ChatGPT and generative artificial intelligence: an exploratory study of key benefits and challenges in operations and supply chain management

Abstract

ChatGPT and generative artificial intelligence (Gen-AI) are transforming firms and supply chains. However, the empirical literature reporting the benefits, challenges, and outlook of these nascent technologies in operations and supply chain management (OSCM) is limited. This study surveys current projects and perceptions of these technologies in US (n = 119) and UK (n = 181) supply chains. We found that projects range from proof-of-concept to full implementation, with a main focus on operational gains, such as improved customer satisfaction, cost minimisation, and process efficiencies. The main challenges concern data, technological and organisational issues. Expected benefits are dominated by cost savings and enhanced customer experience, but also include increased automation and sustainability. Industries were found to cluster around six groups according to perceived benefits and implementation challenges. Our findings contribute to the emerging literature on Gen-AI use in OSCM, and to management practice by mapping the benefits, challenges, outlook, and maturity level of Gen-AI projects in supply chains.

KEYWORDS:

• ChatGPT
• generative AI
• operations
• supply chain
• benefits
• challenges

1. Introduction

Artificial intelligence (AI) has a long tradition in production, operations and supply chain management (OSCM) research (Elliott and Griffiths 1990; Vujosevic 1994) and has recently become a key topic following significant advances in computational power (Dey et al. 2023; Jauhar et al. 2023; Kusiak 2020; Merhi and Harfouche 2023; Wong et al. 2022). AI tools are already used in various OSCM activities, including demand forecasting, inventory management, risk management (Pournader et al. 2021), predictive maintenance, quality control, process optimisation, task scheduling (Rai et al. 2021), sales, maintenance, and operations (Helo and Hao 2022). Indeed, AI has been shown to create competitive advantages (Helo and Hao 2022; Pournader et al. 2021; Sharma et al. 2022), and improve decision-making (Pournader et al. 2021). In the future, AI integration could enable ‘autonomous’ supply chains, with the ability to be self-aware, self-governing, self-determining, and self-optimising (Helo and Hao 2022).

The recent exploits of ChatGPT have brought generative artificial intelligence (Gen-AI) to the fore (Peres et al. 2023). The media have marvelled at its capacity to replicate and surpass human intelligence, notably demonstrated by its ability to pass graduate-level exams¹, such as the United States Medical Licensing Exam (Tiffany et al. 2022), the Uniform Bar Exam (Katz et al. 2023), and an MBA level Operations Management exam (Terwiesch 2023). Scholars have greeted its abilities with both excitement and apprehension (Jo 2023). The emergence of ‘conversational’ generative AI technologies such as ChatGPT allows for easier and more complex interaction with users (Rožanec et al. 2023) and promises faster and more widespread adoption compared to traditional, programming based techniques. Today, Gen-AI is predicted to add trillions of dollars in value annually to the global economy (Chui et al. 2023), and the global market size for Gen-AI technologies and services is projected to reach $126.5 billion by 2031 (Allied Market Research 2023).

While much has been written about the promises of Gen-AI and its ability to mimic human decision-making, the intentions, obstacles, and challenges of its adoption and use in OSCM have received little scholarly attention (e.g. Badini et al. 2023; Fosso Wamba et al. 2023; Hendriksen 2023; Peres et al. 2023; Wang, Liu, and Shen 2023). And yet, the practitioner literature has continually highlighted the transformative potential of Gen-AI (Deighton 2023), with one promising area being operations and supply chain management (Dolgui and Ivanov 2023; Telling 2023; Triantafyllidis and Kuepper 2023; Zagorin 2023). Industry reports have described the business value of these technologies, as well as the required capabilities, associated risks (Deloitte 2023; EY 2023), scalability challenges (PWC 2023), ethical concerns, technology governance and sustainability issues (EY 2023). Chui et al. (2023) even heralded 2023 as the ‘breakout year’ for Gen-AI.

Further scholarly research is urgently required to better understand the adoption and diffusion of Gen-AI technologies across industries in OSCM. This study is an initial effort in this direction by exploring how companies are currently using Gen-AI, the benefits and challenges it presents, and the future envisioned for these technologies. More explicitly, this study intends to answer the following research questions:

How are companies using Gen-AI in OSCM?

What challenges do companies face implementing Gen-AI in OSCM?

What are the future expected benefits of using Gen-AI in OSCM?

How does adoption of Gen-AI in OSCM differ between industries?

To address our research questions, we collected data from middle and upper-level managers across a range of industries in the USA and the UK. Our research is exploratory and descriptive in nature, seeking to capture the emergence of this relatively unexplored area (Punch 2013). Following an exploratory descriptive research design, we use the extant diffusion of innovation (Rogers 1983) and OSCM literatures to build research propositions to structure our work.

The paper is organised as follows. The following section provides a literature review of Gen-AI and develops several propositions to guide our research. The research methods used in data collection and analysis are then discussed in section three. The results are presented in section four, followed by a discussion in section five. Section six concludes the paper.

2. Literature review

2.1. Gen-AI and OSCM

Gen-AI refers to algorithms that can generate new content such as images, videos, documents, and audio based on the data they were trained on (Dwivedi et al. 2023; Peres et al. 2023). The main difference between Gen-AI and traditional AI is that the former can ‘generate novel content, rather than simply analysing and acting on existing data’ (Gozalo-Brizuela and Garrido-Merchan 2023). Also, traditional AI requires explicit programming to execute tasks, whereas Gen-AI may generate content based on human-machine ‘conversations’. Nonetheless, Gen-AI is still built on AI technologies such as neural network and deep learning algorithms. For example, ChatGPT uses large language models (LLM) based on a large number of parameters to train and learn about relations between data and make predictions about what content should be generated. Consequently, Gen-AI requires access to vast collections of data that are mainly sourced from the internet (Huang, Wang, and Yang 2023; Webb, Holyoak, and Lu 2023).

Gen-AI enabled OSCM could lead to a paradigm shift in the field (Jackson, Saenz, and Ivanov 2023) and provide several distinct advantages such as improved sourcing assessment, supply chain risk mitigation, procurement and inventory accuracy, logistics coordination, and customer relations, and more sustainable and ethically conscious OSCM practices (Richey et al. 2023). These technologies also hold promise for enhanced decision-making processes, faster real-time responses to customer queries, and improved customer behaviour forecasting (Precedence Research 2023).

2.2. Adoption and use of AI and Gen-AI

The high operational and strategic potential of Gen-AI has led some practitioners and scholars to go so far as to suggest that Gen-AI is the ‘next big thing for business transformation’ (Gascoigne 2023). However, past IT innovation studies tell us that the acceptance of any technological innovation is mainly driven by organisational perceptions of business value based on an assessment of the benefits and challenges of adoption (e.g. Keating et al. 2010).

A large number of studies have examined the different uses of AI technologies (Wamba et al. 2021), the benefits that accrue from them (Merhi and Harfouche 2023), and the challenges encountered during adoption (Kamoonpuri and Sengar 2023). AI is also seen as an integral part of Industry 4.0 and beyond with manifold opportunities within digital transformations (Ivanov et al. 2021). However, empirical studies on the uses, benefits and challenges of Gen-AI in OSCM are lacking (Fosso Wamba et al. 2023). Table 1 illustrates some of the possible benefits and challenges of Gen-AI based on publicly available business use cases.

Table 1. Illustrations of Gen-AI projects and outcomes.

Use case	Context	Description	Realised and expected benefits
Maersk^a	Customer service	Maersk is using ChatGPT on their website (Maersk.com) to automate responses to frequently asked questions. It enhances the accuracy of queries and contributes to growth in website traffic.	The company expects to offer more Gen-AI based services that enrich the customer experience, such as prompt and accurate responses to customer requests.
Walmart^c	Procurement negotiations	The retailer Walmart uses an AI chatbot that is specialised in autonomous negotiations. The tool considers many variables, such as payment terms, discounts, prices, historical trends, and informs buyers of the best price the supplier is most likely to accept.	The rate of successful deals negotiated by the chatbot is 68% with savings on contracts of 3%.
Unilever^b	Consumer engagement	Unilever is using a GPT API to enhance consumer engagement and relationships by filtering emails, managing spam sorting, and recommending responses to human teams.	The tool has reduced staff time spent preparing replies to queries by 90%. It can analyse consumer questions and store sentiment data for later use.
Mercedes-Benz^d	Production processes	Mercedes-Benz is implementing a voice based ChatGPT in their digital production ecosystem. The AI tool will support key OSCM processes such as production optimisation and quality management.	The company expects ChatGPT to optimise the analysis of production data, and improve error identification, quality management, and process optimisation.
Tel Aviv Sourasky Medical Center^e	Triage processes	The Tel Aviv Sourasky Medical Center became the first hospital to integrate an AI chatbot to support its clinical processes. The tool can assist physicians and nurses with a robust summary of the patient’s conditions and give valuable insights about diagnostics and the next steps in the treatment.	The organisation expects to improve the efficiency and accuracy of its clinical processes. It is expected to significantly reduce in staff burnout by allowing the medical and nursing staff to focus on treatment plans.

^a ‘The AI chatbot ChatGPT is a game-changer!’, 14 March 2023 https://www.maersk.com/news/articles/2023/03/14/the-ai-chatbot-chatgpt-is-a-game-changer.

^b ‘Unilever leverages GPT API to deliver business value’, 10 March 2023, https://www.cio.com/article/464190/unilever-leverages-chatgpt-to-deliver-business-value.html

^c ‘Walmart Is Using AI to Negotiate the Best Price With Some Vendors’, 26 April 2023, https://www.bloomberg.com/news/articles/2023-04-26/walmart-uses-pactum-ai-tools-to-handle-vendor-negotiations

^d ‘Mercedes adopts ChatGPT’, July 7 2023, https://www.just-auto.com/news/mercedes-adopts-chatgpt/

^e ‘Tel Aviv Hospital First Ever To Use AI Intake Including ChatGPT’, August 09 2023, https://nocamels.com/2023/08/tel-aviv-hospital-first-ever-to-use-ai-including-chatgpt-triage/

Following an exploratory descriptive research design (Bickman, Rog, and Hedrick 2009; Punch 2013), we draw on diffusion of innovation theory (Rogers 1983) inter alia, to formulate a set of research propositions to guide our research.

2.2.1. Diffusion of innovation theory: a theoretical prism

Diffusion of innovation theory provides a theoretical prism to explore and describe the nascent use of Gen-AI in organisations. Past IT innovation studies tell us that the acceptance of any technological innovation is mainly driven by organisational perceptions of business value based on an assessment of the specific benefits and challenges related to its adoption (Keating et al. 2010; Mintzberg, Raisinghani, and Théorêt 1976).

The IT innovation literature has notably examined the reasons behind organisational innovation, its facilitators and inhibitors, and the methods used to foster it (Fichman 2000; Fosso Wamba and Chatfield 2009; Keating et al. 2010). This literature can be classified into two main streams: (a) adoption studies that focus on differences in organisational innovativeness by examining, for example, the characteristics of an organisation as well as the timing and the extent of innovation adoption; (b) diffusion studies that examine the ‘rate, pattern, and extent of technology diffusion’ (Fichman 2000, 110) across a given population.

Multiple factors have been found to influence the diffusion of any given innovation, including technological characteristics (e.g. compatibility, complexity, relative advantage, trialability, observability (Rogers 1983)), organisational characteristics (e.g. organisational readiness (Lee and Shim 2007), organisational culture (Fichman 2000), and top management support (Lu et al. 2021)), and environmental characteristics (e.g, competitive pressure (El-Haddadeh et al. 2021), standards (Fichman 2000), government support (Lu et al. 2021), and regulations (El-Haddadeh et al. 2021)).

Rogers (1983) identified five adopter categories. Innovators (2.5% of adopters) are always interested in new ideas and the first to adopt an innovation. They play ‘an important role in the diffusion process: that of launching the new idea in the system by importing the innovation from outside of the system’s boundaries’ (p.248). Early adopters (13.5%) represent the second set of adopters after the innovators and are quick to adopt an innovation. They are usually the ones that help trigger the critical mass when they adopt an innovation. The early majority (34%) adopt just before an average member of the population, but they typically take a while to fully embrace the innovation. The late majority (34%) consistently adopt innovations with caution and skepticism, often only after the average member of a system has done so. Laggards (16%) are usually the last in a population to adopt an innovation as they tend to be very suspicious of innovations.

2.2.2. Research propositions

We formulate four research propositions based on the extant scholarly literature, practitioner and industry publications, and the theoretical prism of diffusion of innovation theory.

The first proposition concerns initial projects and benefits. Compared to prior IT-enabled OSCM, Gen-AI is expected to improve productivity and efficiency (Fosso Wamba et al. 2023). The technology notably has the potential to solve the productivity paradox as discussed for example in Brynjolfsson (1993) and Brynjolfsson, Rock, and Syverson (2018). For example, it could allow for a more reliable and efficient interface between suppliers, shippers, truckers, and warehouses (Mizell-Pleasant 2023). Brynjolfsson, Li, and Raymond (2023) also report that Gen-AI increases operational productivity and positive customer sentiment by transforming the way novice and low-skilled workers learn.

The first wave of Gen-AI implementation in the supply chain is expected to focus on task automation to more efficiently execute repetitive tasks at operational and managerial levels (Gravier 2023; Wiles 2023). These benefits have been largely described in the practioner literature. For example, Young (2023) reported an increase in Gen-AI solutions by logistics companies, essentially in customer support processes. Likewise, Zagorin (2023) highlighted the potential process efficiencies and cost savings from the adoption of Gen-AI in procurement activities. Lamba (2023) and Telling (2023) reported the use of Gen-AI to more efficiently manage supply chains and minimise disruption risks. We, therefore, propose that:

P1. Initial projects and benefits in OSCM will focus on task automation, efficiencies, and improved customer service.

The second proposition concerns the challenges companies face adopting Gen-AI. Several possible challenges have been identified in the literature including privacy issues, data access and a lack of standardisation and governance (Dwivedi et al. 2023). These issues have also been illustrated in the practitioner literature. For example, Young (2023) and Zagorin (2023) underscore the risks inherent in training models with proprietary data. We, therefore, propose that:

P2. Current challenges to Gen-AI implementation in OSCM will concern AI models, technologies, and data access.

A third proposition concerns the expected benefits from Gen-AI adoption. As companies and supply chains are in the early stages of adoption, few scholars have focused on the medium and long-term impacts of these technologies. Nonetheless, the disruptive and transformational potential of Gen-AI is acknowledged (Larsen and Narayan 2023), with some industries expected to reimagine their business models (Agrawal, Gans, and Goldfarb 2022). We, therefore, propose that:

P3. Expected benefits from future Gen-AI projects in OSCM will include gains accruing from business model transformations.

A final proposition concerns the rate of adoption of Gen-AI. Scholars within the two streams of innovation diffusion research have sought to identify the exact timing of the adoption event (Fichman 2000). For Fichman (2000), ‘organisations that adopt relatively early are more innovative than those that adopt later or not at all’ (111), while Rogers (1983) argues that the adoption of any given technological innovation goes through five stages (i.e. knowledge, persuasion, decision to reject or to accept, implementation, and confirmation). As a number of environmental and industry specific characteristics have been shown to influence the rate of innovation adoption and diffusion by organisations, we propose that:

P4. The rate of adoption of Gen-AI in OSCM will vary by industry.

The following section presents the exploratory descriptive data collection and analysis process we followed to verify our propositions.

3. Research method

To examine our research propositions we used an exploratory descriptive approach which focuses on revealing patterns in data rather than on testing theory based hypotheses (Helfat 2007; Oxley, Rivkin, and Ryall 2010). This approach is appropriate when an area is relatively new or unexplored (Punch 2013) and when little or no theory is available to fully explain the phenomenon (Hambrick 2007; Ven 2016; Ven et al. 2015). Exploratory descriptive research seeks to make causal propositions by exploring and describing links between variables that contribute to building a foundation for understanding relationships before formally testing them through explanatory research. Exploratory descriptive research designs have been successfully employed in OSCM research to study issues such as the use of visual management tools (Jaca et al. 2014), manufacturing strategy (Dangayach and Deshmukh 2001), lean production (Jasti and Kodali 2015) and the adoption of lean six sigma (Samarrokhi, Jenab, and Weinsier 2015).

An exploratory design is ‘indispensable in the early stages of studying a phenomenon’ (Malhotra and Grover 1998) and this is notably the case for new and highly mediatised technologies. The limited practical and theoretical understanding of Gen-AI (Susarla et al. 2023) motivated our methodological approach. Furthermore, several calls have been made in the literature for more empirical observations and descriptions of AI adoption and use (Dwivedi et al. 2023; Richey et al. 2023).

The exploratory descriptive research process used in the study is presented in Figure 1.

Figure 1. Descriptive research workflow.

Schéma illustrating the four stages of the research process including the literature review, data collection, descriptive data analysis and conclusions.The research tasks at each stage are listed.

3.1. Data collection and analysis

A questionnaire survey was used to study the benefits and challenges of Gen-AI for OSCM. The instrument consisted of twelve questions and is reproduced in Appendix 1. A market research service provider managed data collection.² Of the 560 individuals that agreed to participate in the study, 300 returned valid questionnaires (response rate of 54%). Table 2 presents the characteristics of the sample.

Table 2. Demographic and professional profile of the sample.

	USA		UK
	n	%	n	%
Gender
Male	86	72%	133	73%
Female	31	26%	48	27%
Other	2	2%	0	0%
Age
18–25	5	4%	10	5%
26–33	31	26%	48	27%
34–41	27	23%	57	31%
42–49	30	25%	36	20%
50+	26	22%	30	17%
Highest educational level
No formal education	1	1%	0	0%
Primary	0	0%	0	0%
Secondary	8	7%	16	9%
Diploma/polytechnic	17	14%	23	13%
Bachelor degree	60	50%	91	50%
Postgraduate degree (Master/Ph.D.)	33	28%	51	28%
Area of decision-making responsibilities
Accounts/finance	39	7%	62	8%
Business strategy	55	10%	87	11%
Customer/client	60	11%	84	11%
Hiring	71	14%	102	13%
Marketing/sales/advertising	49	9%	65	8%
Operations/production	74	14%	114	15%
Human resources	83	16%	131	17%
Research/development	56	11%	64	8%
Supply chain/logistics	32	6%	54	7%
Other	9	2%	12	2%
Number of years working in the organisation
Less than one year	0	0%	9	5%
2–5 years	46	39%	63	35%
6–10 years	39	33%	58	32%
11–15 years	19	16%	26	14%
16–20 years	4	3%	10	6%
Over 20 years	11	9%	15	8%
Management level
Middle management	80	67%	129	71%
Senior management	37	31%	50	28%
Other	2	2%	2	1%
Industry
Accommodation and food service activities	6	5%	8	4%
Administrative and support service activities	1	1%	3	2%
Agriculture, forestry and fishing	2	2%	2	1%
Arts, entertainment and recreation	3	3%	3	2%
Construction	8	6%	12	7%
Education	0	0%	13	7%
Electricity, gas, steam and air conditioning supply	1	1%	1	1%
Financial and insurance activities	7	6%	23	12%
Human health and social work activities	18	15%	12	7%
Information and communication	16	13%	19	10%
Manufacturing	9	8%	13	7%
Mining and quarrying	0	0%	1	1%
Professional, scientific and technical activities	24	20%	17	9%
Public administration and defence; social security	0	0%	12	7%
Real estate activities	5	4%	2	1%
Transportation and storage	3	3%	9	5%
Water supply; sewerage, waste management	0	0%	2	1%
Wholesale and retail trade; vehicle repair	10	8%	17	9%
Other service activities	6	5%	12	7%

The typical respondent was male, between 26 and 49 years old, and held a university-level qualification. On average, they had worked between two to ten years at their company, occupied a middle management position and worked in the area of strategy, operations, human resources, or marketing. The most observed industries were information and communication, professional, scientific, and technical activities, financial and insurance activities, and human health and social work activities.

4. Results

Data analysis involved studying the responses to open questions concerning current Gen-AI projects, and perceived benefits and challenges. We used bi-gram and co-word network analysis of keywords, and hierarchical cluster analysis of industry groups.

4.1. Current Gen-AI projects in OSCM

Respondents were asked to describe their current use of Gen-AI for OSCM. The responses were cleaned and analysed using the R programming language, particularly the packages dplyr, tidyr, and tidytext. Bi-grams were then computed to identify the most frequently used combinations of words. A bi-gram is a string of two adjacent words used in the same sentence by a respondent. Two word strings were considered appropriate to capture compound nouns (e.g. demand forecasting) and verb-noun collections (e.g. reduce costs) that respondents may use to describe their current projects. The 50 most frequent bi-grams were then manually inspected and cleaned, and similar bi-grams were grouped together. The ten most frequently observed applications of Gen-AI are reported in Table 3.

Table 3. Current project description bi-grams.

Rank	Current Gen-AI projects
1	inventory levels
2	demand forecasting
3	informed decisions
4	customer service
5	identify patterns
6	reduce costs
7	improve efficiency
8	delivery times
9	predictive maintenance
10	lead times

Current Gen-AI projects mainly concerned supply chain and operational issues of inventory management, demand forecasting, resource planning, process efficiencies, and time and cost savings. For example, a manager from the UK construction industry described the following project in inventory optimisation:

ChatGPT can be trained on historical inventory data to generate recommendations for optimising inventory levels. By analyzing past sales data and predicting future demand, the model can suggest optimal stocking levels for different products at different times.

Projects were also customer-oriented, and managers reported deploying Gen-AI to improve customer service, satisfaction, and experience. An IT manager from the information and communication industry described a Gen-AI project in customer service:

We will be using ChatGPT to interact with our customers. If our customers have an issue or concern, they can interact with ChatGPT to find a solution.

Current projects were then studied by industry. The results are presented in Table 4.

First ranked projects varied by industry, with some privileging customer relations (e.g. accommodation and food service, financial and insurance, and public administration and defence), operational efficiencies and savings (e.g. agriculture, forestry and fishing, human health and social work, and real estate), or business intelligence and forecasting (e.g. arts, entertainment and recreation, information and communication, and wholesale and retail trade).

Table 4. Top three current projects by industry.

Industry	n	Project 1	Project 2	Project 3
Accommodation and food service activities	14	client relations	capacity planning	cutting costs
Administrative and support service activities	4	marketing practices	optimise operations	customer services
Agriculture, forestry, and fishing	4	operational efficiency	optimise transportation	demand forecasting
Arts, entertainment, and recreation	6	business intelligence	customer queries	generate content
Construction	20	reduce costs	improve efficiency	resource planning
Education	13	personalised learning	customer engagement	educational resources
Electricity, gas, steam, and air conditioning supply	2	drafting ideas	customer relationships
Financial and insurance activities	30	customer service	automated tools	optimise inventory
Human health and social work activities	30	optimise inventory	consumer enquiries	forecast demand
Information and communication	35	demand forecasting	improve operations	customer service
Manufacturing	22	chemical structures	cutting costs	capacity planning demand forecasting
Mining and quarrying	1	customer service	24 h job updates
Professional, scientific, and technical activities	41	predictive maintenance	quality control	customer service
Public administration and defence; compulsory social security	12	customer service	creating insights	streamline onboarding
Real estate activities	7	generate content	improve operations	compliance activities
Transportation and storage	12	customer communications	inventory levels	demand levels arrange delivery
Water supply; sewerage, waste management	2	tendering process	automated workflow	improve planning
Wholesale and retail trade; repair of motor vehicles and motorcycles	27	accurate predictions	optimise inventory	customer trends delivery services
Other service activities	18	customer inquiries	generate content	save time

4.1.1. The current state of Gen-AI adoption

Respondents were then asked to identify the stage of development of their Gen-AI projects. The results are reported in Figure 2.

Figure 2. Current projects by development stage.

Vertical grouped bar chart showing the percentage of projects for the USA, UK and total respondents at the proof-of-concept, prototype, pilot, trial, partial and full implementation phases.

One-third of projects were in an experimental proof-of-concept or prototype phase, one-third were being trialled or tested in a pilot phase, and one-third of projects had already been partially or fully implemented. Table 5 presents the most frequently reported projects and the main industries by project stage.

Table 5. Current project by stage and industry.

Project stage	Top projects	Main industries
Experimental phase (proof-of-concept or prototype)	demand forecasting inventory management automate processes	Accommodation and food service activities Financial and insurance activities Information and communication Transportation and storage
Test phase (pilot or trial)	quality control inventory management customer service	Administrative and support service activities Education Wholesale and retail trade, and vehicle repair Other service activities
Implementation phase (partial/phased or full)	safety compliance inventory management future demand	Agriculture, forestry and fishing Construction Human health and social work activities Mining and quarrying Professional, scientific and technical activities Water supply; sewerage, waste management

The main projects in the experimental phase (i.e. proof-of-concept, prototype) were demand forecasting, inventory management, and process automation. The accommodation and food service activities industry reported the highest proportion of projects in this early phase of Gen-AI adoption.

The most frequently reported projects in the test phase (i.e. pilot or trial) concerned quality control, inventory management, and customer service. Companies in the administrative and support service activities industry reported most projects in this phase.

An IT director at a UK educational institution described the trialled use of ChatGPT for customer service:

We are currently in trials with a government backed project to test and to see how effective the systems are in place. We are constantly monitoring these systems to evaluate how effective they are both for us and our customers. We are doing this to move forward into a new and exciting future within our industry.

Projects that were fully or partially implemented mainly concerned safety compliance, inventory management, and future demand forecasting.

A purchasing manager at a water supply, sewerage and waste management concern reported using ChatGPT in the tendering process:

AI being used mainly in inventory management to improve planning and scheduling. ChatGPT is used operationally to support the tendering process, generating tender detail, contracts, etc., using historical data.

The stage of development varied by area of business activity, with four industries reporting that most projects were still in the proof-of-concept or prototype phase, four industries were mainly in the test phase, and six industries reported that most projects were either fully or partially implemented. Projects were spread across several phases in several industries including arts, entertainment and recreation, electricity, gas, steam and air conditioning supply, manufacturing, public administration, defence, and social security, and real estate activities.

We can compare these rates of Gen-AI deployment to adopter categories from diffusion of innovation theory. Companies that have fully implemented Gen-AI correspond to Rogers (1983) innovators category (10% of companies). The early adopters include those companies that have partially implemented Gen-AI projects (22%), while the other 68% of companies make up the early majority.

4.2. The benefits and obstacles of Gen-AI in OSCM

Respondents were asked to identify five current benefits, challenges, and expected future benefits of Gen-AI technologies. Once again, bi-grams were generated and analysed, and are reported in the following subsections.

4.2.1. The current benefits of Gen-AI

The ten most cited benefits of Gen-AI in OSCM are presented in Table 6.

Table 6. Top ten current benefits.

Rank	Current benefits
1	customer satisfaction
2	cost saving
3	improved efficiency
4	human error
5	inventory management
6	time saving
7	risk management
8	real-time
9	process management
10	enhanced decision

The first ranked benefit from Gen-AI adoption was improved customer satisfaction. For example, an IT specialist from the US described how Gen-AI can lead to an enhanced customer experience:

ChatGPT can provide customers with quick and accurate responses to their queries, leading to improved customer satisfaction and loyalty.

The three following benefits pertained to process efficiencies and reduced errors. A supervisor from the US financial and insurance industry who is currently in the proof-of-concept phase in AI deployment reported the cost-saving potential of Gen-AI:

ChatGPT can identify cost-saving opportunities by analyzing supplier performance, identifying areas of inefficiency in the supply chain, and automating procurement processes. These cost savings can translate into higher profitability for the business.

Managers also saw benefits in improved risk management. For example, a vice president of IT in the financial and insurance industry explained that Gen-AI can ‘improve risk management by generating predictive models that can help identify potential risks and vulnerabilities.’

Gen-AI was also reported to enhance decision-making and process management through a more accurate understanding of patterns, trends, and real-time insights. For example, a manager from a US marketing services company described the decision-making benefits from Gen-AI:

Increased market agility and responsiveness, allowing marketing companies to adjust to new trends and customer preferences quickly.

Keyword analysis was then performed using VOSViewer (Van Eck and Waltman 2010) to examine the verbatim responses of managers concerning all reported benefits from Gen-AI. While bi-gram frequencies allowed us to rank individual benefits, the co-word analysis of keywords revealed connections between the multiple benefits cited by managers.

Co-word analysis involves identifying the co-occurrences of keywords within a corpus and representing them in the form of a network. Each node is a keyword, and a node's size reflects its frequency within the corpus. Links indicate co-occurrences of keywords and the thickness of a link reflects the number of co-occurrences within the corpus. The distance between two keywords in the network indicates the relatedness regarding co-occurrence links. In general, the closer two nodes are to each other, the stronger their relatedness. Related nodes are grouped into coloured clusters.

The results are presented in Figure 3. Four main themes emerged: productivity, speed and cost savings (yellow, orange, and green clusters); accuracy³ and automation (blue clusters); demand management and customer service (red and brown clusters); and communication (purple cluster).

Figure 3. Co-occurrence map of current benefits.

This network representation shows that productivity, accuracy, cost, and speed benefits are often related, as are supply chain operations, optimisation, inventory management, demand forecasting, and customer service. Such relationships were often illustrated in project descriptions. For example, a senior vice president from the US wholesale, retail and repair industry reported that:

Gen-AI will be trained on historical sales data to forecast demand for different product lines and sizes accurately. This can help optimise inventory levels, minimise stockouts, and reduce overstocking.

4.2.2. The expected benefits of Gen-AI in OSCM

The ten most cited expected benefits from the use of Gen-AI in OSCM are presented in Table 7.

Table 7. Top ten future benefits.

Rank	Future benefits
1	cost savings
2	customer experience
3	real-time
4	increased automation
5	improved sustainability
6	improved decision-making
7	increased efficiency
8	increased productivity
9	time-saving
10	increased agility

Several expected benefits were also reported as current benefits by managers, including customer satisfaction, cost savings, improved efficiency, inventory management, and decision-making. However, two new benefits were expected from the use of Gen-AI in the future, notably increased automation, and improved sustainability. These two benefits were both highly cited and ranked in 4th and 5th position respectively. Also, while access to real-time information was cited as a current benefit (8th rank), it was expected to be a future benefit by an even greater number of managers (3rd rank).

Co-word analysis was then performed using VOSViewer to map connections between all expected benefits. The results are presented in Figure 4.

Figure 4. Co-occurrence map of expected future benefits.

Two major and opposing themes can be observed from the co-occurrence map of expected future benefits: productivity improvements (including time, speed, staff, and reduction), and sustainability (including agility, collaboration, safety, and risk). Three smaller themes in the network are cost savings, innovation, and supply chain operations (including real-time insights, prediction models, and scenarios).

Not surprisingly, the theme of sustainability is transversal, and is related to keywords of operations, reductions, abilities, and communication. The following pilot project described by a senior developer from the professional, scientific, and technical activities industry illustrates the sustainability benefits of Gen-AI:

For us ChatGPT and Gen-AI review material and product records to provide a summary used for logistics in the warehouse to avoid bottlenecks. We recently implemented this, and it allows us to analyze the data from various sources along the supply chain to minimise downtime and identify patterns for areas of improvement. It also allows us to track our sustainability in a way that we had not been able to before we implemented this. Another way that we use it in our supply chain management is to reduce our waste.

4.2.3. The challenges of Gen-AI in OSCM

Managers were asked to describe five current challenges related to the adoption and use of Gen-AI in OSCM. The results are reported in Table 8.

Table 8. Top ten challenges.

Rank	Challenges
1	data quality
2	model complexity
3	legacy systems
4	data privacy
5	staff resistance and training
6	data security
7	learning curve
8	3rd party platforms
9	initial time and cost
10	legal implications

The main challenges concerned issues related to data (i.e. data quality, data privacy, data security) and the Gen-AI technologies themselves (i.e. model complexity, choice and use of 3rd party platforms). For example, a manager from the agriculture, forestry and fishing industry lamented the difficulty of obtaining reliable volumes of supply chain data:

Gen-AI models require large amounts of high-quality data to learn from. However, supply chain data can be fragmented, inconsistent, and difficult to obtain, which can limit the accuracy and effectiveness of Gen-AI models.

Existing IT systems and the availability of qualified staff were also reported as major challenges to using Gen-AI in OSCM. For example, a chief technology officer from a US professional, scientific, and technical activities concern that had implemented ChatGPT to optimise fulfilment processes, described the challenges of retraining IT support and operational staff:

Retraining IT to support something that is not fully understood. Retraining staff to use the ChatGPT solution.

Interestingly, development time and cost were seen as a lesser challenge (9^th rank) than issues associated with data quality and model accuracy.

The co-occurrence of keywords was then studied using VOSViewer to detect relationships between challenges. The results are presented in Figure 5.

Figure 5. Co-occurrence map of current challenges.

Two main and opposing themes emerged from the keyword network: employee adoption and use, data quality, biases and ethical considerations. These are consistent with the challenges previously identified in Table 8.

4.3. Industry clusters by Gen-AI benefits and challenges

Industries were clustered into similar groups according to reported current and future benefits and implementation challenges. The thirty most frequent bi-grams were first extracted and ranked for all nineteen industries. Each bi-gram was then coded using an open coding procedure, resulting in a topology of nine current benefits, eight challenges and eleven future benefits. Industries were then clustered using an agglomerative hierarchical clustering algorithm for categorical variables.⁴ Six clusters were identified based on the resulting dendrogram (reproduced in Appendix 2).

The industries by cluster are presented in Table 9. The number of respondents per cluster in the original dataset is shown in brackets. Each cluster has been named to facilitate analysis.

Table 9. Industries by cluster.

Cluster	Industry	Name
1	Accommodation and food service activities (14) Education (13) Financial and insurance activities (30) Human health and social work activities (30) Information and communication (35) Professional, scientific and technical activities (41) Public administration and defence; social security (12) Transportation and storage (12) Wholesale and retail trade;vehicle repair (27)	Services
2	Administrative and support service activities (4) Construction (20) Manufacturing (22)	Building & support
3	Agriculture, forestry and fishing (4) Water supply; sewerage, waste management (2) Other service activities (18)	Land use & other
4	Arts, entertainment and recreation (6)	Arts
5	Electricity, gas, steam and air conditioning supply (2) Mining and quarrying (1)	Extraction
6	Real estate activities (7)	Real estate

The first cluster (‘services’) was the largest and included nine service and retail industries. The second cluster (‘building & support’) comprised three secondary sector industries. The administrative and support services industry also included companies that provide services to the construction and manufacturing industries (e.g. rental and leasing, security, logistics). The third cluster (‘land use & other’) was also comprised of three industries, two of which specifically concern land-based resource use. One industry was in each of the fourth (‘arts’) and sixth (‘real estate’) clusters. The fifth cluster (‘extraction’) included industries that extract, process and sell fossil fuels and minerals.

Industry clusters differed according to the importance of four current benefits (i.e. automation, cost savings, customer satisfaction, and decision support), two challenges (i.e. data quality and avalilability, legacy systems)⁵, and two future benefits (i.e. business development, and efficiency, optimisation and productivity). The key benefits and challenges by cluster are presented in Table 10.

Table 10. Benefits and challenges by cluster.

There were no significant differences between clusters in demographic (i.e. gender, age, educational level) and professional variables (i.e. area of decision-making responsibility, number of years working in the organisation, management level).⁶

While benefits and challenges varied between clusters, there were similarities: most clusters identified efficiency, optimisation, productivity and cost savings as current benefits from using Gen-AI. Industries in clusters one and six also frequently cited improved customer satisfaction, industries in clusters three and four cited automation benefits, and industries in clusters one and two reported enhanced decision support.

Challenges relating to staff resistance, accuracy, biases and errors, and required training and skills were also cited by most clusters. Customer satisfaction was mentioned by industries in most clusters as an important, if not the main expected future benefit from using Gen-AI. Most clusters also expected to make efficiency, optimisation and productivity gains. Each industry cluster is now described and illustrated in turn.

4.3.1. Cluster one: the ‘service’ cluster

The ‘service’ cluster was the largest, comprised of nine industries representing over 70% of respondents in the dataset. A range of benefits from current Gen-AI projects were reported and highly ranked by managers in these industries, the most frequent being customer satisfaction and cost savings. Efficiency, optimisation and productivity gains, and time and error reduction were also cited.

For example, a general manager in a US accommodation and food services concern described the multiple benefits of a partially implemented Gen-AI project:

Linking suppliers and distributors so everyone is aware of changes, shortages, and increases. AI is also used in client relations, as a way to track trends, needs, collecting data, and keeping up with other competitors in the market.

Multiple implementation challenges were faced by industries in this cluster, with the most important concerning the quality and availability of data to train models and the accuracy of information outputs and their biases and errors. Industries in this cluster were confronted by seven of the eight challenges observed in the dataset. The main expected future benefit was customer satisfaction, followed by cost savings, decision support, and automation.

4.3.2. Cluster two: the ‘building & support’ cluster

The ‘building & support’ cluster was comprised of three industries and represents 15% of companies in the dataset. The main reported benefits from the use of Gen-AI were greater efficiency, optimisation and productivity, enhanced decision-making, and cost savings.

For example, an architect in the construction industry described the multiple benefits of a Gen-AI project in its trial phase:

Gen-AI is used in my construction industry to optimise planning and scheduling, minimise material waste, and reduce costs. For example, AI algorithms can analyse construction project data to identify potential bottlenecks and optimise the scheduling of resources to ensure timely delivery of materials and completion of the project. Additionally, Gen-AI is also used to optimise inventory levels by forecasting demand, thereby reducing the risk of stock outs and minimising the need for costly safety stock. Overall, the application of Gen-AI helps us improve efficiency and reduce costs, ultimately leading to increased profitability.

The main challenges faced by cluster two industries involved integrating Gen-AI applications with legacy systems, and accuracy, biases and errors in output. Legal and regulatory requirements were also a challenge, as were staff resistance and the need for training and specific skills. The main expected benefits were efficiency gains in supply chain and operations activities, improved customer satisfaction, and a reduction in manual labour time and errors. Cost savings were also expected.

4.3.3. Cluster three: the ‘land use & other’ cluster

Three industries belonged to the ‘land use & other’ cluster, representing 8% of respondents in the dataset. The most frequently cited benefits include improved decision support and automation.

For example, a compliance manager from the US water supply, sewerage, and waste management industry described the automation and decision support benefits from AI:

For our customers, implementing supply chain AI solutions, the benefits of ChatGPT and Gen-AI are vast, including improved automated workflow and decision-making, guided scenarios and simulations, and intelligent risk management.

Similarly, a sales manager in security services reported the decision support and automation benefits from Gen-AI use in the other service activities industry as follows:

Currently our company has implemented Gen-AI to analyse all of the documentation throughout the sales process, from purchasing orders to additional services required and invoicing documentation. This has currently been through two testing phases on certain projects and has now been rolled out as a trial across the business for analysis by the whole team. We are hoping that by utilising this tool we'll be able to identify any patterns that we have been missing previously in order to capitalise on any gaps or take advantage of current market trends. Additionally, we have just begun to utilise an email routing system and automated AI communication system to help with customer-business relations.

The main challenges faced include the accuracy of information produced, and data quality and availability. Ethical concerns and staff resistance were also significant challenges for this cluster. The most important expected gains from the future use of Gen-AI technologies were enhanced decision support and automation. Improved customer satisfaction, cost savings, more accurate information, and better strategic management were also expected.

4.3.4. Cluster four: the ‘arts’ cluster

The fourth cluster (‘arts’) was comprised of companies in the arts, entertainment and recreation industry. The main benefits derived from Gen-AI use included task automation, cost savings and business development.

For example, a US business owner described how a fully implemented Gen-AI project automates operations:

I use ChatGPT to do keyword research. When I have a product in mind that I want to sell I have ChatGPT come up with long-tail keywords to better position my products in Google. I have also used ChatGPT to write descriptions of my products that include keywords. Even more recently I have been using Gen-AI to create graphic mockups to help me visualise how I want certain products to look.

The main challenges faced were implementation costs and staff resistance. Companies in this cluster expected improved business development, process efficiencies, optimisation, and productivity, and enhanced customer satisfaction from the future use of these technologies.

4.3.5. Cluster five: the ‘extraction’ cluster

The ‘extraction’ cluster was comprised of companies in the electricity, gas, steam and air conditioning supply, and the mining and quarrying industries. The main benefits accruing from Gen-AI use were efficiency, optimisation and productivity gains, reduced time and errors, and cost savings. Expected benefits from the future use of Gen-AI included business development opportunities, process efficiencies, and enhanced customer satisfaction.

For example, a sales coordinator from the UK mining and quarrying industry described expected future benefits for customer service from Gen-AI:

We are gradually implementing (or attempting to) AI in operations. By introducing this to our customers we hope they can get a faster and easier update on the status of their jobs with us than by using the conventional customer service route. This service will enable us to provide 24 h updates. We also hope it will have an effect on our profits. And reduce the time wasted by supply chain operations.

The most important challenge facing industries in this cluster was the necessary training and skills to use these technologies. Staff resistance, implementation costs, data quality and accessibility issues, and the accuracy, biases and errors in outputs were also cited as barriers.

4.3.6. Cluster six: the ‘real estate’ cluster

The ‘real estate’ cluster was the sixth and final industry cluster In the dataset. The current benefits reported from the use of Gen-AI included gains in efficiency, optimisation and productivity, increased customer satisfaction, and content creation.

For example, a UK-based property manager described how Gen-AI is currently used for content creation:

Currently using it for marketing purposes. We use it to describe our spaces to potential tenants on Loopnet. We also use it to generate emails that we send to our clients.

Similarly, a US-based operations manager described the productivity gains and cost savings from pilot testing ChatGPT for content creation:

ChatGPT is used in our company to update listing language as well as writing posts for our blog. By utilising AI, we can cut costs and reduce the time needed to generate these types of information.

The main challenges faced by the real estate industry were legal and regulatory requirements, the necessary training and skills, and ethical concerns. The most frequently reported expected benefits were greater content creation, improved decision support, and a reduction in time and errors.

5. Discussion

This study aimed to improve our understanding of the uses, benefits, and challenges of using Gen-AI in OSCM. The exploratory descriptive design, employing open questions and mixed methods of qualitative analysis, allowed us to uncover the main benefits enjoyed and expected by organisations, and the major challenges they face. The following discussion examines our results in light of our research propositions, identifies several scientific and managerial implications, and considers the study’s limitations and opportunities for future research.

5.1. Confrontation of results with research propositions

The results support our first research proposition that initial projects and benefits in OSCM will focus on task automation, efficiencies, and improved customer service. We found that current projects mainly concerned supply chain and inventory management issues, demand forecasting, resource planning, process efficiencies, time and cost savings. Several projects were customer oriented, aimed at improving service, satisfaction, and experience. Our findings show a strong alignment with the emerging academic and professional literature on Gen-AI. For instance, one of the primary objectives of early Gen-AI adopters is greater customer satisfaction through process efficiencies (Fosso Wamba et al. 2023; Hendriksen 2023; Zagorin 2023) and improved services (Young 2023). However, cluster analysis revealed a more nuanced answer to this first proposition: groups of industries differed according to the importance of four current benefits, including automation, cost savings, customer satisfaction, and the use of Gen-AI in decision support.

These results are consistent with diffusion of innovation theory: the ability of Gen-AI to enhance methods and processes relative to existing arrangements is the main driver of adoption. Rogers (1983, 15) defines relative advantage as ‘the degree to which an innovation is perceived as better than the idea it supersedes’. Companies adopted Gen-AI technologies to improve supply chain performance (as defined by Beamon (1999)), by better using resources (e.g. optimised inventory management), producing outputs (e.g. improved resource planning) and reacting to uncertainty (e.g. enhanced demand forecasting).

Other innovation characteristics as described by Rogers (1983) may have also influenced Gen-AI adoption. Firstly, the strong media attention that generative AI has received (Bartholomew and Mehta 2023) may have increased the observability (16) of its capabilities and relative advantages. By reporting on business cases and the results of trials and experimentations, the press and the professional literature has made it easier to ‘observe relative advantages of the new technology’ (Rogers 1983). Secondly, the conversational nature of some generative AI technologies (e.g. ChatGPT) may have reduced the perceived complexity (15) of these new technologies (Rožanec et al. 2023).

We also found support for our second research proposition that current challenges to Gen-AI implementation in OSCM will concern AI models, technologies, and data access. Our results confirmed this proposition and revealed that data challenges (i.e. data quality, data privacy, and data security) and technological reliability were the main concerns of supply chain industry practitioners. Keyword co-occurrence analysis also revealed the importance of employee adoption and use, data quality, biases and ethical considerations. Again, these findings support the AI scholarly and professional literatures. For example, Cubric (2020) conducted a tertiary study of thirty systematic literature reviews on AI adoption in business and management and reported three types of barriers to adoption: economic (e.g. cost), technical (e.g. data availability) and social (e.g. lack of knowledge). While we also found these three types of factors in the top ten challenges faced by organisations, technical and social issues ranked higher than economic factors, with ‘initial time and costs’ only ranked ninth. The predominance of technical and social factors is consistent with the professional and scholarly literature on Gen-AI in OSCM (e.g. Fosso Wamba et al. 2023; Hendriksen 2023), that has notably highlighted the challenges concerning the availability of qualified human resources (e.g. EY 2023).

These challenges may also be explained by diffusion of innovation theory. Potential adopters assess the compatibility of an innovation with ‘existing values, past experiences, and needs’ (Rogers 1983, 15). Challenges posed by legacy systems, employee reactions and ethical concerns may all influence the time that organisations take to ‘break down the organisational and cultural barriers that stand in AI’s way’ (Fountaine, McCarthy, and Saleh 2019, 4). Resolving these issues could enhance ‘the collaboration of human and artificial intelligence’ and move OSCM processes and activities closer to the promises of ‘Industry 5.0’ (Ivanov 2023).

It is also possible that the challenges and benefits of Gen-AI vary with the solutions deployed by organisations. For example, ‘Inspectorio CAPA recommender’ launched as the ‘world’s first generative-AI tool for supply chain management’⁷ has the ability to ‘assist brands, retailers, suppliers, factories and others’ improve the efficiency and effectiveness of their quality control processes. The system identifies the root-causes of defects or non-conformities in products, and makes recommendations for corrective and preventive actions (Mayer 2023). On the other hand, CoPilot, a conversational Gen-AI-enabled freight management platform that allows shippers to generate interactive reports, maps, and charts based on their shipping data may present different benefts and challenges.⁸

The results partially support our third proposition that expected benefits from future Gen-AI projects in OSCM will include gains accruing from business model transformations. Opportunities for business development or new business models were not frequently reported across the dataset. However, they were observed for industries in clusters three, four and five. One explanation for this result may be the mix of Gen-AI maturity levels and experiences between the companies in our dataset. As we are currently in the early use of generative AI technologies, some companies may need to move beyond tests and trials and begin actively using Gen-AI to better envisage possible changes to business models (Wells and Nieuwenhuis 2018).

The majority of expected future benefits were the same as those currently enjoyed by adopters, and mainly concerned operating efficiencies and customer experience. In addition, some organisations expect Gen-AI to improve the sustainability of OSCM processes. This result is consistent with the extant OSCM literature that has reported how the adoption of digital technologies is associated with more sustainable practices and outcomes (Schilling and Seuring 2023). For example, Gen-AI technologies may help reduce the uncertainties related to transitions to circular supply chains (Ghadge, Wurtmann, and Seuring 2020) and the risks of climate change (Schilling and Seuring 2023). According to a recent practitioner study, Gen-AI could enhance supply chain sustainability through improved supplier selection, risk mitigation, product design, and life-cycle assessment (Simons 2023).

Finally, we found support for our fourth research proposition that the rate of adoption of Gen-AI in OSCM will vary by industry. There were several similarities and differences in project maturity between industries: four industries reported that projects were still in the proof-of-concept or prototype phase, four were mainly in the test phase, and six reported that most projects were either fully or partially implemented. This result is consistent with findings in the diffusion of innovation (MacVaugh and Schiavone 2010) and OSCM literatures (Ivanov et al. 2021) that technology adoption varies across industries.

5.2. Research implications

Our study has numerous research implications. Firstly, it is the first to provide rich descriptions of how companies are using Gen-AI in OSCM across multiple industries. Our results contribute to the emerging literature on the adoption of AI (Dora et al. 2022; Merhi and Harfouche 2023; Olan et al. 2022), and Gen-AI technologies (Chang and Kidman 2023; Euchner 2023; Susarla et al. 2023; Zhang 2023). It also responds to recent calls for more empirical studies of Gen-AI adoption and use (Fosso Wamba et al. 2023; Li, Yue, and Zhao 2023).

Secondly, the emerging literature has already highlighted the limited practical and theoretical understanding of Gen-AI systems (Susarla et al. 2023) by scholars, practitioners, and policymakers. Our findings address this gap by providing a rich view of Gen-AI adoption and use. For example, the six industry clusters enrich our understanding of the benefits, challenges, and the future outlook of Gen-AI across different supply chains. Future studies could consider a more structured approach towards creating clusters, based for example on major SCM functions and activities (e.g. inventory management, product life-cycle management) or established operations management theories such as the theory of performance frontiers or the theory of swift (Schmenner and Swink 1998; Seuring 2009).

Thirdly, at a company level, our findings suggest that most Gen-AI OSCM projects focus on operational performance (e.g. demand forecasting, resource planning, process efficiencies) and require different types of resources to meet the challenges of adoption and use (e.g. human skills, organisation infrastructure, AI and data-driven culture). Future research could develop novel frameworks to explain and guide Gen-AI deployment.

Fourthly, our study highlighted industry differences in the timing of Gen-AI adoption and use. Future studies could explore the reasons for these differences. A finer analysis of industry specific adoption processes could improve our understanding of the role that social systems and innovation characteristics play in Gen-AI adoption.

Finally, diffusion of innovation theory could also be used to explain how, why, and at what rate Gen-AI technologies are spreading across different national cultures. Such studies could explore the need for culturally sensitive applications, policies and regulations to encourage or discourage the development of Gen-AI.

5.3. Managerial implications

Our study has several managerial implications. Firstly, our results provide a list of benefits and challenges faced by organisations when adopting Gen-AI. This list may serve as a checklist for managers across industries considering investing in these technologies.

Secondly, our findings suggest that the main benefits for organisations using Gen-AI in OSCM are enhanced customer satisfaction, cost reduction and improved efficiency. While these objectives are aligned with an operational ‘optimisation’ strategy in this first wave of adoption, supply chain managers and decision-makers should also consider the opportunities of using Gen-AI to support their strategic priorities. For instance, managers could use Gen-AI to identify opportunities in new markets, forecast demand, improve sustainability and transform business models.

Thirdly, managers should also consider the challenges of adopting Gen-AI technologies, notably relating to data (i.e. quality, privacy, and security issues) and the technology itself (i.e. model complexity, use of third party platforms). These issues would ideally be addressed during project feasibility studies prior to adoption.

Finally, our findings show different benefits, challenges and maturity levels between industries, alerting decision-makers to carefully consider the industrial specificities of their Gen-AI projects.

5.4. Limitations

The results of this study should be interpreted in light of several limitations. Firstly, given the novelty of generative AI and the strong media attention it has attracted, respondents may have had difficulty clearly defining reasonable future uses and challenges of these technologies. Also, respondents may have confused Gen-AI and traditional-AI based systems. Secondly, as this study was descriptive and exploratory, our research propositions should now be tested with explanatory research designs. Thirdly, data was collected over a short period of time. Given the fast moving nature of these technologies, future studies could use case studies (e.g. Chien et al. 2020) or longitudinal research to capture changes in perceptions and expectations over time.

6. Conclusion

Our study explored and described how and why companies are using Gen-AI in their OSCM, the challenges they face, and the benefits they expect from future use. Our study makes a number of contributions to this emerging field. Firstly, projects were equally spread between early stage experimentation, pilot phase trials and tests, and partial or full implementation. Secondly, in this early stage of Gen-AI, most applications were focused at the operational level, looking to improve customer satisfaction, and achieve cost savings and process efficiencies. Thirdly, our findings suggest that data quality, technological issues and organisational barriers are major challenges for adopting organisations. Finally, we found that the benefits, the challenges, and the rate of Gen-AI adoption in OSCM varies between industries. These findings enrich and advance the OSCM, AI and emerging Gen-AI literatures, and they provide practitioners with a structured understanding of the possible benefits and challenges of Gen-AI adoption.

Our results point to a number of promising avenues for research on Gen-AI in OSCM. Scholars could investigate how Gen-AI can help address OSCM issues of sustainability; how to overcome challenges relating to data quality and access; how to align organisational resources, both technological and human with Gen-AI projects; and how to generate new business value with Gen-AI in OSCM, beyond improved efficiency and customer service. Future research could also examine differences according to firm size, and in perceptions between junior employees and senior executives.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Additional information

Notes on contributors

Samuel Fosso Wamba

Samuel Fosso Wamba is a Full Professor in Information Systems and Data Science and the Associate Dean for Research at TBS Education, Toulouse, France. He is also a Distinguished Visiting Professor at The University of Johannesburg, South Africa. He earned his Ph.D. in industrial engineering at the Polytechnic School of Montreal, Canada. His research interests include information technology adoption and use, business analytics, artificial intelligence for business, the Internet of Things, and metaverse for business.

Cameron Guthrie

Cameron Guthrie is an Associate Professor in decision and system sciences at Toulouse Business School. He earned his Ph.D. in management science at the Université de Paris Panthéon-Sorbonne. His current research focuses on technology adoption, social innovation, wicked problems, and complex problem-solving. He has published papers in various journals, including European Journal of Information Systems, Technological Forecasting and Social Change, Journal of Retailing and Consumer Services, Knowledge Management Research & Practice, Business Process Management Journal, Industrial Relations, and Journal of Management History.

Maciel M. Queiroz

Maciel M. Queiroz is an Associate Professor of Operations and Supply Chain Management at FGV EAESP, Brazil, and Latin/South America Regional Ambassador of the Academy of Management OSCM Division. Maciel is an Associate Editor in the International Journal of Management Reviews and the International Journal of Logistics Management. He ranks in the World's Top 2% of most cited scientists Stanford/Scopus-Elsevier, 2022 and 2023.

Stefan Minner

Stefan Minner is a Full Professor for Logistics and Supply Chain Management at the School of Management, Technical University of Munich (TUM). His research interests are in global supply chain design, transportation optimisation and inventory management using Operations Research and Machine Learning methods. The work was published in many peer reviewed journals, including Management Science, Manufacturing & Service Operations Management, Operations Research, Production and Operations Management, Transportation Science, Transportation Research Part B, European Journal of Operational Research, Computers and Operations Research and OR Spectrum. For his research output, he is currently listed among the top 1% business professors in German speaking countries by Wirtschaftswoche and among the top 2% researchers worldwide in a citation-based ranking by Stanford University. He serves on several editorial boards of logistics and operations journals. Currently, Stefan Minner is the Editor-in-Chief of the International Journal of Production Economics and of Sustainable Manufacturing and Service Economics. Stefan Minner is a fellow of the International Society for Inventory Research (ISIR) and is currently vice-chairman of the scientific advisory board of the German Logistics Association (BVL), and a member of the Research Committee of the European Logistics Association (ELA).

Notes

1 https://edition.cnn.com/2023/01/26/tech/chatgpt-passes-exams/index.html

2 The questionnaire was administered by Prolific in S1 2023.

3 The large blue node beneath the yellow ‘productivity’ node.

4 The daisy() function from the cluster package was used in R to perform the analysis. A dissimilarity matrix was computed using the Gower dissimilarity measure (Gower 1971).

5 p-value < 0.1

6 p-value > 0.1

7 https://inspectorio.com/blog/inspectorio-introduces-first-generative-ai-driven-supply-chain-management-saas-product

8 https://blog.loadsmart.com/loadsmart-introduces-artificial-intelligence-in-its-freight-management-platform

Appendices

Appendix 1. Questionnaire items

The questionnaire included seven demographic questions and five questions pertaining to generative AI projects. The questions are reproduced below.

Q1 to 7: The respondent was invited to provide his or her gender, age, education level, functional area, seniority, management level, and industry.

Q8: At what stage are you with generative AI projects in OSCM?

1. Proof-of-concept

2. Prototype

3. Pilot

4. Trial

5. Partial or phased implementation

6. Full implementation

Q9: Describe your current use of generative AI in OSCM projects (one to two paragraphs).

Q10: What are the current benefits generated by generative AI in your OSCM? List the top five.

Q11: What are the current challenges related to using generative AI in OSCM? List the top five.

Q12: What are the future benefits expected from using generative AI in your OSCM? List the top five.

Appendix 2. Agglometive Hierarchical Clustering (AHC) dendrogram