اولویت بندی و مدلسازی عوامل مؤثر در نقشه راه بکارگیری یادگیری ماشین در صنعت خرده فروشی با رویکردFCM (نقشه نگاشت فازی)

The retail industry has undergone a fundamental transformation in recent years, with artificial intelligence—and machine learning in particular—emerging as the primary driver of this metamorphosis. Despite the widespread enthusiasm for adopting these technologies, a significant portion of projects fail during the operationalization phase, mainly due to a one-dimensional and purely technical perspective on the phenomenon of intelligentization. The literature on technology management emphasizes the necessity of developing a "technology roadmap"; however, in the retail sector, there remains a substantial gap in providing a comprehensive framework that elucidates the interconnections between technological layers, processes, and strategic objectives. Previous studies have largely been confined to the static identification of components, neglecting the modeling of dynamic and nonlinear relationships among them. Consequently, this research has been designed with the aim of prioritizing and modeling the factors influencing the roadmap for implementing machine learning in the retail industry, employing a systemic approach.



This study is applied in terms of its objective and exploratory in terms of data collection, conducted with a mixed-method (qualitative-quantitative) approach. The statistical population comprised academic and industrial experts in the fields of retail and machine learning. Through purposive sampling based on the principle of theoretical saturation, 15 academic specialists and 15 industry managers and experts were selected. Initially, using the meta-synthesis method, 11 key factors were identified across three layers: infrastructure (sensors, networks, platforms), process (collection, analysis, decision-making, interaction, feedback), and strategic objectives (behavior prediction, personalization, profitability). Subsequently, employing the Best-Worst Method (BWM), criteria weighting was performed, followed by the use of Fuzzy Cognitive Mapping (FCM) to model causal relationships among the factors, calculate centrality indices, and simulate various scenarios. Data analysis was carried out using Excel and Python software.

The findings from the FCM centrality analysis revealed that the "analytical process," with a centrality index of 3.189, ranked first and was the most central factor in the system. The "data collection process," with a centrality of 2.542, secured the second position, while the "decision-making process," with a centrality of 2.361, was ranked third. Furthermore, the BWM results at the objective layer assigned the highest priority to "analyzing and predicting purchasing behavior." At the process layer, the "interaction process" was given priority, and at the infrastructure layer, the highest weight was allocated to "sensors and data-oriented hardware." Scenario simulations demonstrated that the causal network exhibits strong convergence and a fixed-point attractor, with reinforcing (positive) relationships dominating the model. This research, by presenting the first integrated three-layer model extending from physical sensors to commercial profitability, fills a fundamental gap in the literature and recommends that retail managers concentrate their initial investments on equipping stores with intelligent sensor networks and developing interactive platforms to achieve the highest return on investment, coupled with sustained improvements in customer experience and profitability. The intelligentization of retail is a structured journey that, without a layered roadmap, often comes to a halt midway.

A comprehensive examination of the research findings underscores the critical importance of a holistic and systemic perspective when adopting machine learning in retail. The dominance of the analytical process as the most central factor, according to the FCM analysis, suggests that the ability to effectively process, interpret, and derive actionable insights from data is the linchpin of the entire intelligentization ecosystem. This centrality implies that investments in analytical capabilities, including advanced algorithms, skilled data scientists, and robust data processing frameworks, are likely to yield the most significant systemic impact. However, its influence is not isolated; it is deeply intertwined with the data collection and decision-making processes, which were also identified as highly central. This indicates a tight, interdependent triad where the effectiveness of each element is contingent upon the others. Without a reliable data collection process, the analytical engine would be starved of quality input, and without an efficient decision-making process, the insights generated would fail to translate into tangible business actions.

The BWM findings provide a more granular view of priorities within each layer. At the strategic level, the overwhelming priority given to predicting purchasing behavior reaffirms that in the retail sector, the ultimate value of machine learning lies in its ability to decode customer intentions. This predictive capability is the bedrock upon which other strategic objectives, such as personalization and profitability, are built. By understanding what a customer is likely to buy, retailers can personalize offers, optimize inventory, and ultimately drive revenue. The priority assigned to the "interaction process" at the process layer is particularly insightful. It highlights that while backend analytics are crucial, the point of customer engagement—where data is generated and insights are applied in real-time—is a high-leverage area. Platforms and interfaces that facilitate seamless, intelligent interaction are, therefore, not just touchpoints but active contributors to the data ecosystem and the execution of strategic goals. Furthermore, the infrastructure layer's priority on sensors and hardware suggests that the physical embodiment of data capture is the foundational non-negotiable asset. Without the ability to collect granular, real-time data from the physical retail environment, from shelf-level stock monitoring to customer footfall and dwell time, the entire intelligent system lacks the essential raw material for learning and optimization.

The simulation results revealing a strong convergence and a fixed-point attractor in the causal network are of significant practical importance. This indicates that the system, as modeled, is inherently stable and tends towards a predictable equilibrium state when subjected to shocks. More importantly, the dominance of reinforcing relationships suggests that the system's dynamics are characterized by virtuous cycles. For instance, an improvement in the analytical process leads to better decisions, which enhance interaction and customer experience, thereby generating more and richer data for the collection process, which in turn feeds back into the analytical process. This positive feedback loop implies that strategic interventions, if well-placed, can be amplified through the system, leading to exponential gains. Conversely, this reinforcing nature also means that a negative shock to a central factor could be similarly amplified, potentially leading to a downward spiral