Applications of artificial intelligence in supply chain management and exporting:
a systematic literature review
University of Amsterdam Faculty of Economics and Business Bachelor of Science Business Administration Specialisation: Management in the Digital Age
By Farida Assef - 12526746 29th of June 2022
Supervised by Gezim Hoxha
Statement of originality
This document is written by Farida Assef who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
UvA Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
With the growing popularity of AI and its applications, it is important to recognise that it has a transformational impact on business practices. This paper analyses the full extent of existing AI research in the areas of SCM and exporting and provides areas for practical AI applications within SCM. A systematic literature review was conducted in which various bibliometric analyses were performed to showcase the current state of research and identify seminal works, prevalent themes, and influential authors. The introduction highlights the importance of this research and its practical relevance as well as generates insight into the most relevant AI tools.
It is then followed by the methodology which contains an overview of the data specifications.
Bibliometric analyses were conducted on the data retrieved through Scopus which includes descriptive analysis, bibliographical coupling, citation network analysis, co-citation analysis, keyword co-occurrence network analysis, and a global citation score analysis. Based on the core themes and relevant works, five SCM areas in which practical AI applications that have transformational potential were identified and highlighted. Overall, this allows the research to provide a framework for AI implementation within business practices with a focus on how those applications can generate and enhance business value.
Keywords: Artificial Intelligence (AI) · Machine Learning (ML) · Supply Chain Management (SCM) · Supply Chain (SC) · Automation · Operations Management (OM) · Systematic Literature Review (SLR) · Agent-based systems · Genetic algorithms · Expert systems
Table of Contents
Introduction ... 4
Methodology ... 7
Analyses and synthesis ... 11
Descriptive analysis ... 11
Bibliographic coupling ... 12
Citation network analysis ... 14
Co-citation analysis ... 14
Keyword co-occurrence network analysis ... 17
Global citation score analysis ... 19
Practical applications ... 20
Discussion ... 25
Limitations ... 26
Future research direction ... 26
Conclusion ... 27
References ... 29
Supply chains (SCs) are integral parts of the operations of every business. They are complex and integrative business activities that handle the entire process of making and selling commercial goods and services (Grimshaw, 2020). Supply chains comprehensively oversee all associated stages from sourcing and procurement to production and logistics. As supply chains handle an integral bracket of business activities, this makes successful supply chain management (SCM) integral to a company’s success. Nowadays SCM, especially regarding global supply chains, is continuously being affected by predictable and unpredictable events (Baryannis et al., 2018). The negative effects of such circumstances are only enhanced by the higher uncertainty, higher supply chain risks, elevated external environmental turbulence, and an increase in global competition (Min, 2009). In culmination, these instances have great potential in threatening a company’s profitability and overall business operations. Thus, it is vital to identify SCM excellence as an essential part of a business’s strategy and ability to create value. Often the limiting factor of SCM success in an organisation relates to the visibility of the entire SC, risk mitigation abilities, and the lack of available information (Helo & Hao, 2021).
Recently, a new approach that is being undertaken regarding SCM is the digitalisation of business processes to achieve a total system competitive advantage (Tammela et al., 2008).
Currently, there are many IT systems that are implemented within organisations to support SC-related business activities. Such systems include ERP (Enterprise Resource Planning), MES (Manufacturing Execution System), PPC (Production Planning & Control), and many others (Haas, 2019). Whilst these IT systems do prove to bring benefits regarding SCM that cannot be contended with, their mass adoption has led to firms not gaining competitive advantages from their adoption (Hill, 2021). Additionally, they do exhibit a ceiling regarding their abilities due to the fact that they provide fragmented solutions to a business operation that optimally requires a comprehensive overview. Thus, an optimal system would need to encompass the dynamic nature of SCs regarding rapidly changing demands, changes in business processes, and unforeseen circumstances that current IT systems are not capable of providing (Helo & Hao, 2021). Therefore, this leads to the awareness that successful SCM, especially in the context of international activities, requires rapid and effective business response systems to allow supply chains to operate with the highest efficiency in all related business activities and processes. It is then essential to employ more advanced IT systems that have the ability to manage multi-level and highly variable problems (Seyedghorban et al., 2019).
A system that has such potential is artificial intelligence (AI). Artificial intelligence aims to develop nonbiological systems, such as computers and machines, to perform tasks that would typically require human intelligence (Dogru & Keskin, 2020). Whilst AI is not a new field of study, especially within the context of intentional business activities, it has predominately been utilised as a decision-making tool with only recent technological developments in AI uncovering its increasingly vast set of beneficial applications within SCM.
As such, SCM has been identified by Martínez-López & Casillas (2013) as a field that will profit the most from AI applications as using AI can lead to problem-solving with higher accuracy and increased speed all while utilising an extensive amount of inputs (Toorajipour et al., 2021). These beneficial prospects are due to SCM requiring comprehensive, complex, and interconnected decision-making which is precisely where the capabilities of AI reside.
Ultimately, this would allow companies to shift business operations from remote monitoring to actively controlling, monitoring, and optimising their business operations to reap improved efficiency and functionally and more importantly, competitive advantages through the strategic use of autonomous AI-based systems (Kohtamäki et al., 2019).
Furthermore, AI is likely to have numerous impacts on a firm’s internationalisation efforts. Whilst current theoretical evidence in this domain is quite limited, generally, AI has the ability to increase productivity growth, which will result in increased economic growth and provides new opportunities for international trade (Meltzer, 2019). Furthermore, Meltzer (2018) has investigated the application of AI in international business, albeit in a limited scope, to identify that businesses can use AI tools to help develop and manage their global value chains as it can improve forecasts and better mitigate risks associated with international trade.
Moreover, it was seen that integrating AI tools within international trade negotiations can provide increased chances of business success as they can analyse economic trajectories offered by each negotiating partner all whilst integrating different assumptions and accounting for trade barriers, therefore, deducing a business’s optimal trajectory. Such undoubtedly presents opportunities for businesses to generate and capture value as well as competitive advantages.
Current research about applicable AI tools regarding SCM has mostly been focused on three main AI sub-fields (Min, 2009). These are agent-based systems, genetic algorithms, and expert systems. The promise of each sub-field relates to its ability in recognising business patterns, learn business phenomena, seek information, and analyse multi-dimensional data sources.
The first AI tool of note regarding SCM is agent-based systems which approaches problem-solving by dividing a decision problem into multiple sub-problems (Axtell et al.,
2001). It then proceeds to solve these sub-problems by utilising autonomous agents (who can take the form of individual or collective entities). Such autonomous agents are characterised by acting in a manner of bounded rationality. Generally, most agent-based systems are composed of five components (1) numerous agents located at various scales; (2) decision-making heuristics; (3) learning rules or adaptive processes; (4) an interactive topology; and (5) an environment. These five components allow for the creation of a software that can analyse and assess how changes in individual behaviours would trigger changes in the overall system. Based on such, agent-based models have the capability to be employed to handle logistics planning (Satapathy et al. 1998), aggregated demand planning and forecasting (Yu et al. 2002; Liang and Huang 2006), business-to-business negotiations (Ito and Saleh 2000; Lenar and Sobecki 2007), and outsourcing relationship management (Logan, 2000).
The second AI tool is genetic algorithms (GA) which lies within the branch of evolution programs (Min, 2009). GA work by mimicking the process of biological evolution to solve problems and model evolutions and progressions of existing systems (Mitchell, 1995). This AI tool brings forth unique advantages as it is adaptive and innovative in nature due to its ability to search among an extensive number of possible solutions and obtain the most efficient outcome given the system in place. Its most predominant use has been in solving combinatorial optimisation problems to determine the fitness of a solution in the context of a specific problem.
Primarily, the application of genetic algorithms has been focally implemented within businesses to address network design problems, location-allocation problems and even extends to logistics and purchasing problems (Tam, 1998), inventory control (Disney et al., 2000; Haq and Kannan, 2006), and supplier selection (Rao, 2007).
The last AI tool that this paper will focus on is expert systems which refers to a computing system that mimics the cognitive skills and decision-making abilities of human experts (Jackson, 1990). Essentially, they are designed to be able to solve complex problems by using reasoning. This AI tool is divided into two subsystems: the interference engine and the knowledge base. The interference engine applies known rules and facts to deduce new facts all whilst providing an explanation. The knowledge base simply represents the facts and rules.
Since expert systems provide an explanation of their deductive reasoning, it is generally observed to be much more easily understood and thus makes it applicable to solve practical SC problems. The application of expert systems has been focused on increasing productivity in managing logistics (Eom and Karathanos, 1996), providing a sophisticated forecasting method (DeLurgio, 1997), and conducting supplier evaluations (Kwong et al., 2002)
Overall, whilst there has been an increased interest in research concerning the benefits AI can bring to business operations and SCM, this area is largely unexplored within the context of international activities, especially exporting. In recent years, organisations have exhibited an increase in their interest in adopting AI to generate and gain business value and competitive advantages (Enholm et al., 2021). This is seen through a recent study that identified that 80%
of organisations see AI as a strategic opportunity, with 85% also seeing it as a means to gain competitive advantages. However, despite the apparent interest and enthusiasm, organisations are struggling to adopt and leverage AI within their existing operations due to a lack of coherent understanding regarding how AI tools can create value-generating mechanisms. Thus, there is an identifiable research gap that does not directly address these aspects, especially regarding international firm activities. For this reason, a promising research question to address these latencies and provide a conceptual framework that firms can beneficially utilise is:
How can firms implement value-creating AI tools in regard to supply chain management within the context of exporting activities?
The overarching aim of this paper is to identify value-generating AI tools through the conduction of a systematic literature review to identify which tools and applications can be advantageous in managing SCM within the domain of exporting activities. To do so, this paper will be structured as follows. The introduction will be followed by Section 2 which outlines the utilised methodology and describes the search method used. Section 3 titled analyses and synthesis will contain descriptive and bibliographical analyses conducted on the retrieved data set. This will be followed by Section 4 outlining the identified application areas of AI tools.
Such will be followed by Section 5 being the discussion and finally succeeded by the conclusion in Section 6.
The data used in this paper will be gathered using Scopus which is one of the biggest and most commonly used scholarly citation database. Scopus covers journals in a wide variety of top-level subject fields. All journals in its database are reviewed to ensure high-quality (Kulkarni, 2009). The reason for the use of Scopus is due to its various advantages. It is generally seen to be easy to navigate even for novice users (Burnham, 2006). It allows for the ability to search forwards and backwards from any given citation. Additionally, enhanced utility can be achieved through its various and multidisciplinary source searches (Falagas et al., 2007).
All withstanding, it is important to also acknowledge the limitation of Scopus which includes limited content availability, lack of inclusion of diverse source types such as books, and an inadequate mechanism for distinguishing among authors (Wilder & Walters, 2021).
The five-phase process outlined by Denyer and Tranfield (2009) will be utilised as an approach for the methodology of this paper. The phases will be as follows:
(1) A pilot study to gain a deeper understanding of the current literature, as well as construct the criteria for literature selection
(2) Locating the studies will be done through the selected Scopus database and the formulated search strings
(3) Study selection and evaluation will allow for the narrowing down of the various identified taxonomies through the utilisation of inclusion/exclusion criteria
(4) The analysis and synthesis phase will analyse relationships and connections of sources by utilising descriptive and bibliographical analyses
(5) Finally, the results will be reported
Figure 1. Depicting the phases of the systematic literature review.
This first step allows the development of an extensive understanding of the already existing academic literature and any related fields. The concept of ‘artificial intelligence’ can be identified as being the central and primary search string utilised in this paper as it covers all related concepts and can be utilised in combination with other relevant keywords to obtain desired data. The pilot study is essential in getting a grasp of relevant literature and establishing useful inclusion and exclusion criteria.
Locating the studies
To locate all the relevant studies that will provide the base of the retrieved data, it is imperative to select an appropriate search engine and search strings. As mentioned for the reasons above, the research engine used is Scopus. The identification of relevant keywords is essential, and it is necessary to be specific. Given the primary search string that correlates to the research’s relevance, a multitude of search strings will be used that will incorporate
‘artificial intelligence’ AND ‘keyword’. In the identification of keywords, it is important to incorporate keywords that are typically associated and included when considering a comprehensive definition of SCM and exporting activities.
It is important to identify reliable search terms as the inclusion or exclusion of certain keywords is critical and can change displayed results. If available, it is important to include synonyms of relevant concepts to widen the scope of the obtained data as well as include the most commonly used terms. It is also important to recognise that the selection of keywords can be a subjective process, therefore, it is important to interpret the results with such a provision in mind. Additionally, an important consideration is to stray away from terms that are too narrow or specific as such could restrict the overall number of academic sources retrieved.
Table 1. Outlining the search strings used and number of articles retrieved.
Number of retrieved articles in the pilot study
Number of retrieved articles after search parameter
Number of retrieved articles after the application of inclusion/exclusion criteria ( "artificial
intelligence" OR "ai" AND
"supply chain" ) 2,147 264 62
( "artificial intelligence" OR
"ai" AND "supply chain management" OR 'scm' )
1,038 119 34
intelligence" OR "ai" AND '
export' OR 'exporting' ) 543 26 11
Study selection and evaluation
The identified keywords were used in combination to initially obtain a broad scope of data to ensure that academic articles incorporating different taxonomies were collected. The identified keywords highlighted in Table 1 were used as search terms in Scopus in early April 2022. Whilst this initial scope is quite broad resulting in the identification of 3728 articles, it is important to narrow down and limit the identified studies through inclusion and exclusion
criteria. The first search selection criterion outlines the document type in which only documents identified as articles were maintained. Other types of documents such as reviews, conference papers, conference reviews, books, book chapters, etc… were excluded from the search. The second selection criterion targets the life span of the literature in which only sources written from 2008 until 2022 were included. The reason for this is due to this timeframe reflecting the emergence of a large number of new trends and applications that contribute significantly to this research topic. The third search selection criterion involved the parameter pertaining to the subject area. It is of great importance to narrow down the identified literature according to the subject area that explicitly regards this research. For this reason, the relevant subject areas were identified to be business, management and accounting as well as economics, econometrics and finance. The final search selection criterion outlined the language of the academic literature in which only articles and reviews written in English were included.
Subsequently, it is important to establish inclusion and exclusion criteria that will evaluate and screen the retrieved data to ensure high quality. There are four main criteria used for the exclusion of sources based on the quality metric. Firstly, search engine reasoning was used to exclude any papers that had their title, abstract and keywords in English, however, the publication itself was not in the English language. Secondly, a without full-text criterion was used to exclude any academic sources that did not contain sufficient full-text to be assessed or included an insufficient abstract. Thirdly, a non-related criterion was used to exclude any articles in which their research direction did not align with that of this paper. Lastly, a loosely related exclusion criterion was utilised to eliminate papers that did not have a focus on the review, problem-solving ability, tools, or applications of AI.
Additionally, there are two main criteria used for the inclusion of sources based on the quality metric. This ensures that specific articles covering the goal of this research paper are included in the analysis. The first criterion can be identified as partially related which outlines credible sources that place a focus on AI as the main tool, perspective, or focus of a study and its associated academic contribution. The second criterion is a closely related one which would ensure that the collected academic sources are explicitly related to AI.
The initial literature review sampled 3728 articles which were retrieved through the three search strings (see Table 1). After the initial search parameters were adjusted, this resulted in a total of 409 identified articles. Following, the application of the inclusion and exclusion criteria to ensure high-quality and relevant sources were identified was done manually by reading document titles and abstracts whilst accounting for the prevalence of relevant keywords. This left 107 articles as the final sample size that will undergo analysis.
Analyses and synthesis
After collecting and refining the scope of the gathered articles to account for relevance and quality, various analyses need to be conducted on the retrieved dataset. The overarching aim of this is to provide better insight into the gathered data as well as identify prevalent themes, influential publications and authors, and identify and describe relationships and connections present within the data in a multitude of ways.
Regarding bibliographical analyses, VOSviewer was used. The choice of VOSviewer over other bibliometrics analysis tools was predominately due to it having a user-friendly interface, being able to import and analyse Scopus data, and having increased functionality and ability to handle large maps (van Eck & Waltman, 2009).
Publication date. Whilst initially parameters were set regarding the years that the articles were retrieved, which outlined articles published between 2008 and 2022. Figure 2 shows the number of articles published by year. It can be seen that from 2008 until 2018 research within this field was consistently low. Recently, however, there’s been a very significant and sharp increase in this topic with a peak in 2021. Regarding this, is important to note that the number of sources published has not yet reached the definitive amount in 2022 at the time this paper is being written.
Figure 2. Number of articles published by year.
0 10 20 30 40
2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022
Subject area categorisation. The descriptive statistics shown in Figure 3 show the documents published by subject area regarding the articles gathered for this research. The majority of published articles are accounted for by the Business, Management, and Accounting subject area which comprises 37%. The second most prevalent and significant subject area is Decision Sciences and Engineering which account for 19%
and 14% of total published articles respectively. Overall, the articles retrieved cover a total of 10 subject areas.
Country/territory of origin. There are a total of 46 countries and territories affiliated with the 107 articles extracted for analysis. There are three prominent countries regarding the amount of published research. Firstly, the United Kingdom covers a total of 27 documents. Secondly, India has a total of 23 published articles. And thirdly, comes the United States with 21 published documents. The other 43 countries account for the rest of the written articles.
Bibliographic coupling is a similarity measure that uses citation analysis to establish a similarity relationship between documents (Martyn, 1964). Bibliographic coupling refers to the fact that two published works reference or cite a common third work in their bibliography list.
Therefore, this leads to the establishment of both works belonging to a related subject matter.
Additionally, this type of citation analysis provides a coupling strength or a total link strength (TLS) which refers to the amount of citations or references the two analysed documents share.
This means that the higher the coupling strength between two publications, the higher the number of intersections regarding their respective bibliography list. For this research, two bibliographical coupling methods will be conducted. The first will pertain to the country of origin and the second will regard organisations.
Country of origin. For the bibliographical coupling based on country of origin, to ensure that only relevant data was procured specified parameters was set. The maximum number of countries per document was limited to 25. The minimum number of a document of a country was set to 5, further, the minimum number of citations was set to 0. Based on these specifications, a total of 7 countries met the defined criteria. The results showcased in Table 2 outline the most influential countries regarding AI tools in relation to SCM activities. Overall,
Business, Management and Accounting Decision Sciences
Engineering Computer Science Social Sciences
Economics, Econometrics and Finance Environmental Science Psychology Arts and Humanities Energy
Figure 3. Subject area categorisation
the top three leading countries appear to be India, the United Kingdom, and the United States respectively.
Table 2. Bibliographic coupling of countries ranked according to their TLS.
Rank Country Documents Links TLS
1 India 23 6 3227
2 United Kingdom 27 6 2936
3 United States 21 6 1758
4 France 9 6 1477
5 Denmark 5 6 819
6 Brazil 6 6 703
7 China 6 6 128
Organisations. To ensure the procurement of solely relevant data for this measure, explicit specifications were set. The maximum number of organisations per document was set to 25. The minimum number of documents of an organisation was set to 2. Further, the minimum number of citations of an organisation is set at 0. Based on these restrictions, out of 258 total organisations, a total of 6 met the threshold and are outlined in Table 3. This type of analysis showcases the amount of influence of each organisation. In this regard, the most influential organisations can be seen to be the school of innovation, design and engineering of Mälardalens University, the department of operations management of the Copenhagen Business School, and the school of business of Maynooth University each having the same TLS of 516.
Table 3. Bibliographic coupling of organisations ranked according to their TLS.
Rank Organisation Documents Links TLS
1 School of innovation, design and engineering (Mälardalens University) 2 5 516 2 Department of operations management (Copenhagen Business School) 2 5 516
3 School of business (Maynooth University) 2 5 516
4 Institute of economics (Hungarian University) 2 5 306
5 Guildhall school of business and law (London Metropolitan University) 2 5 306
6 Cadi Ayyad University 2 5 24
Citation network analysis
A citation network analysis (CNA) is a review method that aims to map the scientific structure of a field of research as a function of citation practices (McLaren & Bruner, 2022). It can be stated that published research articles that receive a great number of citations in comparison to others would have a greater degree of prominence in the field being investigated.
A citation network comprises of both nodes and links. Generally, the nodes represent publications, and links depict the presence of a citation between connected publications (Orr et al., 2020). The size of each node indicates the number of citations normalised by age. The distance present between one node and the other is calculated through a citation analysis algorithm to determine the relatedness of each publication based on the number of times they cited each other. Additionally, the colours pertaining to the nodes and links represent the respective discipline they belong to.
The overarching purpose of a CNA is threefold. Firstly, it manages to identify the impact that a specific published piece of work has had by identifying which other authors have cited or used their work seminally in their own papers (The Ohio State University, 2022). Secondly, it allows an exploration of a field or topic by highlighting seminal work in that area. Thirdly, it can allow for the establishment of the impact of particular authors by their contributions to a certain discipline.
In a CNA it is not uncommon for certain connected components to have a varying number of nodes. Due to the nature of CNA as described above, for a clearer overview and analysis, it would be important to exclude any isolated nodes from the analysis. This is strengthened by the fact that a citation analysis can only be applied to connected components.
Therefore, Figure 4 displays the result of the CNA which takes into account the 62 excluded isolated nodes. This leaves the target set of 42 connected items. Additionally, Table 4 shows the breakdown of the authors, document titles and links of the top 10 most cited documents in the network which displays the most influential and impactful pieces of work within this subject area.
According to Small (1973), a co-citation analysis refers to a semantic similarity measure that analyses citation relationships in publications. Generally, it is defined as the frequency in which two publications are cited together by other documents. It can be seen that the more co- citations two documents receive together, the higher their co-citation strength. Such a relationship would signify that these two documents are related and share similarities (Surwase
Figure 4. Showing the citation network analysis of the connected component.
Table 4. Showing information regarding the top 10 most cited documents.
Rank Author(s) Title Citations Links
1 Kusiak (2018) Smart manufacturing 473 2
2 Dwivedi et al. (2021) Artificial Intelligence (AI): Multidisciplinary perspectives on emerging
challenges, opportunities, and agenda for research, practice and policy 319 5
3 Min (2010) Artificial intelligence in supply chain management: theory and applications 111 16 4 Kumar et al. (2010) Minimisation of supply chain cost with embedded risk using computational
intelligence approaches 101 1
5 Dubey et al. (2020) Big data analytics and artificial intelligence pathway to operational performance under the effects of entrepreneurial orientation and environmental dynamism: A study of manufacturing organisations
6 Calatayud et al. (2018) The self-thinking supply chain 65 6
7 Tsang et al. (2018) An Internet of Things (IoT)-based risk monitoring system for managing cold
supply chain risks 59 2
8 Toorajipour et al.
(2021) Artificial intelligence in supply chain management: A systematic literature
review 55 13
9 Mahroof (2019) A human-centric perspective exploring the readiness towards smart
warehousing: The case of a large retail distribution warehouse 46 6 10 Benzidia et al. (2021) The impact of big data analytics and artificial intelligence on green supply
chain process integration and hospital environmental performance 35 1
et al., 2011). However, it is important to note whilst it would signify the relatedness of two documents, that does not mean they are necessarily linked.
With this area of research exhibiting growth over recent years as displayed by the increase of published work as displayed in Figure 2, a co-citation analysis allows the identification of the most prevalent and relevant growing themes according to their degree of semantic similarities. As a result, this type of analysis assists in providing a retrospective view of the subject in question and identifies evolutions in this field.
For this analysis, the threshold criteria were determined as follows. The minimum number of citations of a cited reference was set at 2. Based on this, 161 cited references met the threshold out of a total of 7464. Out of the 161 cited references, the largest set of connected items consisted of 149 cited references and for this reason, 12 references were excluded from this analysis.
Figure 5 visualises the results of the co-citation analysis. In this visualisation, the size of each node represents the relative number of citations. From this analysis, influential thematic areas are identified based on the clusters exhibited and the colours of each cluster. To determine what those thematic areas are exactly; it is worthwhile to identify the top 10 publications based on TLS which are outlined in Table 5. Based on this, the prevalent themes in this subject seem to be supply chain risk management, supply chain resilience, using AI as a predictive mechanism, and AI in decision making.
Figure 5. Co-citation network analysis.
Table 5. Showing the top 10 documents from the co-citation analysis ranked by TLS.
Rank Author(s) Title Citations TLS
1 Baryannis et al. (2018)
Supply Chain Risk Management and Artificial Intelligence: State of the Art
and Future Research Directions 8 127
2 Baryannis et al. (2019) Predicting supply chain risks using machine learning: The trade-off between
performance and interpretability 5 101
3 Min (2010) Artificial intelligence in supply chain management: Theory and applications 6 100 4 Carbonneau et al.
Application of machine learning techniques for supply chain demand
forecasting 6 74
5 Ivanov & Dolgui (2020)
Viability of intertwined supply networks: extending the supply chain resilience angles towards survivability. A position paper motivated by COVID-19 outbreak
6 Ivanov et al. (2018) The impact of digital technology and Industry 4.0 on the ripple effect and
supply chain risk analytics 4 65
7 Ivanov (2020)
Predicting the impacts of epidemic outbreaks on global supply chains: A simulation-based analysis on the coronavirus outbreak (COVID-19/SARS-
CoV-2) case 3 61
8 Toorajipour et al.
Artificial intelligence in supply chain management: A systematic literature
review 4 59
9 Tushman & Nadler
(1978) Information Processing as an Integrating Concept in Organizational Design 3 57 10 Duan et al. (2019) Artificial intelligence for decision making in the era of Big Data – evolution,
challenges and research agenda 4 56
Keyword co-occurrence network analysis
A keyword co-occurrence network analysis represents a visualisation of the collective interconnectedness of keywords based on their occurrence within a specified unit of text (Segev, 2021). The purpose of this type of analysis is its usefulness in detecting research trends as well as localising information available in publications based on the presence of certain keywords.
In this research, the co-occurrence nodes in the network represent keywords that have been chosen by the author as well as index keywords. This allows for the keywords to be the best representative of the content of each author’s publication. The inclusion of index keywords in this analysis was done to make sure that synonyms, abbreviations, or alternate spellings are taken into account as a means to increase validity. Overall, what this particular analysis manages to achieve is allowing a deeper understanding of the topic being examined through the revelation of patterns and trends occurring in the research area. Additionally, the keyword co- occurrence network analysis provides support and additional insight to the CNA by including important bodies of literature that are not necessarily connected by citations.
This analysis was performed with the inclusion of both author and index keywords.
Besides that, the following specifications were set. The number of occurrences of a keyword was set to 4. This allowed 25 keywords to meet the threshold out of a potential 837. The resulting network analysis is displayed in Figure 6. Additionally, Figure 7 provides a density visualisation of the network to provide increased insight into each keyword and the frequency by which it occurred. Table 6 goes to highlight the keywords retrieved and groups them by cluster.
From this analysis, the most significant themes identified appear to be ‘artificial intelligence’, ‘supply chain management’, ‘supply chain’, and ‘decision support systems’. The identification of these keywords as being most prevalent strengthens the relevance of the identified articles in relation to providing a comprehensive framework for utilising AI tools within the scope of SCM and exporting activities.
Figure 6. Co-occurrence network visualisation of (author and index) keywords.
Figure 7. Co-occurrence density visualisation of (author and index) keywords.
Table 6. Clusters identified using the VOSviewer clustering algorithm.
Cluster 1 Cluster 2 Cluster 3 Cluster 4
- Artificial intelligence
- Supply chain
- Big data
- Big data analytics
- Logistics - Covid-19
- Supply chain management
- Decision support systems
- Decision support system
- Decision making
- Machine learning
- Bullwhip effects
- Supply chains
- Internet of things
- Sustainable development
- Technology adoption
- Sales - Export - Forecasting
- Supply chain operation
Global citation score analysis
A global citation score (GCS) analysis portrays the cumulated citations a paper has received in a given database regardless of their inclusion in a connected component of a citation network (Khitous et al., 2020). Therefore, a GCS analysis can identify seminal or recent breakthrough studies in a specified field. Generally, works that receive a high GCS are characterised by being influential in their related field. Essentially a GCS analysis is able to
identify important pieces of literature that represents the basis of a field and from which various authors have drawn from for the development of their own contributions.
To identify the seminal pieces of work, Table 7 classifies publications based on their normalised GCS which allows for the identification of the top 10 publications that have had the biggest impact. The normalised GCS manages to account for publications that might have a lower GCS, however, they have a great current appeal and significance in the scientific community. The normalised GCS is calculated by dividing the number of citations a paper has received in 2021 by the number of years since its publication. For this calculation, 2022 is taken as the base year.
The outcome of the various conducted analyses allows the attainment of an overall outlook regarding the current state of research within the field of AI in the context of relevant SCM applications. The analyses also manage to bring insight regarding the existing knowledge, influential literature, core themes, and potential trajectories of where research in this field is heading. Given these outcomes, AI has various applications in the SCM domain which can extend to help a corporation with its internationalisation efforts. Through this literature review, various AI systems have been identified and extracted from influential sources that can help cooperation in SC areas such as SCM inventory control and planning, SC transportation network design, purchasing management, demand planning and forecasting, and e- synchronising the SC.
Inventory control and planning. Inventory presents an area where AI can have tremendous potential concerning how businesses conduct their operations and presents avenues to generate value. Holding inventory has a high annual cost in relation to the overall value of a firm’s product as it can account for 20% to 30% of the total value of the inventory which varies per industry and business size (Tuovila, 2020). Due to inventory being such a substantial amount, this can be an area in which cost reductions can aid businesses in generating increased profit margins as well as allow them to stay competitive in existing or new markets. In order for a business to make cost minimisation efforts, accurate, real-time information is needed that pertains to customer demands, order cycle times, and vendor and supplier specifications amongst other variables. This would allow AI implementation in this domain to be used in decision-making in which expert systems can be especially valuable due to its ability to handle unexpected changes and proactively take inventory and control planning decisions to optimise SCM operations (Min, 2009).
Table 7. Top 10 publications in Scopus ranked by normalised GCS.
Citations received in
Rank Author(s) Title Publication
Year Journal Appears in
the CNA <2018 2018 2019 2020 2021 2022 GCS Normalised GCS
1 Wolfert et al. Big Data in Smart Farming – A review 2017 Agricultural
Systems No 13 98 196 236 280 100 923 56
2 Baryannis et
al. Supply chain risk management and artificial intelligence:
state of the art and future research directions 2019
International Journal of Production
Yes 0 0 6 48 95 28 177 32
Role of multiple stakeholders and the critical success factor theory for the sustainable supplier selection process
International Journal of Production Economics
No 2 14 33 36 41 23 149 10
4 Dweiri et al. Designing an integrated AHP based decision support
system for supplier selection in automotive industry 2016 Expert Systems
with Applications No 31 44 47 34 42 23 221 7
5 Min Artificial intelligence in supply chain management:
Theory and applications 2010
International Journal of Logistics
Research and Applications
Yes 10 5 13 23 48 12 111 4
6 Holopainen et
al. Outlook for the next generation's precision forestry in
Finland 2014 Forests No 34 13 18 22 17 8 112 2
7 Tako &
The application of discrete event simulation and system
dynamics in the logistics and supply chain context 2012 Decision Support
Systems Yes 138 32 40 24 24 6 264 2
8 Neto et al. From closed-loop to sustainable supply chains: The
WEEE case 2010
International Journal of Production
Yes 80 13 20 8 25 6 152 2
9 Kumar et al. Minimisation of supply chain cost with embedded risk
using computational intelligence approaches 2010
International Journal of Production
Yes 40 7 11 16 25 2 101 2
Additionally, expert systems manage to bring additional value due to its ability to identify and capture patterns in inventory control and planning in a SC with extreme accuracy.
More recently, expert systems have been implemented into materials requirement planning.
This allows this AI tool to capture dynamic and complex ongoing activities within the inventory database and then proceed to generate predictions such as optimal stock levels, order replenishing timeframes, and distribution strategies.
Furthermore, there is a co-evolutionary immuno-particle swarm optimisation algorithm that can be utilised within SCM to alleviate inventory replenishment problems that are exhibited by firms who have dispersed locales (Sinha et al., 2012). This algorithm can be especially important regarding exporting activities as firms are able to grasp the dynamic capabilities of various distributed plants and take actions to improve cost-effectiveness. Additionally, its deployment includes the ability to quickly react and adapt to changes and constantly try and find better SCM strategies to implement.
More recently, fuzzy linear programming methods have been utilized for solving inventory planning and control problems when it concerns SCs that have multi-echelon, multi- product, and multi-period characteristics (Peidro et al., 2010). This showcases that fuzzy set theory seems to hold an important position in the future development of inventory control as most recent breakthroughs have been focused on its potential. Additionally, machine learning has been an area of interest with Carbonneau et al. (2008) showcasing the benefits of using neural networks, and support vector machines to forecast potential distortions in demand that could be an underlying cause of a bullwhip effect. Doing so proves to be a beneficial way of introducing risk-mitigating efforts within SCM.
Supply chain transportation network design. Transportation network design problems are often complex in nature which makes it difficult to find optimal solutions. This complexity stems from the fact that various inputs need to be taken into consideration in the decision- making process. Current problems that are being exhibited within this area relate to vehicle routing and scheduling, freight consolidation, and intermodal connectivity. The incorporation of genetic algorithms can be an optimal AI tool to analyse ongoing issues and provide solutions that would optimise route details, fuel costs and other related factors to attain the best possible transportation network operations (Chambers, 1995).
Additionally, the deployment of the ant colony optimisation algorithm can also bring value when designing a SC network. Businesses have been increasingly adopting this algorithm with successful results to aid in decision-making relating to vehicle route planning or approaching the minimum spanning tree problem (Dorigo & Gambardella, 1997). This
algorithm can accommodate increased flexibility and environmental changes which allows it to account for variations that might occur which are two important factors to have when approaching combinatorial problems.
Besides, AI implementation in transportation can optimise resources and bring forth diminished environmental impacts. The main area of impact is its ability to optimise backhauls which refers to an empty truck making a return to its original starting point (Fedortchenko, 2022). Backhauls present to be an opportunity to reduce environmental impacts. AI tools can manage to optimise backhaul bookings and efficient truck loading to make sure that the truck’s journey back to its original starting point not only brings value but diminishes environmental impacts as opposed to having an empty truck. By taking such action, companies could manage to reduce empty running mileage and decrease their environmental footprint.
Purchasing management. One of the many decisions that companies need to make is the make-or-buy decision. It is seldom as simple as contemplating a few factors as it must take into account a broad overview of variables such as volume requirements, available capital, level of associated risks, and the company’s capabilities. Due to this type of decision being multifaceted, it is important for it to be approached by an AI tool that is able to handle systematic decision-making to ensure that the optimal outcome regarding purchasing and supply management is attained. Various algorithms and AI tools have been developed to aid in this area, one of which is expert systems that have been employed by companies as an assist to purchasing managers (Humphreys et al., 2002). These expert systems manage to analyse information to provide evaluations on prospective suppliers which allows managers to attain information required for decision-making easily and quickly.
Moreover, Kim et al. (2002) have developed an agent-based purchasing system that automates online material ordering processes when it concerns supply management decisions that involve a global supply base. Additionally, Cheung et al. (2004) have researched hybrid agent and knowledge-based AI systems and have seen that they have garnered great outcomes for businesses as they have a wide array of capabilities such as evaluating bids from various suppliers based on their performance and order fulfilment capabilities. This brings forth information that allows for the making of more informed and efficient decisions regarding which suppliers to source from.
Also, there have been applications going one step further than simply allowing AI to assist purchasing managers’ decision-making by allowing AI to fully take control and automate the purchasing process. This allows it to identify suitable suppliers, evaluate these suppliers, screen them, and go through all the necessary steps to complete the purchase order. This type
of AI tool deployment could allow the attainment of strategic and advantageous purchasing decisions due to its ability to handle big data.
Demand planning and forecasting. AI tool implementation within this area of SCM has been on the rise in recent years. From the various possible tools, artificial neural networks (ANNs) offer the most potential primarily due to their powerful computational capabilities which allows the handling of highly dynamic data (Kasabov, 2019). In addition, ANNs applications in SCM have grown in popularity due to their unique problem-solving capabilities and their ability to solve data-intensive problems in which rules or algorithms might not already exist or are too complex to codify (Li, 1994; Chen et al., 2008).
Various other AI tools have also been identified to forecast demand with high levels of accuracy and, as a result, improve performance and increase the effectiveness of SCM. These tools include but do not limit to fuzzy models, data mining, and support vector machines. Data mining in particular allows for big databases to contain valuable information that can be utilised in the decision-making process in which its application can be used to control and monitor warehouse activities and the effect that SC efforts and other related changes might have on demand management (Ting et al., 2014).
The increased adoption of AI tools within demand forecasting and management promises to be a great value-generating opportunity for businesses. This is because as opposed to traditional demand forecasting methods that heavily rely on historical data which can be inaccurate or invalid, these new AI tools can predict future demand based on a combination of past, present, and future customer behaviour (Jeong et al., 2002). Similar outcomes have also been achieved through an agent-based demand forecasting technique developed by Yu et al.
(2002) that utilizes data mining techniques alongside human knowledge to be able to predict aggregate demand for new products with extreme accuracy. What is tremendously advantageous about these tools is their ability to forecast consumer demand with no historical data which is especially important when considering new product or service development.
These capabilities can also be beneficial in the context of exporting activities as they would allow the tools to aid in proactively seeking exporting opportunities through the timely analysis of various markets and their potential aggerate demand levels even when no historical data is present to identify international markets with successful exporting potential that firms can enter.
E-synchronised SCM. An important aspect of SC activities is to ensure coordination and integration along both the upstream and downstream SC. To achieve that, it is imperative to ensure that information exchange is maximised regarding SC-wide activities to better coordinate SCM efforts. Information exchange can be seen as vital to achieve this in which AI
tools such as web mining and text mining can significantly improve coordination and synchronicity (Min, 2009). These tools manage to add value by extracting and identifying new patterns within data that relate to customer profiles, supplier profiles, sales and revenue trends, and changes in sourcing practices. Such would allow companies to localise these changes and take the required action to maintain efficient operations.
In the context of international business activities, employing AI tools that aim to synchronise the entire SC can be advantageous as it can account for volatile changes in the environment or unexpected events which allows the business to proactively mitigate any risks.
Additionally, such tools can garner insight into future profitable customer bases, allow them to develop appropriate marketing strategies for various independent markets, and evaluate trading partners to ensure optimal partnership potential. Through a combination of these aspects, a firm can manage to achieve increased revenue as a result of the opportunities that web and text mining can bring.
This research took the form of a systematic literature review to provide a comprehensive overview regarding AI and its applicability in SCM and exporting practices. In doing so, this literature review allows the identification of seminal works in the research domain which allowed the extraction of practical applications in which specific AI tools and methods can be utilised to generate value through their implementation into a firm’s SC.
The results of this research showcased that the field of AI is ever-increasing in popularity with the new potential SCM implications becoming apparent as this field grows and develops further. Five applications have been of focus at the end of this research in areas pertaining to inventory control and planning, SC transportation network design, purchasing and supply management, demand planning and forecasting and e-synchronised SCM. Within each area, it has been seen that various AI tools or algorithms can be implemented to optimise performance and aid the human decision-making process to achieve better outcomes. It has also been seen that AI can be proactive about seeking business opportunities and strategies which can be vital for businesses that want to pursue exporting opportunities or those with existing exporting activities that can be improved. Additionally, its ability to process volatile data and notice precise SC disruptions or changes allows it to be a useful SC risk mitigation tool.
Through the integration of AI, this can manage to have transformation potential on business strategies regarding exporting and SCM processes to provide the business with comparative and competitive advantages.
The results of this research have important practical implications as they can provide a framework that firms can utilise in their pursuit of a digital transformative business strategy.
This can allow businesses an initial step in localising which SC areas can be improved through the integration of AI tools as well as introduce the necessary tools to make such changes.
Further, the results manage to showcase that AI is a very beneficial technology when pursuing internationalisation as it has many aforementioned advantages. Overall, these findings seem to be aligned with the general literature of AI within SCM as it emphasises the existing knowledge but also puts it into the perspective of internationalisation efforts. Whilst it is important to recognise the tremendous potential of AI, some associated constraints could present themselves upon implementation that are tool- or algorithm-specific and which companies need to be taken into consideration. Such would need caution to be exercised to reduce potential risks and smoothen the transition process of AI implementation.
Whilst this research provides a new perspective on AI tool applications, it is important to recognise that it is limited. Only one database was used for the conduction of this systemic literature review which presents a limitation regarding the retrieved literature. By expanding the number of databases new publications can be retrieved to provide a more comprehensive overview. Additionally, there have been a lot of exclusion criteria applied to select the relevant literature that was analysed in this study. Whilst it does provide for more relevant analyses, it also manages to narrow the scope of applicable publications thus leaving out publications that could be relevant but lie outside of the bounds of the applied exclusion criteria. Doing so fails to account for the potential cross-field applicability of certain conducted research and perhaps through its inclusion in future research novel applications of AI tools can be discovered. This is especially important due to research conducted in fields with high potential such as AI has high variability across research areas. Lastly, the inclusion criteria of this literature review allowed the collection of generalisable insights that whilst specific to SCM could still potentially leave out niche applications of AI in SCM practices that could prove to be extremely influential thus hindering the reliability of the analysis.
Future research direction
There are different research directions that could be pursued to provide a more comprehensive understanding of this research area especially due to it being a growing research field. To start, it would be worthwhile for future research to include various other databases
such as Science Direct, JSTOR, and Web of Science. Through doing so, it is possible to retrieve various research publications and publication formats that are available on other databases that might not have been on Scopus which would expand the overall scope of research to identify even more prevalent AI tools and applications that might exist. Further, exploring the applicability and impacts that AI has regarding other types of internationalisation efforts could be fruitful in identifying new application perspectives that have currently not been explored as research into this specific area of AI applicability has been limited. Additionally, it would be critical to outline the success and impeding factors when it comes to the implementation efforts of specific AI tools/systems within SCM, this could provide firms with an increasingly useful framework of what measures to take into account when deploying AI. Another interesting avenue is research into a hybrid AI framework that combines both established successful mathematical modelling and AI tools which can showcase AI systems that have the potential to encompass proactive and predictive problem-solving characteristics as well as have active SC risk mitigation capabilities.
This research paper aimed to discover the value-generating ability of AI tools in the context of SCM and exporting activities. It was important to do so as this field of research is growing in popularity with new applications becoming ever-increasingly relevant for practical business applications. In addition, applications within the context of exporting activities have been limited in nature. This research was worthwhile in bringing forth AI tools that could be applied to generate value regarding SCM activities and that have an impact on exporting practices and general internationalisation efforts. The systematic literature review conducted allowed for the identification of seminal research papers, prevalent themes, and particularly influential publications. Various main themes have been identified through this research in which the existing literature was reviewed to extract relevant SC areas in which AI tools can be implemented to optimize and improve operations. Five main applications were discussed with different AI tools that are relevant to each one. These applications manage to present venues where businesses can transform their operations through introducing digitalisation efforts that will aid human decision-making. As a result, benefits would arise due to the increased computational power that AI tools can bring alongside their unique capabilities of handling volatile data and generating strategic insights. Such results indicate that AI and its accompanying tools and algorithms have very promising applications for SCM business processes which should be embraced to implement a digitalised internationalisation strategy. In
conclusion, whilst the transformational potential of AI has widely been established, this paper provides a framework in which such potential can be harnessed to create uncontested value generating mechanisms for businesses.
Axtell, R. L., Andrews, C. J., & Small, M. J. (2001). Agent-Based Modeling and Industrial Ecology. Journal of Industrial Ecology, 5(4), 10–13.
Baryannis, G., Validi, S., Dani, S., & Antoniou, G. (2018). Supply chain risk management and artificial intelligence: state of the art and future research directions. International Journal of Production Research, 57(7), 2179–2202.
Burnham, J. F. (2006). Scopus database: a review. Biomedical Digital Libraries, 3(1).
Carbonneau, R., Laframboise, K., & Vahidov, R. (2008). Application of machine learning techniques for supply chain demand forecasting. European Journal of Operational Research, 184(3), 1140–1154. https://doi.org/10.1016/j.ejor.2006.12.004
Chambers, L. (1995). Practical Handbook of Genetic Algorithms Applications Volume I (1st ed.). CRC-Press.
Chen, S. H., Jakeman, A. J., & Norton, J. P. (2008). Artificial Intelligence techniques: An introduction to their use for modelling environmental systems. Mathematics and Computers in Simulation, 78(2–3), 379–400.
Cheung, C. F., Wang, W. M., Lo, V., & Lee, W. B. (2004). An agent-oriented and
knowledge-based system for strategic e-procurement. Expert Systems, 21(1), 11–21.
Delurgio, S. A. (1997). Forecasting, Principles, and Application. McGraw-Hill College.
Denyer, D. (2009). Producing a systematic review. In D. Tranfield (Ed.), The SAGE Handbook of Organizational Research Methods (1st ed., pp. 671–689). SAGE Publications Ltd.
Disney, S., Naim, M., & Towill, D. (2000). Genetic algorithm optimisation of a class of inventory control systems. International Journal of Production Economics, 68(3), 259–278. https://doi.org/10.1016/s0925-5273(99)00101-2
Dogru, A. K., & Keskin, B. B. (2020). AI in operations management: applications, challenges and opportunities. Journal of Data, Information and Management, 2(2), 67–74.