Electronic commerce (e-commerce) can be defined as activities or services related to buying and selling products or services over the internet (Holsapple & Singh, 2000; Kalakota & Whinston, 1997). Firms increasingly indulge in e-commerce because of customers' rising demand for online services and its ability to create a competitive advantage (Gielens & Steenkamp, 2019; Hamad et al., 2018; Tan et al., 2019). However, firms struggle with this e-business practice due to its integration with rapidly evolving, easily adopted, and highly affordable information technology (IT). This forces firms to constantly adapt their business models to changing customer needs (Gielens & Steenkamp, 2019; Klaus & Changchit, 2019; Tan et al., 2007). Artificial intelligence (AI) is the latest of such technologies. It is transforming e-commerce through its ability to “correctly interpret external data, to learn from such data, and to use those learnings to achieve specific goals and tasks through flexible adaptation” (Kaplan & Haenlein, 2019. p. 15). Depending on the context, AI could be a system, a tool, a technique, or an algorithm (Akter et al., 2021; Bawack et al., 2021; Benbya et al., 2021). It creates opportunities for firms to gain a competitive advantage by using big data to uniquely meet their customers' needs through personalised services (Deng et al., 2019; Kumar, Rajan, et al., 2019; Kumar, Venugopal, et al., 2019).

AI in e-commerce can be defined as using AI techniques, systems, tools, or algorithms to support activities related to buying and selling products or services over the internet. Research on AI in e-commerce has been going on for the past three decades. About 4000 academic research articles have been published on the topic across multiple disciplines, both at the consumer (de Bellis & Venkataramani Johar, 2020; Sohn & Kwon, 2020) and organisational levels (Campbell et al., 2020; Kietzmann et al., 2018; Vanneschi et al., 2018). However, knowledge on the topic has not been synthesised despite its rapid growth and dispersion. This lack of synthesis makes it difficult for researchers to determine how much the extant literature covers concepts of interest or addresses relevant research gaps. Synthesising research on AI in e-commerce is an essential condition for advancing knowledge by providing the background needed to describe, understand, or explain phenomena, to develop/test new theories, and to develop teaching orientations in this research area (Cram et al., 2020; Paré et al., 2015). Thus, this study aims to synthesise research on AI in e-commerce and propose directions for future research in the IS discipline. The innovative approach of combining bibliometric analysis with an extensive literature review is used to answer two specific research questions: (i) what is the current state of research on AI in e-commerce? (ii) what research should be done next on AI in e-commence in general, and within information systems (IS) research in particular?

This study's findings show that AI in e-commerce primarily focuses on recommender systems and the main research themes are sentiment analysis, optimisation, trust, and personalisation. This study makes timely contributions to ongoing debates on the connections between business strategy and the use of AI technologies (Borges et al., 2020; Dwivedi et al., 2019, 2020). It also contributes to research on how firms can address challenges regarding the use of AI-related benefits and opportunities for new product or service developments and productivity improvements (Makridakis, 2017). Furthermore, no study currently synthesises AI in e-commerce research despite its rapid evolution in the last decade triggered by big data, advanced machine learning (ML) algorithms, and cloud computing. Using well-established e-commerce classification frameworks (Ngai & Wat, 2002; Wareham et al., 2005), this study classifies information systems (IS) literature on AI in e-commerce. These classifications make it easier for researchers and managers to identify relevant literature based on the topic area, research style, and research theme. A future research agenda is proposed based on the gaps revealed during the classification to guide researchers on making meaningful contributions to AI knowledge in e-commerce.

Research method

Bibliometric analysis

Bibliometric analysis has been increasingly used in academic research in general and in IS research to evaluate the quality, impact, and influence of authors, journals, and institutions in a specific research area (Hassan & Loebbecke, 2017; Lowry et al., 2004, 2013). It has also been used extensively to understand AI research on specific fields or topics (Hinojo-Lucena et al., 2019; Tran et al., 2019; Zhao, Dai, et al., 2020; Zhao, Lou, et al., 2020). In this study, a bibliometric analysis was conducted to understand research on AI in e-commerce using the approach Aria and Cuccurullo (2017) proposed. This methodology involves three main phases: data collection, data analysis, and data visualisation & reporting. The data collection phase involves querying, selecting, and exporting data from selected databases. This study's data sample was obtained by querying the Web of Science (WoS) core databases for publications from 1975 to 2020. This database was chosen over others like Google Scholar or Scopus because WoS provides better quality bibliometric information due to its lower rate of duplicate records (Aria et al., 2020) and greater coverage of high-impact journals (Aghaei Chadegani et al., 2013). The following search string was used to query the title, keywords, and abstracts of all documents in the WoS collection:

(‘‘Electronic Commerce’’ OR ‘‘Electronic business’’ OR ‘‘Internet Commerce’’ OR “e-business” OR “ebusiness” OR "e-commerce” OR “ecommerce” OR “online shopping” OR “online purchase” OR “internet shopping” OR “e-purchase” OR “online store” OR “electronic shopping”).

AND (“Artificial intelligence” OR “Artificial neural network” OR “case-based reasoning” OR “cognitive computing” OR “cognitive science” OR “computer vision” OR “data mining” OR “data science” OR “deep learning” OR “expert system” OR “fuzzy linguistic modelling” OR “fuzzy logic” OR “genetic algorithm” OR “image recognition” OR “k-means” OR “knowledge-based system” OR “logic programming” OR “machine learning” OR “machine vision” OR “natural language processing” OR “neural network” OR “pattern recognition” OR “recommendation system” OR “recommender system” OR “semantic network” OR “speech recognition” OR “support vector machine” OR “SVM” OR “text mining”).

This search string led to 4414 documents that made up the initial dataset of this study. For quality reasons, only document types tagged as articles, reviews, and proceeding papers were selected for this study because they are most likely to have undergone a rigorous peer-review process before publication (Milian et al., 2019). Thus, editorial material, letters, news items, meeting abstracts, and retracted publications were removed from the dataset, leaving 4335 documents that made up the final dataset used for bibliometric analysis. Figure 1 summarises the data collection phase.

Fig. 1
figure 1

Summary of the data collection phase

Table 1 summarises the main information about the dataset regarding the timespan, document sources, document types, document contents, authors, and author collaborations. The dataset consists of documents from 2599 sources, published by 8663 authors and 84,474 references.

Table 1 Main information about the dataset

BibliometrixFootnote 1 is the R package used to conduct bibliometric analysis (Aria & Cuccurullo, 2017). This package has been extensively used to conduct bibliometric studies published in top-tier journals. It incorporates the most renowned bibliometric tools for citation analysis (Esfahani et al., 2019; Fosso Wamba, 2020; Pourkhani et al., 2019). It was specifically used to analyse the sources, documents, conceptual, and intellectual structure of AI in e-commerce research. Publication sources and their source impacts were analysed based on their h-index quality factors (Hirsch, 2010). The most significant, impactful, prestigious, influential, and quality publication sources, affiliations, and countries regarding research on AI in e-commerce were identified. This contributed to the identification of the most relevant disciplines in this area of research. Documents were analysed using total citations to identify the most cited documents in the dataset. Through content analysis, the most relevant topics/concepts, AI technologies/techniques, research methods, and application domains were identified.

Furthermore, citation analysis and reference publication year spectroscopy (RPYS) were used to identify research contributions that form the foundations of research on AI in e-commerce (Marx et al., 2014; Rhaiem & Bornmann, 2018). These techniques were also used to identify the most significant changes in the research area. Co-word network analysis on author-provided keywords using the Louvain clustering algorithm was used to understand the research area's conceptual structure. This algorithm is a greedy optimisation method used to identify communities in large networks by comparing the density of links inside communities with links between communities (Blondel et al., 2008). This study used it to identify key research themes by analysing author-provided keywords. Co-citation network analysis using the Louvain clustering algorithm was also used to analyse publication sources through which journals communities were identified. It further contributed to identifying the most relevant disciplines in this research area by revealing journal clusters.

The bibliometric analysis results were reported from functionalist, normative, and interpretive perspectives (Hassan & Loebbecke, 2017). The functionalist perspective presents the results of the key concepts and topics investigated in this research area. The normative perspective focuses on the foundations and norms of the research area. The interpretive perspective emphasises the main themes that drive AI in e-commerce research.

Literature review

An extensive review and classification of IS literature on AI in e-commerce complemented the bibliometric analysis. It provides more details on how research in this area is conducted in the IS discipline. The review was delimited to the most impactful and influential management information systems (MIS) journals identified during the bibliometric analysis and completed by other well-established MIS journals known for their contributions to e-commerce research (Ngai & Wat, 2002; Wareham et al., 2005). Thus, 20 journals were selected for this review: Decision sciences, Decision support systems, Electronic commerce research and applications, Electronic markets, E-service journal, European journal of information systems, Information and management, Information sciences, Information systems research, International journal of electronic commerce, International journal of information management, Journal of information systems, Journal of information technology, Journal of management information systems, Journal of organisational computing and electronic commerce, Journal of strategic information systems, Journal of the association for information systems, Knowledge-based systems, Management science, MIS Quarterly.

The literature review was conducted in three stages (Templier & Paré, 2015; Webster & Watson, 2002): (i) identify and analyse all relevant articles from the targeted journals found in the bibliometric dataset (ii) use the keyword string to search for other relevant articles found on the official publication platforms of the targeted journals, and (iii) identify relevant articles from the references of the articles identified in stages one and two found within the target journals. All articles with content that did not focus on AI in e-commerce were eliminated. This process led to a final dataset of 229 research articles on AI in e-commerce. The articles were classified into three main categories: by topic area (Ngai & Wat, 2002), by research style (Wareham et al., 2005), and by research themes (from bibliometric analysis).

Classification by topic area involved classifying relevant literature into four broad categories: (i) applications, (ii) technological issues, (iii) support and implementation, and (iv) others. Applications refer to the specific domain in which the research was conducted (marketing, advertising, retailing…). Technological issues contain e-commerce research by AI technologies, systems, algorithms, or methodologies that support or enhance e-commerce applications. Support and Implementation include articles that discuss how AI supports public policy and corporate strategy. Others contain all other studies that do not fall into any of the above categories. It includes articles on foundational concepts, adoption, and usage. Classification by research style involved organizing the relevant literature by type of AI studied, the research approach, and the research method used in the studies. The research themes identified in the bibliometric analysis stage were used to classify the relevant IS literature by research theme.


Results of the bibliometric analysis

Scientific publications on AI in e-commerce began in 1991 with an annual publication growth rate of 10.45%. Figure 2 presents the number of publications per year. Observe the steady increase in the number of publications since 2013.

Fig. 2
figure 2

Number of publications on AI in e-commerce per year

Institutions in Asia, especially China, are leading this research area. The leading institution is Beijing University of Posts and Telecommunications, with 88 articles, followed by Hong Kong Polytechnic University with 84 articles. Table 2 presents the top 20 institutions publishing on AI in e-commerce.

Table 2 Top 20 institutions publishing on AI in e-commerce

As expected, China-based affiliations appear most frequently in publications (4261 times). They have over 2.5 times as many appearances as US-based affiliations (1481 times). Interestingly, publications with US-based affiliations attract more citations than those in China. Table 3 presents the number of times authors from a given country feature in publications and the corresponding total number of citations.

Table 3 Top 20 countries contributing to research on AI in e-commerce

Functional perspective

Analysing the most globally cited documentsFootnote 2 in the dataset (those with 100 citations) reveals that recommender systems are the main topic of interest in this research area (Appendix Table 10). Recommender systems are software agents that make recommendations for consumers by implicitly or explicitly evoking their interests or preferences (Bo et al., 2007). The topic has been investigated in many flavours, including hybrid recommender systems (Burke, 2002), personalised recommender systems (Cho et al., 2002), collaborative recommender systems (Lin et al., 2002) and social recommender systems (Li et al., 2013). The central concept of interest is personalisation, specifically leveraging recommender systems to offer more personalised product/service recommendations to customers using e-commerce platforms. Thus, designing recommender systems that surpass existing ones is the leading orientation of AI in e-commerce research. Researchers have mostly adopted experimental rather than theory-driven research designs to meet this overarching research objective. Research efforts focus more on improving the performance of recommendations using advanced AI algorithms than on understanding and modelling the interests and preferences of individual consumers. Nevertheless, the advanced AI algorithms developed are trained primarily using customer product reviews.

Interpretive perspective

Four themes characterise research on AI in e-commerce: sentiment analysis, trust & personalisation, optimisation, AI concepts, and related technologies. The keyword clusters that led to the identification of these themes are presented in Table 4. The sentiment analysis theme represents the stream of research focused on interpreting and classifying emotions and opinions within text data in e-commerce using AI techniques like ML and natural language processing (NLP). The trust and personalisation theme represents research that focuses on establishing trust and making personalised recommendations for consumers in e-commerce using AI techniques like collaborative filtering, case-based reasoning, and clustering algorithms. The optimisation theme represents research that focuses on using AI algorithms like genetic algorithms to solve optimisation problems in e-commerce. Finally, the AI concepts and related technologies theme represent research that focuses on using different techniques and concepts used in the research area.

Table 4 Main research themes for AI in e-commerce

Normative perspective

Research on AI in e-commerce is published in two main journal subject areas: computer science & AI and business & management. This result confirms the multidisciplinary nature of this research area, which has both business and technical orientations. Table 5 presents the most active publication outlets in each subject area. The outlets listed in the table could help researchers from different disciplines to select the proper outlet for their research results. It could also help researchers identify the outlets wherein they are most likely to find relevant information for their research on AI in e-commerce.

Table 5 Publication structure of research on AI in e-commerce

However, some disciplines set the foundations and standards of research on AI in e-commerce through the impact of their contributions to its body of knowledge. Analysing document references shows that the most cited contributions come from journals in the IS, computer science, AI, management science, and operations research disciplines (Table 6). It shows the importance of these disciplines to AI's foundations and standards in e-commerce research and their major publication outlets.

Table 6 Most local cited sources on AI in e-commerce

The IS discipline is a significant contributor to AI in e-commerce research, given that 24 out of the 40 top publications in the area can be assimilated to IS sources. Table 7 also shows that 7 out of the top 10 most impactful publication sources are assimilated to the IS discipline. The leading paper from the IS field reviews approaches to automatic schema matching (Rahm & Bernstein, 2001) and it is the second most globally cited paper in the research area. Meanwhile, the leading paper from the MIS subfield reviews recommender system application developments (Lu et al., 2015).

Table 7 Publication source impact by h-index

Collaborative filtering, recommender systems, social information filtering, latent Dirichlet allocation, and matrix factoring techniques are the foundational topics in research on AI in e-commerce (Table 8). They were identified by analysing the most cited references in the dataset. These references were mostly literature reviews and documents that discussed the basic ideas and concepts behind specific technologies or techniques used in recommender systems.

Table 8 Foundational studies on AI in e-commerce

Furthermore, the specific documents that set the foundations of research on AI in e-commerce and present the most significant historical contributions and turning points in the field were identified using RPYS (Appendix Table 11). 2001, 2005, 2007, 2011, and 2015 are the years with the highest number of documents referenced by the documents in the sample. The most cited studies published in 2001 focused on recommendation algorithms, especially item-based collaborative filtering, random forest, gradient boosting machine, and data mining. The main concept of interest was how to personalise product recommendations. In 2005, the most referenced documents focused on enhancing recommendation systems using hybrid collaborative filtering, advanced machine learning tools and techniques, and topic diversification. That year also contributed a solid foundation for research on trust in recommender systems. In 2007, significant contributions continued on enhanced collaborative filtering techniques for recommender systems. Meanwhile, Bo & Benbasat (2007) set the basis for research on recommender systems' characteristics, use, and impact, shifting from traditional studies focused on underlying algorithms towards a more consumer-centric approach. In 2011, major contributions were made to enhance recommender systems, like developing a new library for support vector machines (Chang & Lin, 2011) and the Scikit-learn package for machine learning in Python (Pedregosa et al., 2011). In 2015, the most critical contributions primarily focused on deep learning algorithms, especially with an essential contribution to using them in recommender systems (Wang et al., 2015).

Results of the literature review study

Classification by topic area

Most articles on AI in e-commerce focus on technological issues (107 articles, 47%), followed by applications (87 articles, 38%), support and implementation (20 articles, 9%), then others (15 articles, 6%). Specifically, most articles focus on AI algorithms, models, and methodologies that support or improve e-commerce applications (76 articles, 33.2%) or emphasise the applications of AI in marketing, advertising, and sales-related issues (38 articles, 16.6%). Figure 3 presents the distribution of articles, while Appendix Table 12 presents the articles in each topic area.

Fig. 3
figure 3

Classification of MIS literature on AI in e-commerce by topic area

Classification by research style

Most authors discuss AI algorithms, models, computational approaches, or methodologies (168 articles, 73%). Specifically, current research focuses on how AI algorithms like ML, deep learning (DL), NLP, and related techniques could be used to model and understand phenomena in the e-commerce environment. It also focuses on studies that involve designing intelligent agent algorithms that support learning processes in e-commerce systems. Many studies also focus on AI as systems (31 articles, 14%), especially on recommender systems and expert systems that leverage AI algorithms in the back end. The “others” category harboured all articles that did not clearly refer to AI as either an algorithm or as a system (30 articles, 13%) (see Fig. 4 and Appendix Table 13).

Fig. 4
figure 4

Classification of MIS literature on AI in e-commerce by type of AI

Furthermore, most publications use the design science research approach (198 articles, 86%). Researchers prefer this approach because it allows them to develop their algorithms and models or improve existing ones, thereby creating a new IS artefact (see Fig. 5 and Appendix Table 14).

Fig. 5
figure 5

Classification of MIS literature on AI in e-commerce by research approach

Also, authors adopt experimental methods in their papers (157 articles, 69%), especially those who adopted a design science research approach. They mostly use experiments to test their algorithms or prove their concepts (see Fig. 6 and Appendix Table 15).

Fig. 6
figure 6

Classification of MIS literature on AI in e-commerce by research method

Classification by research theme

Based on the main research themes on AI in e-commerce identified during the bibliometric analysis, most authors published on optimisation (63 articles, 27%). They mostly focused on optimising recommender accuracy (25 articles), prediction accuracy (29 articles), and other optimisation aspects (9 articles) like storage optimisation. This trend was followed by publications on trust & personalisation (31 articles, 14%), wherein more articles were published on personalisation (17 articles) than on trust (14 articles). Twenty-nine articles focused on sentiment analysis (13%). The rest of the papers focus on AI design, tools and techniques (46 articles), decision support (30 articles), customer behaviour (13 articles), AI concepts (9 articles), and intelligent agents (8 articles) (see Fig. 7 and Appendix Table 16).

Fig. 7
figure 7

Classification of MIS literature on AI in e-commerce by current research themes


This study's overall objective was to synthesise research on AI in e-commerce and propose avenues for future research. Thus, it sought to answer two research questions: (i) what is the current state of research on AI in e-commerce? (ii) what research should be done next on AI in e-commerce in general and within IS research in particular? This section summarises the findings of the bibliometric analysis and literature review. It highlights some key insights from the results, starting with the leading role of China and the USA in this research area. This highlight is followed by discussions on the focus of current research on recommender systems, the extensive use of design science and experiments in this research area, and a limited focus on modelling consumer behaviour. This section also discusses the little research found on some research themes and the limited number of publications from some research areas. Implications for research and practice are discussed at the end of this section.

Need for more research from other countries

Research on AI in e-commerce has been rising steadily since 2013. Overall, these results indicate a growing interest in the applications of AI in e-commerce. China-based institutions lead this research area, although US-based affiliations attract more citations. Tables 2 and 3 indicate that China is in the leading position regarding research on AI in e-commerce. Observe that Amazon Inc. (USA), (China), Alibaba Group Holding Ltd. (China), (China), Meituan (China), Wayfair (USA), eBay (USA), and Groupon (USA) are referenced among the largest e-commerce companies in the world (in terms of market capitalisation, revenue, and the number of employees).Footnote 3 These companies are primarily from China and the USA. These findings correlate with Table 3, which could indicate that China and the USA are investing more in the research and development of AI applications in e-commerce (especially China, based on Table 2) because of the positions they occupy in the industry. This logic would imply that companies seeking to penetrate the e-commerce industry and remain competitive should also consider investing more in the research and development of AI applications in the area. The list of universities provided could become partner universities for countries with institutions that have less experience in the research area. Especially with the COVID-19 pandemic, e-commerce has become a global practice. Thus, other countries need to contribute more research on the realities of e-commerce in their respective contexts to develop more globally acceptable AI solutions in e-commerce practices. It is essential because different countries approach e-commerce differently. For example, although Amazon’s marketplace is well-developed in continents like Europe, Asia, and North America, it has difficulty penetrating Africa because the context is very different (culturally and infrastructurally). While mobile wallet payment systems are fully developed on the African continent, Amazon’s marketplace does not accommodate this payment method. Therefore, it would be impossible for many Africans to use Amazon’s Alexa to purchase products online. What does this mean for research on digital inclusion? Are there any other cross-cultural differences between countries that affect the adoption and use of AI in e-commerce? Are there any legal boundaries that affect the implementation and internationalisation of AI in e-commerce? Such questions highlight the need for more country-specific research on AI in e-commerce to ensure more inclusion.

Focus on recommender systems

AI in e-commerce research is essentially focused on recommender systems in the past years. The results indicate that in the last 20 years, AI in e-commerce research has primarily focused on using AI algorithms to enhance recommender systems. This trend is understandable because recommender systems have become an integral part of almost every e-commerce platform nowadays (Dokyun Lee & Hosanagar, 2021; Stöckli & Khobzi, 2021). As years go by, observe how novel AI algorithms have been proposed, the most recent being deep learning (Chaudhuri et al., 2021; Liu et al., 2020; Xiong et al., 2021; Zhang et al., 2021). Thus, researchers are increasingly interested in how advanced AI algorithms can enable recommender systems in e-commerce platforms to correctly interpret external data, learn from such data, and use those learnings to improve the quality of user recommendations through flexible adaptation. With the advent of AI-powered chatbots and voice assistants, firms increasingly include these technologies in their e-commerce platforms (Ngai et al., 2021). Thus, researchers are increasingly interested in conversational recommender systems (De Carolis et al., 2017; Jannach et al., 2021; Viswanathan et al., 2020). These systems can play the role of recommender systems and interact with the user through natural language (Iovine et al., 2020). Thus, conversational recommender systems is an up-and-coming research area for AI-powered recommender systems, especially given the ubiquitous presence of voice assistants in society today. Therefore, researchers may want to investigate how conversational recommender systems can be designed effectively and the factors that influence their adoption.

Limited research themes

The main research themes in AI in e-commerce are sentiment analysis, trust, personalisation, and optimisation. Researchers have focused on these themes to provide more personalised recommendations to recommendation system users. Personalising recommendations based on users’ sentiment and trust circle has been significantly researched. Extensive research has also been conducted on how to optimise the algorithmic performance of recommender systems. ML, DL, NLP are the leading AI algorithms and techniques currently researched in this area. The foundational topics for applying these algorithms include collaborative filtering, latent Dirichlet allocation, matrix factoring techniques, and social information filtering.

Current research shows how using AI for personalisation would enable firms to deliver high-quality customer experiences through precise personalisation based on real-time information (Huang & Rust, 2018, 2020). It is highly effective in data-rich environments and can help firms to significantly improve customer satisfaction, acquisition, and retention rates, thereby ideal for service personalisation (Huang & Rust, 2018). AI could enable firms to personalise products based on preferences, personalise prices based on willingness to pay, personalise frontline interactions, and personalise promotional content in real-time (Huang & Rust, 2021).

Research also shows how AI could help firms optimise product prices by channel and customer (Huang & Rust, 2021; Huang & Rust, 2020) and develop accurate and personalised recommendations for customers. It is beneficial when the firm lacks initial data on customers that it can use to make recommendations (cold start problem) (Guan et al., 2019; Wang, Feng, et al., 2018; Wang, Jhou, et al., 2018; Wang, Li, et al., 2018; Wang, Lu, et al., 2018). It also gives firms the ability to automatically estimate optimal prices for their products/services and define dynamic pricing strategies that increase profits and revenue (Bauer & Jannach, 2018; Greenstein-Messica & Rokach, 2018). It also gives firms the ability to predict consumer behaviours like customer churn (Bose & Chen, 2009), preferences based on their personalities (Buettner, 2017), engagement (Ayvaz et al., 2021; Sung et al., 2021; Yim et al., 2021), and customer payment default (Vanneschi et al., 2018). AI also gives firms the ability to predict product, service, or feature demand and sales (Cardoso & Gomide, 2007; Castillo et al., 2017; Ryoba et al., 2021), thereby giving firms the ability to anticipate and dynamically adjust their advertising and sales strategies (Chen et al., 2014; Greenstein-Messica & Rokach, 2020). Even further, it gives firms the ability to predict the success or failure of these strategies (Chen & Chung, 2015).

Researchers have shown that using AI to build trust-based recommender systems can help e-commerce firms increase user acceptance of the recommendations made by e-commerce platforms (Bedi & Vashisth, 2014). This trust is created by accurately measuring the level of trust customers have in the recommendations made by the firm’s e-commerce platforms (Fang et al., 2018) or by making recommendations based on the recommendations of people the customers’ trust in their social sphere (Guo et al., 2014; Zhang et al., 2017).

Sentiment analytics using AI could give e-commerce firms the ability to provide accurate and personalised recommendations to customers by assessing their opinions expressed online such as through customer reviews (Al-Natour & Turetken, 2020; Qiu et al., 2018). It has also proven effective in helping brands better understand their customers over time and predict their behaviours (Das & Chen, 2007; Ghiassi et al., 2016; Pengnate & Riggins, 2020). For example, it helps firms better understand customer requirements for product improvements (Ou et al., 2018; Qi et al., 2016) and predict product sales based on customer sentiments (Li, Wang, et al., 2019; Li, Wu, et al., 2019; Li, Zhang, et al., 2019). Thus, firms can accurately guide their customers towards discovering desirable products (Liang & Wang, 2019) and predict the prices they would be willing to pay for products based on their sentiments (Tseng et al., 2018). Thus, firms that use AI-powered sentiment analytics would have the ability to constantly adapt their product development, sales, and pricing strategies while improving the quality of their e-commerce services and personalised recommendations for their customers.

While the current research themes are exciting and remain relevant in today’s context, it highlights the need for researchers to explore other research themes. For example, privacy, explainable, and ethical AI are trendy research themes in AI research nowadays. These themes are relevant to research on AI in e-commerce as well. Thus, developing these research themes would make significant contributions to research on AI in e-commerce. In the IS discipline, marketing & advertising is where AI applications in e-commerce have been researched the most. This finding complements Davenport et al. (2020)’s argument, suggesting that marketing functions have the most to gain from AI. Most publications focus on technological issues like algorithms, support systems, and security. Very few studies investigated privacy, and none was found on topics like ethical, explainable, or sustainable AI. Therefore, future research should pay more attention to other relevant application domains like education & training, auctions, electronic payment systems, inter-organisational e-commerce, travel, hospitality, and leisure (Blöcher & Alt, 2021; Manthiou et al., 2021; Neuhofer et al., 2021). To this end, questions that may interest researchers include, what are the privacy challenges caused by using AI in e-commerce? How can AI improve e-commerce services in education and training? How can AI improve e-commerce services in healthcare? How can AI bring about sustainable e-commerce practices?

Furthermore, research on AI in e-commerce is published in two main journal categories: computer science & AI and business & management. Most citations come from the information systems, computer science, artificial intelligence, management science, and operations research disciplines. Thus, researchers interested in research on AI in e-commerce are most likely to find relevant information in such journals (see Tables 5 and 6). Researchers seeking to publish their research on AI in e-commerce can also target such journals. However, researchers are encouraged to publish their work in other equally important journal categories. For example, law and government-oriented journals would greatly benefit from research on AI in e-commerce. International laws and government policies could affect how AI is used in e-commerce. For example, due to the General Data Protection Regulation (GDPR), how firms use AI algorithms and applications to analyse user data in Europe may differ from how they would in the US. Such factors may have profound performance implications given that AI systems are as good as the volume and quality of data they can access for analysis. Thus, future research in categories other than those currently researched would benefit the research community.

More experiment than theory-driven research

Most of the research done on AI in e-commerce have adopted experimental approaches. Very few adopted theory-driven designs. This trend is also observed in IS research, where 69% of the studies used experimental research methods and 86% adopted a design science research approach instead of the positivist research approach often adopted in general e-commerce research (Wareham et al., 2005). However, this study's findings complement a recent review that shows that laboratory experiments and secondary data analysis were becoming increasingly popular in e-commerce research. Given that recommender systems support customer decision-making, this study also complements recent studies that show the rising use of design science research methods in decision support systems research (Arnott & Pervan, 2014) and in IS research in general (Jeyaraj & Zadeh, 2020). This finding could be explained by the fact that researchers primarily focused on enhancing the performance of AI algorithms used in recommender systems. Therefore, to test the performance of their algorithms in the real world, the researchers have to build a prototype and test it in real-life contexts. Using performance accuracy scores, the researchers would then tell the extent to which their proposed algorithm is performant. However, ML has been highlighted as a powerful tool that can help advance theory in behavioural IS research (Abdel-Karim et al., 2021). Therefore, key research questions on AI in e-commerce could be approached using ML as a tool for theory testing in behavioural studies. Researchers could consider going beyond using AI algorithms for optimising recommender systems to understand its users' behaviour. In Fig. 4, observe that 73% of IS researcher papers reviewed approached AI as an algorithm or methodology to solve problems in e-commerce. Only 14% approached AI as a system. Researchers can adopt both approaches in the same study in the sense that they can leverage ML algorithms to understand human interactions with AI systems, not just for optimisation. This approach could provide users with insights by answering questions regarding the adoption and use of AI systems.

Furthermore, only 6% of the studies focus on consumer behaviour. Thus, most researchers on AI in e-commerce this far have focused more on algorithm performance than on modelling the behaviour of consumers who use AI systems. It is also clear that behavioural aspects of using recommender systems are often overlooked (Adomavicius et al., 2013). There is relatively limited research on the adoption, use, characteristics, and impact of AI algorithms or systems on its users. This issue was raised as a fundamental problem in this research area (Bo et al., 2007) and seems to remain the case today. However, understanding consumer behaviour could help improve the accuracy of AI algorithms. Thus, behavioural science researchers need to conduct more research on modelling consumer behaviours regarding consumers' acceptance, adoption, use, and post-adoption behaviours targeted by AI applications in e-commerce. As AI algorithms, systems, and use cases multiply in e-commerce, studies explaining their unique characteristics, adoption, use, and impact at different levels (individual, organisational, and societal) should also increase. It implies adopting a more theory-driven approach to research on AI in e-commerce. Therefore, behavioural science researchers should be looking into questions on the behavioural factors that affect the adoption of AI in e-commerce.

Implications for research

This study contributes to research by innovatively synthesising the literature on AI in e-commerce. Despite the recent evolution of AI and the steady rise of research on how it could affect e-commerce environments, no review has been conducted to understand this research area's state and evolution. Yet, a recent study shows that e-commerce and AI are currently key research topics and themes in the IS discipline (Jeyaraj & Zadeh, 2020). This paper has attempted to fill this research gap by providing researchers with a global view of AI research in e-commerce. It offers a multidimensional view of the knowledge structure and citation behaviour in this research area by presenting the study's findings from functional, normative, and interpretive perspectives. Specifically, it reveals the most relevant topics, concepts, and themes on AI in e-commerce from a multidisciplinary perspective.

This contribution could help researchers evaluate the value and contributions of their research topics in the research area with respect to other disciplines and choose the best publication outlets for their research projects. This study also reveals the importance of AI in designing recommender systems and shows the foundational literature on which this research area is built. Thus, researchers could use this study to design the content of AI or e-commerce courses in universities and higher education institutions. Its content provides future researchers and practitioners with the foundational knowledge required to build quality recommender systems. Researchers could also use this study to inform their fields on the relevance of their research topics and the specific gaps to fill therein. For example, this study reveals the extent to which the IS discipline has appropriated research on AI in e-commerce. It also shows contributions of the IS discipline to the current research themes, making it easier for IS researchers to identify research gaps as well as gaps between IS theory and practice.

Implications for practice

This study shows that AI in e-commerce primarily focuses on recommender systems. It highlights sentiment analysis, optimisation, trust, and personalisation as the core themes in the research area. Thus, managers could tap into these resources to improve the quality of their recommender systems. Specifically, it could help them understand how to develop optimised, personalised, trust-based and sentiment-based analytics supported by uniquely designed AI algorithms. This knowledge would make imitating or replicating the quality of recommendations rendered through e-commerce platforms practically impossible for competitors. Firms willing to use AI in e-commerce would need unique access and ownership of customer data, AI algorithms, and expertise in analytics (De Smedt et al., 2021; Kandula et al., 2021; Shi et al., 2020). The competition cannot imitate these resources because they are unique to the firm, especially if patented (Pantano & Pizzi, 2020). Also, this research paper classifies IS literature on AI in e-commerce by topic area, research style, and research theme. Thus, IS practitioners interested in implementing AI in e-commerce platforms would easily find the research papers that best meet their needs. It saves them the time to search for articles themselves, which may not always be relevant and reliable.


This study has some limitations. It was challenging to select a category for each article in the sample dataset. Most of those articles could be rightfully placed in several categories of the classification frameworks. However, assigning articles to a single category in each framework simplifies the research area's conceptualisation and understanding (Wareham et al., 2005). Thus, categories were assigned to each article based on the most apparent orientation from the papers' titles, keywords, and abstracts. Another challenge was whether or not to include a research paper in the review. For example, although some studies on recommender systems featured in the keyword search results, the authors did not specify if the system's underlying algorithms were AI algorithms. Consequently, such articles were not classified to ensure that those included in this review certainly had an AI orientation. Despite our efforts, we humbly acknowledge that this study may have missed some publications, and others may have been published since this paper started the review process. Thus, in no way does this study claim to be exhaustive but rather extensive. Nonetheless, the findings from our rigorous literature review process strongly match the bibliometric analysis findings and those from similar studies we referenced. Therefore, we believe our contributions to IS research on AI in e-commerce remain relevant.

Future research

In addition to recommendations for future research discussed in the previous sections, the findings of this study are critically analysed through the lens of recent AI research published in leading IS journals. The aim is to identify other potential gaps for future research on AI in e-commerce that could interest the IS community.

One of the fundamental issues with AI research in IS today is understanding the AI concept (Ågerfalk, 2020). Our findings show that researchers have mostly considered algorithms and techniques like ML, DL, and NLP AI in their e-commerce research. Are these algorithms and techniques AI? Does the fact that an algorithm helps to analyse data and make predictions about e-commerce activities mean that the algorithm is AI? It is crucial for researchers to clearly explain what they mean by AI and differentiate between different types of AI used in their studies to avoid ambiguity. This explanation would help prevent confusion between AI and business intelligence & analytics in e-commerce. It would also help distinguish between AI as a social actor and AI as a technology with the computational capability to perform cognitive functions.

A second fundamental issue with AI research in IS is context (Ågerfalk, 2020). Using the same data, an AI system would/should be able to interpret the message communicated or sought by the user based on context. Context gives meaning to the data, making the AI system’s output relevant in the real world. Research on AI in e-commerce did not show much importance to context. Many authors used existing datasets to test their algorithms without connecting them to a social context. Thus, it is difficult to assess whether the performance of the proposed algorithms is relevant in every social context. Future research should consider using AI algorithms to analyse behavioural data alongside ‘hard’ data (facts) to identify patterns and draw conclusions in specific contexts. It implies answering the crucial question, what type of AI best suits which e-commerce context? Thus, researchers would need to collaborate with practitioners to better understand and delineate contexts (Ågerfalk, 2020) of investigation rather than make general claims on fraud detection or product prices, for example.

The IS community is also interested in understanding ethical choices and challenges organisations face when adopting AI systems and algorithms. What ethical decisions do e-commerce firms need to make when implementing AI solutions? What are the ethical challenges e-commerce firms face when implementing AI solutions? Following a sociotechnical approach, firms seeking to implement AI systems need to make ethical choices. These include transparent vs black-boxed algorithms, slow & careful vs expedited & timely designs, passive vs active implementation approach, obscure vs open system implementation, compliance vs risk-taking, and contextualised vs standardised use of AI systems (Marabelli et al., 2021). Thus, future research on AI in e-commerce should investigate how e-commerce firms address these ethical choices when implementing their AI solutions and the challenges they face in the process.

AI and the future of work is another primary source of controversy in the IS community (Huysman, 2020; Willcocks, 2020a, b). Several researchers are investigating how AI is transforming the work configurations of organisations. Workplace technology platforms are increasingly observed to integrate office applications, social media features and AI-driven self-learning capabilities (Baptista et al., 2020; Grønsund & Aanestad, 2020; Lyytinen et al., 2020). Is this emergent digital/human work configuration also happening in e-commerce firms? How is this changing the future of work in the e-commerce industry?

IS researchers have increasingly called for research on how AI transforms decision making. For example, they are interested in understanding how AI could help augment mental processing, change managerial mindsets and actions, and affect the rationality of economic agents (Brynjolfsson et al., 2021). A recent study also makes several research propositions for IS researchers regarding conceptual and theoretical development, AI-human interaction, and AI implementation in the context of decision making (Duan et al., 2019). This study shows that decision-making is not a fundamental research theme as it accounts for only 13% of the research papers reviewed. Thus, future research on AI in e-commerce should contribute to developing this AI research theme in the e-commerce context. It involves proposing answers to questions like how AI affects managerial mindsets and actions in e-commerce? How is AI affecting the rationality of consumers who use e-commerce platforms?

This study shows that relatively few research papers on AI in e-commerce are theory-driven. Most adopted experimental research methods and design science research approaches wherein they use AI algorithms to explain phenomena. The IS community is increasingly interested in developing theories using AI algorithms (Abdel-Karim et al., 2021). Contrary to traditional theory development approaches, such theories developed based on AI algorithms like ML are called to be focused, context-specific, and as transparent as possible (Chiarini Tremblay et al., 2021). Thus, rather than altogether abandoning the algorithm-oriented approach used for AI in e-commerce research, researchers who master it should develop skills to use it as a basis for theorising.

Last but not least, more research is needed on the role of AI-powered voice-based AI in e-commerce. It is becoming common for consumers to use intelligent personal assistants like Google’s Google Assistant, Amazon’s Alexa, and Apple’s Siri for shopping activities since many retail organisations are making them an integral part of their e-commerce platforms (de Barcelos Silva et al., 2020). Given the rising adoption of smart speakers by consumers, research on voice commerce should become a priority for researchers on AI in e-commerce. Yet, this study shows that researchers are still mostly focused on web-based, social networking (social commerce), and mobile (m-commerce) platforms. Therefore, research on the factors that affect the adoption and use of voice assistants in e-commerce and the impact on consumers and e-commerce firms would make valuable contributions to e-commerce research. Table 9 summarises the main research directions recommended in this paper.

Table 9 Future research questions for AI in e-commerce research


AI has emerged as a technology that can differentiate between two competing firms in e-commerce environments. This study presents the state of research of AI in e-commerce based on bibliometric analysis and a literature review of IS research. The bibliometric analysis highlights China and the USA as leaders in this research area. Recommender systems are the most investigated technology. The main research themes in this area of research are optimisation, trust & personalisation, sentiment analysis, and AI concepts & related technologies. Most research papers on AI in e-commerce are published in computer science, AI, business, and management outlets. Researchers in the IS discipline has focused on AI applications and technology-related issues like algorithm performance. Their focus has been more on AI algorithms and methodologies than AI systems. Also, most studies have adopted a design science research approach and experiment-style research methods. In addition to the core research themes of the area, IS researchers have also focused their research on AI design, tools and techniques, decision support, consumer behaviour, AI concepts, and intelligent agents. The paper discusses opportunities for future research revealed directly by analysing the results of this study. It also discusses future research directions based on current debates on AI research in the IS community. Thus, we hope that this paper will help inform future research on AI in e-commerce.