Quick commerce (Q-commerce) relies on fast order fulfilment, high-density last-mile coordination, dynamic routing, and the real-time allocation of couriers, stock, and orders. These characteristics make Q-commerce an appropriate setting for exploring agentic artificial intelligence (AI) and other autonomous coordination technologies, such as multiagent systems, reinforcement learning, and decision support systems. This conceptual and bibliometric review examines how multi-agent systems, reinforcement learning, autonomous decision-making, simulation, auction/trading mechanisms, and bounded large language model (LLM)-assisted decision support have been used to coordinate Q-commerce-related rapid fulfilment and last-mile delivery. A Scopus search of titles, abstracts, and author keywords initially identified 29 records. After title, abstract, and full-text screening, 16 eligible studies were retained and analysed through keyword co-occurrence analysis, overlay visualization, and thematic synthesis. The findings reveal that existing research is predominantly oriented toward technical and operational topics, including multi-agent reinforcement learning, delivery optimization, courier coordination, delivery-zone partitioning, platform-based allocation, and sustainable last-mile delivery. However, sociotechnical concerns remain underexplored, including human autonomy, accountability, transparency, privacy, fairness, safety, and platform governance. The review shows that the evidence base is dominated by multi-agent and autonomous logistics coordination rather than by mature deployments of agentic AI in Q-commerce. It therefore positions agentic AI as an emerging conceptual extension of autonomous logistics coordination that requires further sociotechnical, governance, and empirical validation
Amidst the rapid development of the metaverse and generative artificial intelligence, virtual influencers (VIs) are emerging as a significant force in digital marketing. Drawing upon meaning transfer theory and the uncanny valley theory, this research employs four experiments to investigate the mechanisms and boundary conditions through which a VI’s facial realism affects brand attitude. Study 1 (N = 298) demonstrates that high (vs. low) VI facial realism significantly enhances brand attitude. Study 2 (N = 289) reveals a dual-pathway mechanism: while perceived VI coolness supports a serial meaning transfer process to brand coolness, perceived VI eeriness acts primarily as a direct affective barrier rather than operating through sequential cognitive reassessment (i.e., brand authenticity). Furthermore, Studies 3 (N = 253) and 4 (N = 247) examine two perceptual boundary conditions. Specifically, high facial attractiveness attenuates the positive effect of facial realism on brand attitude, serving as an aesthetic buffer. Conversely, high facial familiarity unexpectedly amplifies this effect, leading to heightened consumer scrutiny. This research not only enriches the literature on virtual endorsers but also provides actionable managerial guidance for VI design and digital marketing practices.
Broadcasters play a central role in the growth of live commerce platforms. However, broadcasters’ charismatic leadership has received limited attention in existing research. Drawing on charismatic leadership theory, this study identifies three dimensions of broadcasters’ charismatic leadership: vision articulation, vision implementation, and charismatic communication style. Building on the consumer inspiration literature, this study investigates how broadcasters’ charismatic leadership influences consumer inspiration and ultimately online impulsive buying. Data were collected from 598 consumers with live commerce shopping experience through an online survey using validated measurement scales. The proposed research model was examined using partial least squares structural equation modeling (PLS-SEM). The results indicate that all three dimensions are positively associated with the inspired-by state. Furthermore, inspired-by is positively related to inspired-to, which is subsequently associated with online impulsive buying. The findings also support the serial mediating roles of inspired-by and inspired-to in linking broadcasters’ charismatic leadership to online impulsive buying. In addition, impulsive buying tendency moderates the effects of broadcasters’ charismatic leadership. This study contributes to the live commerce literature by examining the role of broadcasters’ charismatic leadership in online impulsive buying. The findings offer practical guidance for live streaming platforms on how to utilize broadcasters’ charismatic leadership to enhance sales performance.
E-marketplaces increasingly operate with rich product-level commercial criteria, such as price, rating, review density, and category structure, while their collaborative interaction signals remain sparse, restricted by privacy regulation, or intentionally excluded from the deployable architecture. Regulatory pressure under the European Union Digital Services Act and the forthcoming AI Act, together with platform-governance limits, is turning this collaborative-filtering-free (CF-free) regime into a deployment constraint that operators inherit rather than choose, while requiring that ranking decisions remain auditable. This paper proposes Hybrid RL-TOPSIS, a CF-free decision-support framework coupling an entropy-weighted Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) ranker with a tabular Q-learning behavioral module through a transparent linear fusion rule, so that every recommendation decomposes into a structural and a behavioral contribution. Across 50 deterministic 400-item Amazon India bootstrap catalogs, Hybrid RL-TOPSIS reaches F1@7 of 0.901 versus 0.574 for RL-only and 0.259 for TOPSIS-only, with paired significance tests confirming both gaps (Delta F1 = +0.326, p = 5.24 x 10^-37; Delta F1 = +0.641, p = 6.38 x 10^-51). The advantage persists under reduced reward shaping and gradual concept drift. Boundary checks on the McAuley Home and Kitchen 5-core dataset, sparse collaborative benchmarks, CUDA-based deep recommenders, LinUCB, and varying catalog sizes identify when interaction-rich architectures remain preferable, defining the operational niche rather than claiming universal superiority. A SHAP audit of hybrid scores and intrinsic RL-versus-TOPSIS contribution tables make the framework auditable at criterion level for managerial review and regulatory disclosure under tight compute budgets (16 KB Q-table, no GPU, sub-millisecond inference).
Direct price competition between online platforms and third-party sellers has intensified as platforms introduce similar first-party products. While prior research focuses on price-cutting dynamics, we examine a less explored scenario: how sellers respond when platform owners raise their prices. Using large-scale transaction data from a leading pharmaceutical e-commerce platform in China, we analyze third-party sellers' pricing responses to platform-initiated price increases. We argue that prior interactions create rigid role frames through which sellers interpret the platform owner, thereby shaping their pricing responses. The empirical results show that sellers with greater reliance on platform logistics are more likely to follow price increases, whereas sellers with higher product overlap are less likely to align upward. Seller size further moderates these effects, with larger sellers exhibiting weaker pricing responses. This study contributes to the literature on platform governance and strategic pricing while offering practical implications for pricing management in hybrid platform ecosystems.
Live commerce continuously acquires live viewers and effectively converts them into purchases. Live streamers effectively facilitate this process. This study innovates by employing differential equations to model the dynamic fluctuations of live streaming viewership, and analyze how the dynamics of live streaming viewership influence the efforts of the brand owner and the performance of the live streamer, and propose a “reciprocal subsidy” contract to achieve system coordination in live commerce. The results show that as live streamers’ core competencies (core fan cohort and viewer-to-buyer conversion rate) are strengthened, the operational efficiency of live commerce improves. Centralized decision-making significantly outperforms decentralized decision-making in optimizing system performance. While the “reciprocal subsidy” contract enables coordination, its viability depends on constraining the commission ratio within a specific range acceptable to both the live streamer and the brand owner. Furthermore, the applicability of the contract diminishes as the core fan cohort expands. This study provides actionable insights for managing live commerce operations under dynamic live streaming viewership.
The rapid growth of e-commerce has been accompanied by increasingly sophisticated fraud, costing online platforms over $48 billion annually. Fraudulent entities-fake review accounts, payment fraud rings, and money laundering operations-often exploit the gap between carefully constructed profile attributes and their actual transaction behavior. Detecting such fraud in large-scale online transaction networks is challenging due to the scarcity of labeled fraud cases and the complex, multi-faceted nature of fraudulent behavior. We propose CSGAD, a semi-supervised graph-based fraud detection framework that models the inconsistency between two complementary views of entity behavior in e-commerce networks: (1) the structural view, encoding transaction connectivity patterns such as who transacts with whom and through which intermediaries, and (2) the semantic view, encoding account profile attributes and transaction characteristics. Our key insight is that legitimate e-commerce participants exhibit consistency between their network behavior and profile attributes, whereas fraudulent entities show characteristic cross-view divergence. CSGAD employs dual specialized encoders-a transaction pattern encoder based on Graph Convolutional Networks and an account profile encoder with controlled Laplacian smoothing-linked through an InfoNCE contrastive objective that maximizes cross-view mutual information. Extensive experiments on eight real-world datasets, including Amazon product review networks (13K+ entities; collected 2014-2015), the T-Finance online payment platform (39K accounts; collected 2019-2020), the Elliptic Bitcoin transaction network (203K transactions; January-September 2019), and the DGraph digital commerce network (3.7M users; collected 2020-2021), demonstrate state-of-the-art fraud detection. Notably, CSGAD achieves 3.9% and 3.6% AUROC improvements on T-Finance and Elliptic over the strongest baseline (SpaceGNN), while requiring only 1%-5% labeled fraud cases. A detailed case study on Amazon review networks confirms that CSGAD identifies coordinated fake review clusters by detecting profile-behavior inconsistencies that conventional methods miss. We further discuss deployment considerations, scalability, and practical implications for e-commerce platform operators.
Understanding how customer satisfaction evolves over time has become increasingly critical in today’s rapidly changing service environments. While the Kano model has long served as a foundational framework for categorizing quality attributes, traditional approaches often fail to capture the dynamic and nuanced shifts within these attributes. This study addresses these limitations by introducing a revised dynamic Kano model. Using advanced topic modeling, temporal analysis, and sentiment evaluation, this study develops a refined framework that systematically tracks and analyzes quality transitions over time. By leveraging online user reviews, the study suggests a new way of categorizing customer requirements that allows for a dynamic reinterpretation of the traditional Kano model, offering fresh insights into how customer needs evolve. As a result, eight types of attributes are defined: star performer, critical concern, understated essential, silent necessity, fan favorite, disappointing, quiet charm, and low impact. To demonstrate the applicability of this study, we selected a ChatGPT—a generative AI service—as our case study. This approach not only advances theoretical understanding of dynamic quality characteristics but also offers a practical framework to facilitate more proactive responses to evolving customer needs.
Artificial intelligence (AI) is reshaping e-business by enabling advanced personalization in immersive digital commerce. Accurately predicting consumer preferences in such environments remains challenging due to high inter-subject variability in neurophysiological responses. To address this challenge, this study proposes an AIenabled multimodal neuromarketing framework that integrates electroencephalogram (EEG) and eye-tracking data to predict consumer preferences in virtual reality (VR) shopping environments. The framework employs a hybrid deep learning architecture combining convolutional neural networks (CNNs), bidirectional gated recurrent units (BGRUs), attention mechanisms, and adversarial domain adaptation to learn subject-invariant representations from multimodal data streams. In a VR shopping case study with 30 participants, the proposed multimodal model achieved up to 80.89 % accuracy and outperformed single-modality approaches under leave-one-subject-out (LOSO) evaluation. The findings highlight the complementary value of EEG and eyetracking data and demonstrate the framework's potential for enhancing customer experience personalization and strategic AI adoption in immersive e-business. This work addresses real-world challenges and strategic considerations of AI integration in e-business, promoting responsible and data-driven innovation in digital commerce.
On social media platforms, User-provided Monetary Incentives (UMI) have emerged as a relatively novel form of monetary incentive in recent years, yet the research on UMI remains limited. This study investigates how UMI influences content contribution behavior: the effects on contribution quantity and quality, the temporal dynamics, and the heterogeneous effects across different contributor groups. We collect a large-scale dataset with approximately 15 million posts and 390 thousand contributors from a leading online finance forum (Xueqiu.com) in China, and then conduct a quasi-experimental design using propensity score matching (PSM) followed by a difference-in-differences (DID) estimator to identify the causal effect of UMI on content contributions. The results show that UMI significantly increases both contribution quantity and quality. Interestingly, the effect is neither permanent nor fragile: although the effect decays over time, it persists with both statistical significance and considerable magnitude through even 16 weeks. Moreover, professional identity and social connectedness moderate UMI's influence: when compared to non-professional contributors, the professional contributors exhibit only stronger short-term quality improvements, while the contributors with larger follower networks show both stronger short-term quantity improvements and enduring quality improvements. This study is the first attempt to offer a systematic and comprehensive investigation of UMI's effects on content contributions using a quasi-experiment based on a large-scale dataset. This study also introduces a dynamic view of incentive effectiveness, revealing temporal decay in UMI's impact, and highlights how professional identity and social connectedness shape heterogeneous responses.
This study advances omnichannel retail research by shifting focus to the post-adoption stage, examining how digital-native consumers evaluate and sustain engagement in fashion omnichannel environments. Extending UTAUT, the model integrates key behavioural drivers (performance expectancy, effort expectancy, and trust) with contextual inhibitors (privacy concerns, uncertainty, and situational factors) relevant to digital commerce. A sequential mixed-methods design was adopted. Insights from 24 qualitative interviews informed a survey of 589 omnichannel shoppers, which was analyzed using PLS-SEM. Performance expectancy, effort expectancy, and trust all significantly enhance perceived value, which in turn strongly predicts satisfaction. Privacy concerns and uncertainty diminish perceived value, while situational factors show no significant effect. By contextualising UTAUT for an emerging market and combining consumer narratives with empirical validation, the study offers actionable implications for strengthening perceived value and satisfaction among young omnichannel consumers.
This study investigates how users' desire for personalised online experiences and their perceptions of a website's ethical data practices shape trust and cookie acceptance decisions across different website types. Using an online survey administered across five major Chinese web portals, the research captures users' situational preferences regarding four categories of cookies. It examines how technological infrastructures, commercial incentives, and legal compliance jointly influence cookie-based data strategies. The results indicate that transparent user controls and reward-based incentives significantly enhance users' willingness to share data. At the same time, ethical standards and trust cues play a critical role in sustaining long-term data authorisation. These findings demonstrate that businesses can strengthen user trust by balancing personalisation benefits with responsible data governance. Overall, the study advances understanding of cookie acceptance by integrating technological, commercial, and regulatory perspectives and offers novel insights into sustainable data donation mechanisms in digital marketing.
Traditional key opinion leader (KOL) live streaming mode poses challenges for brand manufacturers due to high commissions and pricing control issues. Manufacturers start to invest in artificial intelligence (AI) live streaming to gain control over live streaming. This paper constructs a dual-channel supply chain consisting of a manufacturer focused on live streaming and a retailer, where the manufacturer sells products through both direct channel and the retailer's retail channel. Using game theory, we study the manufacturer's choice between AI and KOL live streaming modes. We also investigate the impact of the consumer perceived value enhancement brought by live streaming on channel decisions and AI investment. Results show that only when market size is large, the manufacturer chooses AI live streaming. Importantly, when adopting AI live streaming, the manufacturer is advised to consider AI investment based on market size, rather than solely pursuing higher AI levels. Additionally, the enhancement of consumer perceived value is expected to be as high as possible under AI live streaming. However, under KOL live streaming, if the KOL commission rate is high, an increase in perceived value enhancement can significantly decrease the manufacturer's profit. Finally, when the market size is moderate, AI live streaming can achieve a win-win situation for the manufacturer, the retailer, and consumers.
Customer expectations for quick delivery have shifted significantly. Today, speed is the new currency, with customers wanting faster delivery while prioritising cheaper delivery. Hence, some changes have been made to address this faster delivery process in the retail industry. This research will address what factors need to be considered in adopting the quick-commerce service in retail and address the interdependence among factors and subfactors using supply chain agility. To address these problem the information was gathered from experts in the retailing area from academics and industry and proposed a hybrid multiple criteria decision-making model incorporating the Decision-Making Trial and Evaluation Laboratory (DEMATEL)-based Analytic Network Process (DANP). The findings result in 19 subfactors categorised into five factors from the literature review, which might be important in adopting a q-commerce service. The cause-and-effect relationship has been drawn among the factors and subfactors. The top subfactor, based on the influencing weights, is classified into three tiers, in which the first tier factors are core critical factors, the second tier is for enhancing the efficiency of the organisation, and the third tier is for enhancing customer satisfaction, engagement and trust for q-commerce (quick commerce) service. To successfully adopt q-commerce in retail, artificial intelligence and machine learning facilitate advanced demand forecasting, route optimization, and inventory management, thereby enhancing operational responsiveness in highly dynamic urban environments. Additionally, real-time order tracking, intelligent warehousing, and temperature-controlled transportation are made possible by Internet of Things technology, guaranteeing product quality and supply chain transparency. This knowledge helps to clarify how managers and merchants can use these elements to improve the q-commerce service. This study is novel in that it highlights the factors that contribute to the retail adoption of q-commerce and explores the interdependencies between these aspects.
Recent advancements in aspect-based sentiment analysis (ABSA) have increasingly shifted focus towards dialogue-based scenarios, termed conversational aspect-based sentiment analysis (CABSA), a critical area for advancing ABSA applications in business contexts. Despite this progress, current methodologies inadequately exploit the inherent reply structures within dialogues, and existing tagging schemes exhibit notable redundancy. To address these limitations, we introduce a novel dialogue graph attention network (DiaGAT) that effectively harnesses both reply relations and syntactic information. Our approach involves transforming dialogue sequences into a graph structure based on reply and syntactic relations. We then implement a dual hierarchical graph attention network (GAT) architecture, comprising an utterance GAT and a token GAT, to facilitate multi-level information propagation. The utterance GAT module enables efficient message passing within the utterance subgraph, enhancing the model’s ability to comprehend dialogue context and extract semantic information. Comprehensive experimental evaluations demonstrate that DiaGAT establishes a new state of the art on the DiaASQ benchmark, surpassing even large language models (LLM) in effectiveness and precision. This article effectively captures and interprets consumer sentiment, offering valuable insights into customer attitudes and emotional trends. These insights can serve as a practical decision-making tool for business customer service teams, helping them tailor responses and improve overall customer satisfaction.
