at bridging this gap by focusing on the concep- [6] then envisaged a smarter supply chain, which a supply chain
tual development of ASC. We present a set of theoretical artefacts that capable of autonomous learning and decision making without human
conceptualise ASC developments, forming a prerequisite for implement- intervention. Through a study of nearly 400 in-person conversations
ing ASCs technically. Through a case study, we explore how a prototyp- with global supply chain executives across industries, this article sum-
ical ASC system could be implemented using the proposed conceptual marised five key supply chain challenges and recommends that supply
framework and serving as an example to be used in further technical de- chains must be smart, efficient and demand-driven. The envisioned sup-
velopment. Additionally, we introduce an ASC maturity model, to serve ply chains were characterised by being instrumented, interconnected,
as a reference framework for delineating different levels of supply chain and intelligent, leveraging instruments like sensors, RFID tags, and
autonomy, and for measuring the stages of technological development advanced analytics to shift from a sense-and-response approach to
necessary to achieve supply chain autonomy. predict-and-act strategies. Wu et al. [7] further developed this concept
Specifically, the main contributions of this paper are threefold,
by introducing a smart supply chain and defining six characteristic,
summarised as follows:
including automation, integration, and innovation, in addition to the
• We present a formal definition of ASCs and describe their defining characteristics proposed by Butner [6]. This type of supply chain in-
characteristics. volves process automation, emphasises cross-process collaboration, and
• We present a five-layer conceptual model — the MIISI model — integrates prediction and decision-making with human involvement.
for constructing ASC systems. While Butner [6] laid the groundwork for the initial concept, Wu et al.
• We present a seven-level supply chain autonomy reference model, [7] refined it by conducting a literature review and updating it with
which assesses the autonomy level of an enterprise’s supply chain the advances found in industry and literature in the time between each
and offers a reference trajectory towards achieving full supply paper’s publication. The aim was to investigate various technologies
chain autonomy. relevant to smart SCM. Furthermore, this study briefly discussed the

2
L. Xu et al. Journal of Industrial Information Integration 42 (2024) 100698

essential elements and implementation stages required to develop such services [32] through various AI technologies. The use of emerging
an innovative supply chain model. However, despite presenting close technologies in SCM automation has been since reviewed in [18,33].
ideas, these articles have not thoroughly explored the conceptual or However, these automation efforts are often confined to individual
theoretical aspects of ASCs. tasks or specific supply chains [18,22]. With modern supply chains
Calatayud and Katz [22] proposed to achieve a connected supply becoming more interconnected, effective management requires col-
chain for enhancing risk management. This approach emphasises the laboration and integration across organisational boundaries [16,34].
importance of enabling both physical connectivity and information Thus, integrating a suite of automated functions throughout end-to-
systems connectivity within the supply chain, seamlessly integrating end value chains is the next step. Vast majority of academic discourse
the flow of material, information and finance. Compared to the con- on automation has utilised a multi-agent system (MAS) approach [35–
nectivity proposed in Wong et al. [20], this connectivity extends to a 37], designed for distributed problem solving. Approximately ten years
more generic connection between the physical and the digital. Built after its introduction to the Computer Science literature, first studies
upon this connectivity, Calatayud et al. [11] proposed a new supply on supply chain automation appeared. Despite perceived potential for
chain model referred to as the self-thinking supply chain, characterised various tasks such as supplier selection [38], cost management [39],
by autonomous and predictive capabilities. Similar to Wu et al. [7], this and risk management [40], the multi-agent automation approach re-
study also employed a systematic literature review approach to identify mains under-researched and has limited decision-making capabilities.
the characteristics of a self-thinking supply chain. Rather than explor- Recent AI advancements, particularly in deep reinforcement learn-
ing a broad range of techniques like Wu et al. [7], this study identified ing [31,41] and LLMs such as GPT-4 [42] and Llama [43], could
IoT and AI as the primary technologies enabling cyber–physical con- empower software agents with advanced decision-making capabilities,
nectivity and unmanned, automated decision-making within the supply thus enabling automation in decision makings in complex supply chain
chain. This self-thinking supply chain introduced continuous monitor- scenarios [26].
ing and rapid response mechanisms, resulting in enhanced agility and Moreover, this MAS approach has relatively low adoption in in-
adaptability to manage risks and disruptions effectively. This model is dustries [44,45]. As discussed by Karnouskos et al. [45], aside from
similar in many aspects of ASC. However, similar to previous studies, it business-relevant factors, the lack of terminologies and standards con-
only presents its concepts and characteristics without considering how tributed to this limited acceptance.
the conceptual elements are embodied in practice and they were not This is particularly evident in the emerging domain of ASC, which
designed to systematically conceptualise ASCs. Moreover, existing stud- involves a network of cooperative yet competitive distributed stake-
ies focus on the applications of IoT and/or AI in specific functions or holders. Therefore, a generalised conceptual framework containing es-
specific supply chains, comprehensively reviewed in [7,11,23]. These sential elements such as definitions, terminologies, and technical matu-
do not create a generalisable conceptual framework. rity levels is crucial for both the conceptual and technical development
Nitsche [24] and Nitsche et al. [25] proposed a conceptual frame- of ASC.
work outlining application areas and prerequisites for achieving au-
3. The autonomous supply chain model
tomation in SCM and logistics. Unlike previous studies, this work
provides concrete concepts and emphasises conceptualisation of the
Aside from introducing preliminaries, this section describes key
automating both physical and information processes. Although this
concepts for conceptualising the ASC model, including its definition and
framework provides a common basis for further discussion on automa-
characterising features.
tion between research and practice, it does not provide a portfolio of
conceptual artefacts to systematically guide the technical development
3.1. Preliminaries
of ASCs.
While earlier supply chain models aimed to achieve decision-making
Before dealing with the ASC model, we first briefly describe relevant
autonomy, recent advancements in modern AI technologies, such as
preliminaries: networked supply chains and SCM.
large language models (LLMs) and embodied agents, have paved the
way for the development of true ASCs [26]. Although the benefits
3.1.1. Networked supply chains
of ASC are recognised, the relevant literature on dealing with ASC
A supply chain is a set of business entities involved in the move-
conceptual models is very limited. Most of the existing studies are
ment of materials, services, information, and finance from sources to
practitioner-orientated with surface discussion of the features and
customers, both upstream and downstream [10,46]. Fig. 1 illustrates an
applications of ASCs. For example, BlueYonder [8] underscored the
extended supply chain, with entities that constitute the static aspects of
importance of data and outlined a seven-step approach for companies
a supply chain, along with additional service providers such as third-
to realise true ASCs. Furthermore, based on interviews with 50 supply
party logistics (3PL) and financial providers that offer specific services
chain executives, NelsonHall [9] provided guidelines to assist supply to the stakeholders in the supply chain. The dynamic aspects of a supply
chain leaders in understanding the challenges and potential approaches chain are represented by three key flows: material flow, financial flow,
to ASCs. While offering strategic insights, these practitioner-focused and information flow, forming a complex adaptive network [47,48].
reports lack systematic conceptual or technical frameworks to achieve The supply chain depicted in Fig. 1 is visually simplified as a linear
the envisioned supply chain. We thus in this paper aim to fill this gap, structure and includes only a subset of entities for clarity. However, it
dealing with the theoretical aspects of developing ASCs. is important to note that modern supply chains have evolved into more
complex and large-scale structures characterised by interdependence
2.2. Automation in SCM and connectivity, forming a networked structure [49].

Digitally transforming supply chains has gained momentum during 3.1.2. Supply chain management
the Industry 4.0 revolution [14]. Many modern technologies, such as Effective supply chain design and management is crucial to ev-
IoT, intelligent agents, and robotics, have been adopted to automate ery company [50]. SCM forms the backbone of most economies and
various aspects of SCM. These automation functions encompass a broad successful multinationals today. The significance of managing supply
range of tasks, from mundane activities like picking and packing or- chains can be, at least, traced back to the creation of the assembly line
ders in warehouses using robotic process automation [27,28] to more in the early 20th century. However, the term ‘‘supply chain manage-
complex processes such as demand forecasting and planning [29], ment’’ was coined only a few decades ago by Keith Oliver in a 1982
inventory management [30], delivery optimisation [31], and customer interview with The Financial Times [51], and gained prominence in the

3
L. Xu et al. Journal of Industrial Information Integration 42 (2024) 100698

Fig. 1. Illustration of an extended supply chain.

late 1990s. Since then, various definitions of SCM have been proposed, Just as with autonomous vehicles (AVs), the most crucial feature of
each with a different focus, evolving to align with changing business an ASC is autonomy. The term ‘‘autonomy’’ is defined by the Merriam-
environments and technology advancements [10,46,50,52]. Webster dictionary as ‘‘the quality or state of being self-governing’’.1
To reduce confusion and ambiguity, Mentzer et al. [46] exam- The key aspect here is ‘‘self-governing’’ — the capability to determine,
ined prior SCM concepts and definitions and proposed a consistent conduct and control the actions or behaviours independently, without
means to conceptualise SCM. They proposed the concept of supply external input. When applied to vehicles, autonomy refers to the ve-
chain orientation (SCO), which represents the management philosophy hicle’s capacity to self-govern its driving. Similarly, in supply chains,
that organisations must view SCM activities and coordination from autonomy implies the ability to self-govern its operations. However,
a systemic and strategic perspective. This SCO then forms the basis the distributed and decentralised nature of ASCs differentiates then
for SCM, which is defined as the coordinated set of inter-firm and from monolithic autonomous systems like AVs. Monolithic systems
intra-firm actions implemented to embody this philosophy [46]. By operate within a singular organisational scope without the need to
demarcating SCO from SCM and positing SCO as a major antecedent coordinate with other self-governing entities. Most existing autonomous
of SCM, this definition provides a holistic, strategic, and broader view systems are monolithic. For example, an AV achieves autonomy by
on SCM conceptualisation [53]. This influential definition highlights orchestrating the functions of its diverse components either within the
inter-functional and inter-corporate coordination, aligning with the vehicle or through remote servers, involving a central governing entity.
consistent claim about the strategic importance of integrating upstream A supply chain is a network of interconnected entities, each of
and downstream supply chains [10,15,16,50]. which may have independent decision-making capability. This network
Market environments and technologies have undergone significant is established with the goal of making products available to meet
changes since the development of these concepts nearly two decades customer demand. To achieve this goal, a supply chain requires co-
ago. As described in Section 1, business environments have evolved ordination among the various entities that constitute it, as well as
to become more volatile and customer-centric. Additionally, new ad- coordination among the components within each individual entity.
vanced technologies such as AI, IoT, and advanced robotics have Whether self-organised, intentionally designed, or a combination of
emerged and are now maturing. These technologies are transforming both, supply chains exhibit varying degrees of connections. This in-
SCM, automating numerous supply chain functions. Recently, an article cludes the connections between the entities themselves and between
the internal components within each entity. We classify these connec-
by Lyall et al. [54] even asserted ‘the death of SCM’, arguing that digital
tions into two types: external connections and internal connections. In
technologies are making traditional SCM obsolete.
external connections, we further identify two classes of relationships
However, a closer examination offers an alternative interpreta-
based on the degree of coupling [55] between supply chain entities:
tion: while the implementation methods are changing, the core el-
tight and loose external connections, as described by:
ements of SCM — their strategic nature, customer value creation,
and inter-organisational collaboration — remain pertinent today [10, • Tight External Connection: This denotes a close relationship be-
17,53]. Rather than outdating SCM, technologies are disrupting its tween two entities, characterised by high interdependency. Such
practices, paving the way for new SCM models. The ASC proposed in connections are often intentionally designed and can be consid-
this paper represents such a technology-enabled SCM model, built upon ered ‘‘hard-wired’’ due to their strong and deliberate linkage.
the foundational elements that have long been recognised within SCM. • Loose External Connection: This refers to an arm’s length relation-
ship between two entities, characterised by low interdependency.
3.2. Defining the ASC: Type of connections, structural entities, and its These connections offer higher flexibility, as the entities involved
definition maintain a higher degree of independence.

Tight and loose external connections allow companies to be linked

The ASC concept or similar ideas have been discussed in both either through predefined configurations or through emergent arrange-
academic literature (e.g., [7,11]) and industrial reports ([8,9,54], etc.) ments when structuring supply chains. Unlike external connections,
over the past few years, mostly as derivative of the advancement of which require the establishment of common rules or principles to facil-
AI, IoT and/or advanced robotics. While these interpretations of the itate interaction, internal connections are intrinsic to the organisation
ASC vary, they are commonly characterised by technology-enabled of each individual company and therefore under the same span of
features, including automated processes, continuous monitoring, and control. Internal connections describe the interdependency between
unmanned decision-making. It is worth noting that, as of now, there units within an entity, where all units coordinate coherently towards
is no well-defined definition of ASC in the literature. Therefore, this
section provides a definition of the ASC and describes its defining
1
characteristics and relevant concepts. https://www.merriam-webster.com/dictionary/autonomy.

4
L. Xu et al. Journal of Industrial Information Integration 42 (2024) 100698

Fig. 2. Illustration of the ASC structure.

a common objectives. These connections are defined by the company

itself and reflect its internal organisational structure and dynamics.
To ensure the efficient functioning of an ASC, close collaboration
among all supply chain stakeholders is essential. These stakeholders
include both the entities within the supply chain and their internal
operational divisions. We introduce a new concept, the Structural
Entity, representing the local authority at an external connection point.
Structural entities are defined as follows:

A structural entity is the entity along the supply chain network

that assembles and controls the flows of materials, information and
finance. Structural entities gather the essential data needed by other
entities during decision-making.

Structural entities compose the main structure of an ASC, bridging

external and internal connections, enabling entities of varying auton-
omy levels to interact. These entities often represent firms or groups
of supply chain entities with shared interests. We illustrate an ASC
composed of structural entities and internal units in Fig. 2, in which
structural entities and internal units are denoted by black-filled circles
and unfilled circles, respectively. As shown in Fig. 2, the supply chain
consists of a collection of tightly or loosely connected organisations
denoted by grey circles, each composed of a set of internal units.
Regardless of whether these organisations are autonomous or not, all Fig. 3. The two-dimensional autonomy manifold. The labelled circles, numbered 1 to
structural entities in an ASC must operate coherently without human 4, represent the four regions: intelligence-skewed, automation-skewed, balanced, and
ideal.
intervention. Based on the concepts we presented, we define an ASC as
follows:

An autonomous supply chain (ASC) is a self-governing supply chain 3.3. Characterising the ASC: Two dimensional autonomy and six charac-
built upon intelligence and automation, in which key structural teristics
entities are capable of making and enforcing their decisions with
little or no human intervention.
The previous section presents the definition of an ASC, based on
As presented in the definition above, intelligence and automation the factors of intelligence and automation. This section examines the
are fundamental elements of an ASC. Intelligence refers to the ability manifold formed by these two factors and subsequently defines the
to make decisions and develop solutions to problems in a dynamic and characteristics of ASCs.
uncertain environment [56], while automation involves the capability
to execute solutions automatically. These two aspects form the two 3.3.1. Two dimensional autonomy
dimensions of the autonomy manifold, as shown in Fig. 3. Further
As discussed in Section 3.2, autonomy has two dimensions: intel-
details regarding this manifold are discussed in Section 3.3.1.
ligence and automation. These two dimensions form a bounded space
As a distributed and decentralised autonomous system, an ASC com-
within which the autonomy of a supply chain entity can be evaluated.
prises a set of key structural entities that are autonomous and inclined
The extent of each dimension is constrained by the current level of
to coordinate with each other for mutual benefit. Some of these struc-
technological maturity, but there is potential for expansion through
tural entities may be grouped due to the scale and performance needs
of the supply chain. Therefore, at least some representative structural technological advancement.
entities must possess autonomy. These autonomous structural entities The resulting autonomy manifold is shown in Fig. 3. As depicted,
function as connection points, managing materials, information, and the autonomy manifold can be divided into four regions: intelligence-
financial flows. To ensure smooth operation of an ASC, well-defined skewed, automation-skewed, balanced, and ideal. These regions are
communication and interaction mechanisms among entities must be annotated with circled numbers from one to four. Dashed horizontal
in place. These mechanisms ensure both entity autonomy and coordi- and vertical lines indicate the sufficiency degree of technological maturity
nated behaviour across all entities. They specify how and what entities in their respective dimensions. Entities located within the skewed
communicate, including languages, protocols, and terminologies for regions have achieved greater technological development in one dimen-
effective communication and interaction. sion than the other, resulting in a compromised level of autonomy. For

5
L. Xu et al. Journal of Industrial Information Integration 42 (2024) 100698

instance, an intelligence-skewed entity can automatically propose solu- on specific technologies; rather, they focus on describing how tech-
tions but might struggle to effectively execute these solutions without nological advancements enhance and enrich ASC implementations.
human assistance. Conversely, an entity skewed towards automation The first four characteristics are downward dependent, with higher-
possesses the capability to perform planned actions but may lack the layer characteristics facilitated by the lower ones. In contrast, the
ability to generate appropriate actionable plans. In an ASC, entities, final two characteristics, automated and intelligent, are relatively less
particularly structural entities, should strive to position themselves in interdependent; they are manifest features of an ASC.
the spectrum of dark regions (the diagonal area in Fig. 3), where they
achieve a relative balance between intelligence and automation. The 4. A conceptual framework: the MIISI model
darkest region (denoted by circled four) represents the ideal region,
situated above both technological maturity lines. Entities in this region In this section, we present a layered conceptual model — the MIISI
attain sufficient autonomy, with well-balanced and sufficient unmanned model — for constructing ASCs, detailing the main functions and data
decision-making and execution capability. Importantly, within the bal- associated with each layer within our proposed model.
anced and ideal regions, the two dimensions may not necessarily be The previously defined bottom-up, layered six characteristics sug-
equal; one dimension might outpace the other within a certain range gest potential conceptual partitions for a functional ASC system that
and time frame, allowing entities to take measures to rebalance and encompasses diverse entities with varying degree of autonomy. It is
enhance their autonomy. thus natural to translate the descriptions of each characteristic into
corresponding abstraction layers. The two characteristics, ‘‘intelligent’’
3.3.2. The six characteristics and ‘‘automated’’, have a relative independence from each other and
The core defining characteristic of an ASC is autonomy, which directly contribute to the manifestation of an ASC. Therefore, they
is further characterised by the two dimensions depicted in Fig. 3. are grouped into a single layer termed the ‘‘Manifestation’’ layer. The
Previous studies, such as Butner [6], Wu et al. [7] and Calatayud remaining characteristics each relates to a separate layer, collectively
et al. [11], envisaged future supply chain models and summarised facilitating the manifestation layer. This partition results in a five-
their features, as described in Section 2. These characteristics often layer conceptual model, each building upon the previous one. As
focus on technology-enabled aspects such as continuous monitoring, illustrated in Fig. 4, these layers from top to bottom are: Manifes-
data connectivity, process integration and automation, and predictive tation, Integration, Interconnection, Standardisation, Instrumentation.
analytics. While these features are critical for the operating of an ASC, This conceptual model is referred to as the MIISI model, an acronym
they were initially conceptualised in response to the development of for the five layers in top-down order.
corresponding emerging technologies and were often treated as isolated Each layer in the MIISI model has distinct overarching functions,
concepts, without fully considering their interrelationships. each building upon the layers beneath it. The MIISI model assigns
We herein propose six bottom-up layered characteristics for defin- different levels of abstraction and scope of view to each layer, with
ing ASCs: instrumented, standardised, interconnected, integrated, au- higher layers being more abstract and having a broader scope of view.
tomated, and intelligent. These six characteristics are defined without The bottom layer deals with concrete objects and has a local view,
being tied to specific implementation technologies. Furthermore, they whereas the top layer has an integrated, global view of the supply
are logically separate but functionally connected. These characteristics chain, focusing on specific operational processes. The middle layers
are: facilitate bridging the bottom and top layers by standardising data flow
and processes and establishing connections with entities.
(1) Instrumented: Data connectivity is facilitated through installed The main functions and the data handled in each layer are detailed
instruments. Various devices (sensors, actuators, tags, etc.) are as follows:
employed by supply chain entities for real-time data collection
and transmission, tracking, monitoring, and analytics. (1) Instrumentation Layer: This layer is responsible for collecting and
aggregating raw data necessary to monitor the environment and
(2) Standardised: Common standards and rules govern processes
establish connectivity. It involves utilising a diverse array of
related to inter-entity interaction. Procedures, standards, pro-
equipment, devices, protocols, and systems to generate, capture,
tocols, guidelines, regulatory frameworks, and data formats for
store, and transmit data. This layer also specifies the scheme
data exchange between entities are established.
and metadata of any attributes (type, format, variety, quality,
(3) Interconnected: Supply chain entities, internal units within enti-
purpose, and owner, etc.) of objects needed to enable automated
ties, and other objects and systems that support supply chain
functions or processes. Furthermore, it supports two types of
operations are connected. This extensive interconnection allows
connectivity:
entities to interact with others under certain guidelines and/or
protocols. • Local Connectivity: Enables communication between phys-
(4) Integrated: A broad arc of integration is implemented, enabling ical and/or digital things within an organisation.
entities to coordinate with upstream and/or downstream coun- • Wide Connectivity: Allows connections to external net-
terparts in shared operational activities. works owned by other entities.
(5a) Automated: Automation is enabled and leveraged. Workflows are
designed to allow machines to perform efficiently and effec- Both types of connectivity support intra- and inter-company
tively, and tasks and processes in the supply chains are auto- information sharing, which is a critical aspect for enabling ASC
mated. functionalities.
(5b) Intelligent : Key entities possess autonomous decision-making ca- (2) Standardisation Layer: This layer is responsible for setting stan-
pabilities in the function they are tasked to manage. Entities can dards and establishing procedures for data sharing and ex-
reason based on their context and propose suitable actions with change, thus enabling process automation. It defines data
minimal human intervention. schemes, data interchange formats, and agreed-upon procedures
to facilitate automated data exchange between entities involved
While some of the characteristics presented above have been in- in tasks. Two types of data exchange are considered: internal
troduced in prior work such as Butner [6] and Wu et al. [7], they data exchange within an organisation and external data ex-
have been adopted and redefined to suit the context of ASCs. We change with entities outside the organisation. Data for each type
organise these characteristics into a bottom-up layered structure to of exchange needs to be defined in a standardised, machine-
provide a framework for characterising ASCs. They are not reliant readable format to enable automated data manipulation and

6
L. Xu et al. Journal of Industrial Information Integration 42 (2024) 100698

Fig. 4. Illustration of the proposed MIISI model.

interpretation. However, entities may have different data shar- necessary for these different types of connections. This layer
ing policies regarding data security and privacy. This layer facilitates the customised formation of supply chain structures
identifies processes needed to be standardised for interconnec- and enables coordinated behaviour and joint decision-making
tion and interoperability, defining agreed-upon procedures for among participants, thereby supporting the implementation of
performing processes related to the three flows. Standardised various integration strategies.
processes and data exchange procedures facilitate the connection (4) Integration Layer: This layer coordinates and aligns the
and coordination between entities. behaviours of interconnected entities in the supply chain, fa-
(3) Interconnection Layer: This layer is responsible for managing cilitating collaboration within individual entities and between
connections between entities, physical objects, or digital ob- various participants in the supply chain, with the ultimate goal
jects across the supply chain. It identifies entities, establishes, of enabling them to operate cohesively as a unified whole. It
manages, and terminates connections between two or more par- ensures that decentralised and geographically dispersed stake-
ties. Key functionalities of this layer include registration and holders collaborate effectively towards common goals that ben-
naming services, authentication, message delivery, and error efit the entire supply chain. Moreover, it supports different types
detection and correction. Utilisation of shared infrastructure ser- of integration with varying breadths, ranging from the nar-
vices might be necessary for implementing registration, naming, rowest inward-facing integration to the broadest outward-facing
and reliable message delivery. Protocols are also established integration [15]. To achieve this, this layer needs to establish
within this layer to regulate connection. This layer facilitates protocols for negotiation, define agreed-upon procedures for
the connection of entities, physical objects, and digital objects optimising collaborative processes, and provide standardised
across the supply chain as required, allowing ad hoc or dynamic methods for preparing shareable data (such as data encryption,
configuration of supply chain networks. compression, and conversion) to enable a wide range of inte-
This layer considers three types of connections: ad hoc, tempo- gration strategies. Additionally, this layer converts data into a
rary, and established. These three connections are categorised format suitable for the layer immediately above it. Leveraging
based on the purpose of the connection and the relationship the capabilities of the interconnection layer, this layer ensures
between the connected parties, detailed as below: collaborative decision-making and enables coherent operational
activities among entities.
• Ad Hoc Connection: This involves connecting two un-
(5) Manifestation Layer: This layer is responsible for managing the
known parties to address specific needs or problems,
day-to-day operations of an ASC. It includes a set of intelli-
whether for short or long-term purposes. These connec-
gent and automated applications, devices, and machinery that
tions are often spontaneous and solution-focused.
interact with humans and perform operations within the sup-
• Temporary Connection: This refers to the connection be-
ply chain. However, this layer is not designed to take over
tween two unrecognised parties, but they connect only
every business process or activity. Instead, its primary focus is
for short-term purposes. These connections are typically
on managing the automated movement of three key elements:
formed for specific projects or short-term collaborations.
information, finance, and product. Two categories of flow are
• Established Connection: This involves mutually trusted
defined in this layer based on the conveyance manner: digital
parties that may have an established contract or a long-
and physical. Digital flow relates to intangible items that can
standing collaboration. These connections are stable and
be represented in the form of bits, while physical flow involves
ongoing, often involving formal agreements and long-term
tangible objects that exist in the form of atoms. Consequently, the
cooperation.
flows of information and finance are considered digital flows,
Various design considerations, including authentication, connec- as they can be represented by bits and transmitted along the
tion modes, data security and privacy, among others, may be supply chain using digital communication methods such as the

7
L. Xu et al. Journal of Industrial Information Integration 42 (2024) 100698

Internet. On the other hand, the movement of tangible products Model (CMM) [59,60] are two widely recognised frameworks. The
constitutes a physical flow, where their movement needs to be TRLs, originating from NASA in the 1970s, include a nine-level scale
completed through transportation systems. It is important to used to assess the maturity of specific technologies. The CMM serves as
note that intangible products, which are not physical in nature a process maturity framework, delineating five levels to measure the
and often exist digitally (e.g., licences and software), can also be maturity of an organisation’s software processes. Both models provide
represented as bits and moved through the Internet. Therefore, an incremental path for continuous improvement and have been gained
their movement is considered as a special form of digital flow adoption across various industries.
within this layer. Additionally, models exist for appraising the developmental stages
This layer comprises a range of self-regulating utilities that of autonomous systems. Dumitrescu et al. [61] outlined five develop-
embody the dimensions of autonomy described earlier (see Sec- ment stages of technical systems, such as robots, vehicles, machines and
tion 3.3.1). These utilities have the capability to make decisions software, leading towards autonomous systems. For a concise review
autonomously, including managing supply chain planning across in English, one can refer to Nitsche [24]. Notably, the Society of
various time horizons, proposing real-time solutions for emer- Automotive Engineers (SAE) developed a classification system [62]
gent events, and predicting and responding to contingent events. that includes six levels of vehicle autonomy, ranging from 0 (fully
Additionally, they can execute courses of action to implement manual) to 5 (fully autonomous). This system has gained recognition as
decisions, such as rerouting to avoid traffic congestion, replen- a standard for assessing the degree of driving automation in vehicles.
ishing inventory, and selecting alternative suppliers. Entities Other works include the three levels of object intelligence [19] and the
equipped with intelligent and automated systems can then col- three phrases for implementing a smart supply chain [7].
laborate to plan, control, and execute the flow of both digital Drawing inspiration from these models, particularly the SAE driving
bits and physical atoms along the supply chain. This layer mainly automation levels, we define seven distinct levels for evaluating supply
handles operational and application-related data. chain autonomy. These levels, called Supply Chain Autonomy Levels
(SCALs), as shown in Fig. 5, range from level 0 (L0: fully manual) to 6
Digital flows, such as information and financial flows, are often
(L6: fully autonomous), providing a phased process towards full supply
considered to be driven by physical flows and therefore secondary.
chain autonomy. The proposed seven SCALs are detailed as follows:
However, we argue that ASCs should treat digital flows with equal
importance to physical flows, ensuring the efficient and seamless move- L0 No Automation: supply chains at this level are operated entirely
ment of data across the supply chain. Prioritising digital flows does not manually. Human operators are solely responsible for manag-
undermine the importance of physical flows; instead, it enhances the ing and performing all aspects of the supply chain operations.
management of physical flows. Even if physical flows cannot be entirely Although supportive systems or tools may be in use, human
automated and requires manual interventions, human involvement can intervention remains constant throughout the process.
align it with the fully automated digital flow, thus addressing issues L1 Function Automation: At this level, certain functions within the
related to human errors and inefficiencies in moving data. supply chain have been automated. However, these automated
This model stratifies an ASC into five abstract and conceptual layers. functions are often disconnected from each other and do not
The lower three layers form the foundational infrastructure of ASCs, form a continuous process. Human operators are responsible
enabling connections and facilitating coordination among supply chain for connecting manual and automated functions throughout the
entities. The top two layers are focused on the everyday management process. While humans may be on standby during the execution
and operation of ASCs, involving applications responsible for various of certain automated functions, their active involvement is still
functions, with minimal or no human intervention. This conceptual prevalent, and they retain overall control over supply chain
model describes an unmanned SCM system but does not necessarily operations.
imply the complete replacement of human workers with machines in L2 Process Automation: At this level, one or more processes in the
the ASC era. While automation may reduce the numbers of frontline supply chain are automated and streamlined. A process com-
human workers [57], it also creates opportunities for new types of jobs prises a sequence of functions that are executed in a predeter-
or ‘‘job of tomorrow’’ [58]. Machines will play a crucial role in ASCs, mined order. Although hands-off operations are allowed during
but humans will remain integral. Their role will shift from mundane automated processes, human operators must remain vigilant and
and repetitive tasks to more strategic activities, such as guiding ma- ready to intervene upon request. The key distinction between
chines in supplier selection. Achieving full supply chain autonomy is a L2 and L1 is the integration of multiple automated functions,
complex journey that requires both technological advances and man- creating a connected series of actions that automatically execute
agerial efforts. The next section discusses the stages towards realising a specific process. Supply chains at this level can perform auto-
a fully autonomous supply chain. mated processes and respond precisely to predefined situations.
These supply chains exhibit fewer human errors, enhanced ef-
5. The seven supply chain autonomy levels ficiency, and reduced reliance on frontline operators. However,
additional personnel may be required to oversee the execution
Achieving supply chain autonomy is not an all-or-nothing propo- of automated processes. While humans maintain overall control
sition; rather, it is a gradual process that unfolds over time. Compa- of supply chain operations, their role primarily involves super-
nies progressively integrate automated functionalities into their supply vision as they standby during the execution of processes for the
chains. A maturity reference model outlining the stages of supply chain majority of the time.
autonomy development would benefit all stakeholders by providing a L3 Holistic Automation: When the operation of a supply chain be-
shared framework for discussing the technical development of supply comes nearly entirely automated, it progresses to L3. At this
chain autonomy within both academic and industrial spheres. Addition- level, all major processes within the supply chain are automated
ally, it can be used as a benchmark for comparison, aiding executives and connected to work coherently with human intervention.
in making informed decisions. However, such models are currently Compared to L2, an L3 supply chain encompasses a greater num-
lacking in both academia and industry. This section seeks to bridge this ber of automated processes, even involving actions that require
gap by introducing a consistent framework for assessing the level of external entities’ participation. Automation permeates nearly
supply chain autonomy. every aspect of the supply chain, but it primarily focuses on
Several models exist for evaluating technological developments. executing operational decisions rather than automating decision
The Technology Readiness Levels (TRLs) and the Capacity Maturity making. Supply chains at this level are highly automated and

8
L. Xu et al. Journal of Industrial Information Integration 42 (2024) 100698

Fig. 5. The seven supply chain autonomy levels (SCALs).

capable of automatically performing the majority of tasks under These seven levels represent a trajectory leading towards full supply
human guidance. The need for front workers is further reduced, chain autonomy (see Fig. 5 for an illustration of these levels). This
although more individuals may be involved in supervising and trajectory comprises two distinct phases: automation, encompassing the
monitoring the execution of automated processes. Humans still initial four levels (L1–L4), and autonomisation, which includes the three
control decision making, particularly on tactical and strategic higher levels (L4–L7) in the SCAL reference model. As the level of
levels. autonomy increases, the supply chain incorporates more automated
L4 Limited Autonomy: The transition from L3 to L4 signifies a sub- functions and acquires self-decision-making capabilities, becoming in-
stantial technological advancement, although it may appear sub- creasingly capable of handling more complex scenarios with less human
tle or negligible from a human perspective. The defining char- intervention. Throughout this trajectory, human involvement decreases
acteristic of an L4 supply chain is its capability for self-decision- while machine involvement increases, as illustrated by the shrinking
making. Specifically, the supply chain can autonomously per- light grey area and expanding dark grey area in Fig. 5, While the
ceive its environment and make decisions based on its accu- differentiation between consecutive levels within each phrase is based
mulated knowledge, without requiring human input. However, on the degree of automation, the distinction between the two phrases
this self-decision-making is confined to specific functions, human is substantial. During the automation phase, humans mainly operate
intervention is still essential beyond these functions. For exam- the supply chain, whereas in the autonomisation phase, the supply
ple, the supply chain can decide to replenish inventory without chain starts to manage its own operations. Exactly assigning an existing
human instruction in anticipation of an upcoming demand surge, supply chain to a specific level may not always be possible due to
but it may need human guidance in selecting appropriate sup- complexity of real-world supply chains. This spectrum of supply chain
pliers. At L4, the supply chain shows a considerable level of autonomy provides a lens through which companies can examine the
automation and begins to acquire limited autonomy, with a current stages of automation development in their supply chains and
limited self-learning capability to expand its knowledge base. Al- guide their automation strategies.
though human involvement is still necessary, it mostly revolves
around making decisions in complex situations and monitoring 6. A case study: Autonomous meat supply chain
the execution of automated processes.
L5 Conditional Autonomy: A supply chain reaches L5 when it can In this section, we provide a brief overview of an ASC proto-
perform all functions autonomously without human intervention type implementation, demonstrated through a case study involving the
under certain conditions. This supply chain can operate inde- development of a platform for automated meat procurement and whole-
pendently in predefined circumstances, while humans remain sale. For reasons of brevity, we exclude technical details and focus on
alert to respond to any requests for intervention in unexpected demonstrating how the proposed conceptual model — the MIISI model
events. At this level, humans maintain a high-level control over — guides the design and development of a concrete ASC. More details
the supply chain operations and can be in a state of ‘‘mind off’’ of this system, including its design and technical implementations, can
for extended periods. be found in Xu et al. [63].
L6 Full Autonomy: An L6 supply chain achieve complete automa-
tion, possessing full self-learning and self-decision-making ca- 6.1. Design
pabilities, and can operate autonomously for extended periods.
This supply chain can handle all situations with minimal or This particular application involves a local meat company called
even zero human attention or interaction, even in unanticipated the Cambridge Meat Company (CMC). The CMC specialises in the
situations. At this level, human involvement is kept to a minimal wholesale procurement of meat, such as chicken, beef, and lamb,
level. Human focus may shift to strategic activities that are not and its subsequent distribution to local restaurants. It is important to
necessarily directly involved in the everyday operation of the emphasise that the CMC is a hypothetical company created solely for
supply chain. illustrative purposes. Its supply chain includes several key participants:

9
L. Xu et al. Journal of Industrial Information Integration 42 (2024) 100698

Fig. 6. The physical and digital flows of the meat supply chain under study.

• Suppliers: Entities such as meat producers or suppliers who pro- Logistics occurs after the ordering process, and negotiation and
vide meat products. coordination are integrated into these two processes to resolve conflicts
• Wholesaler: The CMC itself, acting as an intermediary that pur- and achieve coherent behaviours. In this system, a set of ‘‘autonomous’’
chases meat from suppliers and sells it to retailers. agents is developed to act on behalf of these stakeholders shown in
• Retailers: Businesses such restaurants or local stores that purchase Fig. 6. These agents collectively manage the operations of the meat
meat products from the CMC. supply chain, which include selecting suppliers, placing orders, nego-
• Logistics Companies: Companies responsible for managing the lo- tiating delivery options, invoicing, and monitoring logistics. With the
gistics of transporting goods, specifically meat, within the supply exception of order placement, where users must complete the order
chain. form and initiate the process by clicking the ‘Launch’ button, all actions
• 3PL Providers: External logistics service companies that offer de- within this supply chain are executed automatically.
livery services to logistics companies, transporting meat products
from their source to assigned destination. 6.2. Showcase

This supply chain consists of two main processes: replenishment, We demonstrate the functionality of this ASC system by showcasing
where the CMC procures meat from suppliers to restock its inventory, an automated meat procurement process. Fig. 7 shows the startup
and wholesale, where the CMC acts as a wholesaler, providing meat interface of the developed system. As shown in Fig. 7, all four panels
products to retailers, such as local restaurants. Both processes need of this system are in their initial state with empty content and default
logistics services for order fulfilment, with the sellers responsible for values. For example, the system can order three types of meat: chicken,
handling delivery arrangements. beef, and lamb, each with a default ordering quantity of 50 kg. Fig. 8,
To achieve full automation of these two processes, multiple deci- on the other hand, presents the system’s interface during its running,
sions must be made autonomously. These decisions include creating specifically, in the middle of a delivery during the CMC’s replenishment
proposals, accepting or declining proposals, and selecting appropriate process.
delivery options. In real-world scenarios, these decision-making pro- We showcase the ASC system using the default settings. After click-
cesses are often highly complex, requiring one to consider numerous ing the ‘‘Launch’’ button, the system starts the replenishment process
short- and long-term factors. For illustration purposes, we simplify autonomously, without requiring human intervention. Specifically, this
these decisions by only considering a set of simple predefined rules. process involves procuring specified quantities of meat products (50 kg
Using these simplified settings, we implemented an autonomous meat each of pork, lamb, and beef) from selected suppliers and transporting
supply chain system (see Fig. 7) following the MIISI model introduced them to the CMC using the logistics services provided by a designed
in Section 4. 3PL provider. This process include various automated functions, such
This autonomous meat supply chain is simulated, with no actual as supplier selection, inventory updating, logistics arrangement, trans-
purchases and transportation occurring. This supply chain is shown in portation monitoring, and delivery service assessment. These functions
Fig. 6, where its two types of flows are represented by dashed arrows are executed in a predefined order, triggered when prerequisite condi-
and solid arrows, denoting potential interactions between stakeholders. tions are met. The automation involved in this process is facilitated by
To develop a system capable of automating meat procurement and the backend multi-agent system, where autonomous agents act on be-
wholesale, we employ a MAS approach [35,36], which is well-suited half of the respective supply chain entities they represent. These agents
for connecting distributed entities and thus facilitating supply chain in- including meat suppliers, the CMC, logistics companies, and 3PLs.
tegration. Due to space limitation, we omit the implementation details These agents represent structural entities (described in Section 3.2),
in this paper, as they are beyond its scope. gathering necessary data and making decisions.
The resulting platform is a web-based system accessible via a web This system also includes a wholesale process, in which the CMC
browser. As shown in Fig. 7, the system’s interface consists of four supplies meat products to local retailers. While there are variations in
panels: ordering (top left), transport monitoring (central), agent chat the selection of 3PLs, the wholesale process shares many similarities
room (right), and product ambient condition monitoring (bottom). with the replenishment process. Therefore, we omit the description of
These panels correspond to the three basic procedures in a supply this process. When connected sequentially, these two automated pro-
chain: cesses represent a common goal of supply chain management, facilitat-
ing the movement of products from the source, through an intermediary
• Ordering : This panel handles purchasing-related functions such (the CMC), to their ultimate destination.
as supplier selection, order placing, and order confirmation. The Although these two processes are simulated and do not involve
ordering panel is the interface for ordering. the physical movement of products, this system effectively illustrates
• Logistics Monitoring : This pertains to shipping goods, tracking their the architectural guidelines for developing ASCs. This system can be
locations, and monitoring ambient conditions. This process is broadly separated into the five layers described in the MISSI model.
depicted in the central and bottom panels of the interface. In the instrumentation layer, a set of three sensors were deployed to
• Negotiation and Coordination: Supply chain stakeholders inter- gather real-time environmental data, offering the upper layers insights
act with one another to resolve disagreements and collaborate into the surrounding conditions of the products. Subsequently, these
effectively, as shown in the agent chat room panel. data were processed and transformed into standardised formats for

10
L. Xu et al. Journal of Industrial Information Integration 42 (2024) 100698

Fig. 7. A screenshot of the system interface in its startup state, with all four panels are in their initial state with empty content and default values.

Fig. 8. A screenshot of the system interface during a replenishment process. In the central map area, a delivery vehicle is indicated by a red solid circle, positioned at around the
midpoint of its journey along the route outlined in blue. A infotip with black background is visible, displaying the delivery tracking number. In the bottom area of the interface,
real-time IoT readings for temperature, humidity, and illumination are shown, providing insights into the ambient conditions of the products being transported. In the right side
of the interface, the chat history between all involved agents is displayed, visualising the processes of negotiation and coordination among stakeholders in the supply chain.

various purposes, such as controlling product ambient conditions and structure entities to achieve a coherent and streamlined process. In
assessing delivery services, which may involve data exchange. These this system, this is achieved through interaction among agents and
data handling activities constitute the standardisation layer. Standardis- their internal units via the Contract Net protocol [64] — a task-
ation enables interconnection and interoperability. This system adopted sharing protocol in multi-agent systems. Built upon these four layers,
a MAS approach to provides a mechanism for connecting distributed automation and intelligence-related technologies can be adopted to
and heterogeneous entities and objects. This approach enables connec- realise autonomous supply chain functions. These functions collectively
tivity and, consequently, facilitates data exchange and collaboration. form the manifestation layer. In this system, functions within this layer,
Integration is further facilitated by allowing the functions of different such as supplier selection, are more ‘‘automated’’ than ‘‘intelligent’’

11
L. Xu et al. Journal of Industrial Information Integration 42 (2024) 100698

since their decision-making relies on predefined rules. However, these • Trustworthiness: ASCs entail collaboration among multiple entities
functions are interconnected and coordinated to logically construct responsible for managing the supply chain, each driven by its
a loosely-coupled yet coherent system. By following the five layer own objectives and potentially engaged in competition with oth-
conceptual design, the developed ASC system, albeit with limitations, ers. Agents that represent these entities, including those utilising
demonstrate the feasibility of applying the proposed theories for the machine learning models such as BERT [65], GPT-3 [66], and
design and development of ASC systems. Llama [43] for decision making, must be reliable and trustworthy.
Their collaborative actions should result in mutual benefits for all
7. Discussion and considerations supply chain participants, cultivating trust among them.
• Platform Neutrality: ASC system development may rely on co-
The previous section presents a cast study, an ASC system imple- created or vendor platforms for delivering various services. These
mentation employing a MAS approach. This MAS approach provides platforms must maintain neutrality, treating all parties impar-
a framework that logically links physically distributed entities within tially and refraining from favouring any specific entity This en-
a supply chain, forming an integrated structure. Within this struc- sures fairness and transparency within the supply chain ecosys-
ture, representative agents interact with others through messaging tem, fostering healthy competition and collaboration among par-
to ensure coherence. It is important to note that this system falls ticipants.
considerably short of being considered as a complete autonomous
system, let alone a fully-fledged realistic ASC system. Instead, it is an To address these challenges outlined above, an ASC system must
experimental proof-of-concept system based on the MIISI model. Its implement a reliable and resilient architecture, along with effective
implementation mainly focuses on three infrastructure layers of the interaction mechanisms. Additionally, it should establish appropriate
model: instrumentation, interconnection, and integration, which are data access and authorisation policies.
enabled by the adopted MAS approach. The remaining two layers, In ASC development, managerial and cultural aspects are crucial
standardisation and manifestation, receive comparatively less attention considerations. The transition to ASCs involves automating many man-
in this implementation. ual processes, which has raised concerns about potential job losses
This work is an initial attempt to create a realistic ASC system across the supply chain [8]. However, the proliferation of connected
based on the proposed conceptual model. According to the autonomy and smart supply chain technologies is transforming employees into
manifold presented in Fig. 3, this system falls within the automation- strategic decision-makers, as highlighted in a report by BlueYonder
skewed region. However, both the intelligence and automation levels of [8]. With routine tasks increasingly managed by autonomous solutions,
the system are currently low. The intelligence dimension mainly focuses employees can redirect their focus towards strategic activities or new
on decision-making. In current system, decision-making is based on a roles. This transformation results in a shift in the nature of required job
straightforward set of rules that consider only a limited set of factors. roles. Companies should proactively design programmes to empower
For example, one of these rules involves taking price into account employees, providing them with the skills and expertise to work effec-
when selecting potential suppliers. To effectively tackling real-world tively with autonomous systems (or intelligent agents) and adapt to the
challenges in the supply chain domain, it is imperative to enhance evolving demands of their roles.
the decision-making capabilities, enabling them to make informed Moreover, a survey conducted by Chain [67] highlights the growing
decisions and propose appropriate courses of actions. Regarding the importance of leadership and strategic management skills in the fu-
automation dimension, the system mainly focuses on automating data ture. Given the multitude of autonomous functions and systems within
flow and process execution. However, this automation is constrained, ASCs, supply chain managers must prioritise these skills in strategic
only occurring along predefined data transmission paths. Additionally, planning. The ASC roadmap should also include small and middle-
as discussed in Section 6, the system does not consider financial and sized enterprises (SMEs), which play a significant role in the supply
product flow, which are an integral part of a real-world supply chain. networks of large companies. Unlike larger companies with substan-
A successful ASC system shall comprehensively tackle both of these tial R&D budgets, SMEs often have limited resources for investing
two dimensions. This means enhancing both decision-making and au- in new technologies. Therefore, it is crucial to develop appropriate
tomation capabilities, extending autonomy across a broad spectrum technologies and offer governmental and policy support to ensure SMEs
of supply chain activities, which may include automatic context-based can participate in ASC development. Additionally, legal and regulatory
decision making and the automated handling of physical flow through frameworks must be established to govern ASC-related activities, in-
advanced robotics. cluding intellectual property (IP), human–robot collaboration, and data
In addition to the technological aspects relevant to ASC develop- security and privacy.
ment, advancing the ASC agenda needs addressing other crucial design
considerations. These include:
8. Conclusion and future work
• Cyber Security: The adoption of technologies to enhance connec-
tivity along the supply chain exposes companies to increased Modern supply chains have become increasingly networked, which
cyber risks. Organisations must adopt new and effective risk man- not only heightens the cascading of risks but also complicates risk man-
agement strategies and tools to proactively prepare for potential agement [6,10]. Moreover, the turbulent and uncertain environments
cyber threats [14]. ASC systems, being highly interconnected, in recent years exacerbated this situation. To adapt to disruptions,
are particularly vulnerable to cyber attacks, which can trigger supply chains must evolve to be digitalised, more automated, smarter,
disruptions propagating throughout the entire supply chain. Con- and structurally flexible and resilient. These requirements call for the
sequently, ASC development requires proactive measures to safe- new supply chain model: the autonomous supply chain (ASC), a self-
guard against various cyber attacks. governing supply chain with minimal or even no human intervention.
• Data Security and Privacy: The ASC era will witness the collec- Equipped with automation and intelligent decision-making capabilities,
tion, access, and exchange of vast and diverse data of various ASCs can self-adjust and promptly respond to uncertainties.
types. These data may include sensitive information critical to a As a concept that has been evolved over many years, systematic
company’s core competitive advantage or restricted to authorised studies on the conceptual development of ASCs remain lacking, es-
individuals and entities. As such, the design of ASC must incor- pecially in contrast to the technical explorations that adopt various
porate robust mechanisms to ensure data security and privacy, modern technologies (AI, IoT, robotics, etc.) to digitally enhance supply
protecting it from unauthorised access and malicious threats. chain tasks. This paper thus aims to bridge this knowledge gap.

12
L. Xu et al. Journal of Industrial Information Integration 42 (2024) 100698

Specifically, we have provided a formal definition of ASC, based Acknowledgements

on a newly defined concept of structural entities. To better concep-
tualise ASC, we presented a two-dimensional autonomy manifold to This work was funded by the Research England’s Connecting Ca-
examine the autonomy development of a supply chain entity. Building pability Fund (grant number: CCf18-7157): Promoting the Internet of
upon these characteristics, we proposed a conceptual model, the MIISI Things via Collaboration between HEIs and Industry (Pitch-In) and the
model, which conceptualises ASCs as composed of five abstract and EPSRC Connected Everything Network Plus under grant
downward dependent layers: Manifestation, Integration, Interconnection, EP/S036113/1.
Standardisation, and Instrumentation. Importantly, this model is agnostic
to specific technologies, allowing for different implementations in each References
layer. Additionally, we introduced the auxiliary concepts of internal
connection and external connection, with the latter categorised into [1] R.B. Handfield, G. Graham, L. Burns, Corona virus, tariffs, trade wars and supply
three distinct types within the context of ASCs: ad hoc, temporary, and chain evolutionary design, Int. J. Oper. Prod. Manage. (2020).
[2] W.C. Shih, Global supply chains in a post-pandemic world, Harv. Bus. Rev.
established connections.
(2020) URL https://hbr.org/2020/09/global-supply-chains-in-a-post-pandemic-
Achieving a fully autonomous supply chain is an incremental journey world.
that requires both technological and managerial efforts. To support this [3] K. Kilpatrick, Supply chain implications of the Russia-Ukraine conflict, 2022,
journey, we introduced a seven-level supply chain autonomy reference URL https://www2.deloitte.com/uk/en/insights/focus/supply-chain/supply-
chain-war-russia-ukraine.html. (Accessed 23 September 2022).
model designed to assess a company’s current supply chain autonomy
[4] B.L. MacCarthy, D. Ivanov, The Digital Supply Chain, Elsevier, 2022.
level (SCAL). These seven levels are a structured roadmap, providing [5] P.K.R. Maddikunta, Q.-V. Pham, B. Prabadevi, N. Deepa, K. Dev, T.R. Gadekallu,
companies with a reference trajectory for progressively achieving full R. Ruby, M. Liyanage, Industry 5.0: A survey on enabling technologies and
supply chain autonomy. In addition to these theoretical efforts, we potential applications, J. Ind. Inf. Integr. 26 (2022) 100257.
briefly presented an ASC implementation case study on the meat supply [6] K. Butner, The smarter supply chain of the future, Strateg. Leadersh. 38 (1)
(2010) 22–31.
chain. This implementation, employing the MAS approach, showcases [7] L. Wu, X. Yue, A. Jin, D.C. Yen, Smart supply chain management: a review and
the realisation of an ASC system based on the proposed conceptual implications for future research, Int. J. Logist. Manag. 27 (2) (2016) 395–417.
framework. [8] BlueYonder, The autonomous supply chain: seven steps to get there,
This study is among the first to formally address the conceptual 2020, URL https://blueyonder.com/knowledge-center/collateral/7-steps-to-an-
autonomous-supply-chain-white-paper. (Accessed 03 March 2022).
development of ASCs, even as the term has gained attention for many
[9] NelsonHall, Moving to an autonomous supply chain, 2021, URL . (Accessed 12
years in news and industry reports. The conceptual artefacts presented January 2022).
in this paper, including the ASC definition, MIISI model, auxiliary [10] M. Christopher, Logistics & Supply Chain Management, Pearson UK, 2016.
concepts, and SCAL reference model, together provide a valuable foun- [11] A. Calatayud, J. Mangan, M. Christopher, The self-thinking supply chain, Supply
dation for future research and implementations on ASCs. We anticipate Chain Manag.: Int. J. 24 (1) (2019) 22–38.
[12] R. Geissbauer, S. Schrauf, J. Schneider, M. Hermans, Connected and au-
that this study will inspire further research, both conceptual and tech- tonomous supply chain ecosysyem 2025, 2020, URL https://www.pwc.com/gx/
nical, thereby contributing to the continual evolution of ASCs. As the en/industries/industrial-manufacturing/digital-supply-chain.html. (Accessed 03
backbone of any business, supply chains with autonomy can certainly March 2022).
facilitate full information integration between multiple distributed sup- [13] C. Kuntze, S. Lal, K. Seibert, Launching the journey to autonomous supply chain
planning, 2020, URL https://www.mckinsey.com/business-functions/operations/
ply chain entities. We thus expect this study to also contribute to the
our-insights/launching-the-journey-to-autonomous-supply-chain-planning. (Ac-
industrial information integration community by providing conceptual cessed 03 March 2022).
frameworks for integrating information through the lens of supply [14] WEF, Impact of the fourth industrial revolution on supply chain, 2017,
chains. URL https://www.weforum.org/whitepapers/impact-of-the-fourth-industrial-
revolution-on-supply-chains. (Accessed 07 June 2024).
This work is an initial effort towards supply chain autonomy, and
[15] M.T. Frohlich, R. Westbrook, Arcs of integration: an international study of supply
future work is needed to investigate this emerging area. Our future chain strategies, J. Oper. Manage. 19 (2) (2001) 185–200.
work will be threefold: (1) we will continue to interrogate the pro- [16] P. Childerhouse, D.R. Towill, Arcs of supply chain integration, Int. J. Prod. Res.
posed conceptual constructs and conduct qualitative studies on their 49 (24) (2011) 7441–7468.
use in real-world SCM practices; (2) we will extend our research by [17] G.C. Stevens, M. Johnson, Integrating the supply chain . . . 25 years on, Int. J.
Phys. Distrib. Logist. Manage. 46 (1) (2016) 19–42.
implementing ASCs at higher SCALs in new case studies across more
[18] L. Xu, S. Mak, A. Brintrup, Will bots take over the supply chain? Revisiting
industrial sectors; and (3) we will investigate the potential of leveraging agent-based supply chain automation, Int. J. Prod. Econ. 241 (2021) 108279.
large language models to enhance the decision-making capabilities of [19] A. Brintrup, D. McFarlane, D. Ranasinghe, T.S. López, K. Owens, Will intelligent
agents within the context of ASCs. assets take off? Toward self-serving aircraft, IEEE Intell. Syst. 26 (03) (2011)
66–75.
[20] C.Y. Wong, D. McFarlane, A.A. Zaharudin, V. Agarwal, The intelligent product
CRediT authorship contribution statement driven supply chain, in: IEEE International Conference on Systems, Man and
Cybernetics, Vol. 4, IEEE, 2002, pp. 6–pp.
[21] D. McFarlane, V. Marík, P. Valckenaers, Guest editors’ introduction: Intelligent
Liming Xu: Writing – original draft, Methodology, Investigation,
control in the manufacturing supply chain, IEEE Intell. Syst. 20 (1) (2005) 24–26.
Data curation. Stephen Mak: Writing – review & editing. Yaniv Pros- [22] A. Calatayud, R. Katz, The Connected Supply Chain: Enhancing Risk Management
elkov: Writing – review & editing. Alexandra Brintrup: Writing – in a Changing World, Technical Report IDB-DP-508, Inter-American Development
review & editing, Supervision, Funding acquisition. Bank (IDB), 2017.
[23] M. Ben-Daya, E. Hassini, Z. Bahroun, Internet of things and supply chain
management: a literature review, Int. J. Prod. Res. 57 (15–16) (2019)
Declaration of competing interest 4719–4742.
[24] B. Nitsche, Exploring the potentials of automation in logistics and supply chain
The authors declare that they have no known competing finan- management: Paving the way for autonomous supply chains, 5, (3) 2021, p. 51,

cial interests or personal relationships that could have appeared to

[25] B. Nitsche, F. Straube, M. Wirth, Application areas and antecedents of automation
influence the work reported in this paper. in logistics and supply chain management: a conceptual framework, in: Supply
Chain Forum: An International Journal, Vol. 22, Taylor & Francis, 2021, pp.
223–239.
Data availability
[26] L. Xu, S. Almahri, S. Mak, A. Brintrup, Multi-agent systems and foundation
models enable autonomous supply chains: Opportunities and challenges, 2023,
Data will be made available on request. Available at SSRN 4695075. in press in INCOM’24.

13
L. Xu et al. Journal of Industrial Information Integration 42 (2024) 100698

[27] J. Ribeiro, R. Lima, T. Eckhardt, S. Paiva, Robotic process automation and [47] T.Y. Choi, K.J. Dooley, M. Rungtusanatham, Supply networks and complex
artificial intelligence in industry 4.0–a literature review, Procedia Comput. Sci. adaptive systems: control versus emergence, J. Oper. Manag. 19 (3) (2001)
181 (2021) 51–58. 351–366.
[28] N. Nalgozhina, R. Uskenbayeva, Automating hybrid business processes with RPA: [48] Y. Proselkov, J. Zhang, L. Xu, E. Hofmann, T.Y. Choi, D. Rogers, A. Brintrup,
optimizing warehouse management, Procedia Comput. Sci. 231 (2024) 391–396. Financial ripple effect in complex adaptive supply networks: an agent-based
[29] M.Z. Babai, J.E. Boylan, B. Rostami-Tabar, Demand forecasting in supply chains: model, Int. J. Prod. Res. 62 (3) (2024) 823–845.
a review of aggregation and hierarchical approaches, Int. J. Prod. Res. 60 (1) [49] A. Brintrup, Y. Wang, A. Tiwari, Supply networks as complex systems: a
(2022) 324–348. network-science-based characterization, IEEE Syst. J. 11 (4) (2015) 2170–2181.
[30] G.P. Cachon, M. Fisher, Supply chain inventory management and the value of [50] G.C. Stevens, Integrating the supply chain, Int. J. Phys. Distrib. Mater. Manage.
shared information, Manag. Sci. 46 (8) (2000) 1032–1048. (1989).
[31] S. Mak, L. Xu, T. Pearce, M. Ostroumov, A. Brintrup, Fair collaborative vehicle [51] R.K. Oliver, M. Webber, Supply chain management: Logistics catches up with
routing: A deep multi-agent reinforcement learning approach, Transp. Res. C 157 strategy, outlook, cit. Christopher, MC (1992), logistics, the strategic issue, 1982.
(2023) 104376. [52] M.C. Cooper, D.M. Lambert, J.D. Pagh, Supply chain management: more than a
[32] L. Cui, S. Huang, F. Wei, C. Tan, C. Duan, M. Zhou, Superagent: A customer new name for logistics, Int. J. Logist. Manag. 8 (1) (1997) 1–14.
service chatbot for e-commerce websites, in: Proceedings of ACL 2017, System [53] S. Min, Z.G. Zacharia, C.D. Smith, Defining supply chain management: in the
Demonstrations, 2017, pp. 97–102. past, present, and future, J. Bus. Logist. 40 (1) (2019) 44–55.
[33] C. Hendriksen, Artificial intelligence for supply chain management: Disruptive [54] A. Lyall, P. Mercier, S. Gstettner, The death of supply chain management,
innovation or innovative disruption? J. Supply Chain Manag. 59 (3) (2023) Harv. Bus. Rev. (2018) URL https://hbr.org/2018/06/the-death-of-supply-chain-
65–76. management.
[34] A. Khanuja, R.K. Jain, Supply chain integration: a review of enablers, dimensions [55] J.S. Brown, S. Durchslag, J. Hagel III, Loosening up: How process networks
and performance, Benchmark.: Int. J. 27 (1) (2020) 264–301. unlock the power of specialization, McKinsey Q. 2 (2002) 59–69.
[35] M. Wooldridge, N.R. Jennings, Intelligent agents: Theory and practice, Knowl. [56] J.P. Gunderson, L.F. Gunderson, Intelligence ≠ autonomy ≠ capability, in:
Eng. Rev. 10 (2) (1995) 115–152. Proceedings of the 2004 Performance Metrics for Intelligent Systems Workshop,
[36] M.S. Fox, M. Barbuceanu, R. Teigen, Agent-oriented supply-chain management, Gaithhersburg, MD, US, 2004.
in: Information-Based Manufacturing, Springer, 2001, pp. 81–104. [57] J. Hawksworth, R. Berriman, S. Goel, Will robots really steal our jobs?
[37] L. Xu, Y. Proselkov, S. Schoepf, D. Minarsch, M. Minaricova, A. Brintrup, An international analysis of the potential long term impact of automa-
Implementation of autonomous supply chains for digital twinning: a multi-agent tion, 2018, URL https://www.pwc.co.uk/services/economics-policy/insights/the-
approach, IFAC-PapersOnLine 56 (2) (2023) 11076–11081. impact-of-automation-on-jobs.html.
[38] P. Ghadimi, F.G. Toosi, C. Heavey, A multi-agent systems approach for sus- [58] N. Orduña, Why robots won’t steel your job, Harv. Bus. Rev. (2021) URL
tainable supplier selection and order allocation in a partnership supply chain, https://hbr.org/2021/03/why-robots-wont-steal-your-job.
European J. Oper. Res. 269 (1) (2018) 286–301. [59] W.S. Humphrey, Characterizing the software process: a maturity framework, IEEE
[39] J. Fu, Y. Fu, An adaptive multi-agent system for cost collaborative management Softw. 5 (2) (1988) 73–79.
in supply chains, Eng. Appl. Artif. Intell. 44 (2015) 91–100. [60] M.C. Paulk, B. Curtis, M.B. Chrissis, C.V. Weber, Capability maturity model,
[40] L.A. Bearzotti, E. Salomone, O.J. Chiotti, An autonomous multi-agent approach version 1.1, IEEE Softw. 10 (4) (1993) 18–27.
to supply chain event management, Int. J. Prod. Econ. 135 (1) (2012) 468–478. [61] R. Dumitrescu, J. Gausemeier, P. Slusallek, S. Cieslik, G. Demme, T. Falkowski,
[41] D. Silver, A. Huang, C.J. Maddison, A. Guez, L. Sifre, G. Van Den Driessche, J. H. Hoffmann, S. Kadner, F. Reinhart, T. Westermann, et al., Studie" Autonome
Schrittwieser, I. Antonoglou, V. Panneershelvam, M. Lanctot, et al., Mastering Systeme", Technical report, Studien zum deutschen Innovationssystem, 2018.
the game of go with deep neural networks and tree search, Nature 529 (7587) [62] SAE J3016, Taxonomy and definitions for terms related to driving automation
(2016) 484–489. systems for on-road motor vehicles, 2021, URL https://www.sae.org/standards/
[42] OpenAI, GPT-4 technical report, 2024, arXiv:2303.08774. content/j3016_202104/. (Accessed 08 June 2022).
[43] H. Touvron, T. Lavril, G. Izacard, X. Martinet, M.-A. Lachaux, T. Lacroix, B. [63] L. Xu, S. Mak, M. Minaricova, A. Brintrup, On implementing autonomous
Rozière, N. Goyal, E. Hambro, F. Azhar, et al., Llama: Open and efficient supply chains: a multi-agent system approach, Computers in Industry 161 (2024)
foundation language models, 2023, arXiv preprint arXiv:2302.13971. 104120.
[44] V. Marik, V. Gorodetsky, P. Skobelev, Multi-agent technology for industrial [64] R.G. Smith, R. Davis, Frameworks for cooperation in distributed problem solving,
applications: Barriers and trends, in: 2020 IEEE International Conference on IEEE Trans. Syst. Man Cybern. 11 (1) (1981) 61–70.
Systems, Man, and Cybernetics, SMC’20, IEEE, 2020, pp. 1980–1987. [65] J. Devlin, M.-W. Chang, K. Lee, K. Toutanova, Bert: Pre-training of deep
[45] S. Karnouskos, P. Leitao, L. Ribeiro, A.W. Colombo, Industrial agents as a bidirectional transformers for language understanding, 2018, arXiv preprint
key enabler for realizing industrial cyber-physical systems: Multiagent systems arXiv:1810.04805.
entering industry 4.0, IEEE Ind. Electron. Mag. 14 (3) (2020) 18–32. [66] C. Raffel, N. Shazeer, A. Roberts, K. Lee, S. Narang, M. Matena, Y. Zhou, W.
[46] J.T. Mentzer, W. DeWitt, J.S. Keebler, S. Min, N.W. Nix, C.D. Smith, Z.G. Li, P.J. Liu, Exploring the limits of transfer learning with a unified text-to-text
Zacharia, Defining supply chain management, J. Bus. Logist. 22 (2) (2001) 1–25. transformer, J. Mach. Learn. Res. 21 (1) (2020) 5485–5551.
[67] D.S. Chain, The supply chain talent shortage: from gap to crisis, 2017,
URL http://dhl.lookbookhq.com/ao_thought-leadership_talent-gap/research-
report_talent-shortage-from-gap-to-crisis. (Accessed 18 August 2022).

14