Can Artificial Intelligence Enhance the Stability of Supply Chain Systems for Sports Enterprises? Insights from Systems Theory and Supply Chain Management Theory

Abstract

In the digital economy, the effective use of artificial intelligence (AI) is crucial for maintaining supply chain stability (SCS) in sports enterprises (SEs). Leveraging systems theory and supply chain management theory, we construct a dual machine learning model (DML) to empirically assess the impact of AI on the SCS of SE. This analysis is based on panel data from 45 Chinese listed SEs over the period 2012–2023. The results indicate that AI significantly enhances supplier stability but notably reduces customer stability in SE. Talent attraction emerges as the primary mechanism, while logistics efficiency fails to fulfill its anticipated role. The impact of AI on SCS in SE exhibits heterogeneity based on enterprise type and profitability status. Our findings offer valuable insights for harnessing the potential of AI and fostering its deeper integration into the supply chains of SE.

1. Introduction

Sport is an efficient and flexible tool for sustainable development, playing a crucial role in achieving United Nations SDGs 3, 4, 5, 11, 16, and 17 [1]. The sports industry’s diversified revenue structure has enhanced its resilience to risks. In addition to traditional media rights and event ticket sales, revenue streams from sports team/franchise valuations, commercial and sponsorship rights, and betting rights are expected to grow at an annualized rate of 6.7% to 7.0% over the next three to five years. Despite the positive industry outlook, SEs continue to face multiple challenges, including economic volatility, geopolitical conflicts, and shifting consumer behavior [2] (see Figure 1). The WFSGI × McKinsey 2025 Global Sporting Goods Industry Report reveals that only 30% of companies have achieved both revenue and margin growth since 2018, while 84% of executives are concerned about the impact of supply chain disruptions. Maintaining SCS has thus become the primary challenge for SE in realizing a promising future.

SCS in SE is defined as the ability to maintain a balance between supply and demand, ensure a stable market for the production, exchange, distribution, and circulation of products or services, and sustain a network of functioning upstream and downstream companies [3]. Established studies identify supply chain structure [4], supply chain efficiency [5], supply chain security, and supply chain resilience [6] as key factors influencing SCS. Additionally, several studies have examined the impact of factors such as macroeconomic performance, climate risk [7], geopolitics [8], and market competition [9] on the SCS of SE. However, with the advent of the big data era and the accelerated digital transformation of SE, challenges such as highly fragmented supply chain information flows—stemming from the dispersion of upstream and downstream information—and the growing disparity between high demand volatility and slow response to sudden market changes have significantly hindered SEs’ ability to maintain SCS [10,11]. In this context, AI technology, leveraging big data, excels at identifying patterns in complex data and making predictive optimizations. By integrating dispersed information through intelligent algorithms, it enhances data visibility and transparency, enabling real-time monitoring and analysis while ensuring the smooth transmission of upstream and downstream information flows [12]. This has demonstrated significant potential in enhancing the SCS of SE.

There is a limited number of academic studies on AI and SCS in SE. Therefore, combining supply chain management theory and systems theory, we select China—one of the largest countries globally in terms of sports participation (over 500 million people) and the sports consumption market (with the national sports industry projected to exceed 5 trillion yuan by 2025). We collect and screen panel data from 45 listed sports companies, sourced from public databases, to serve as the research sample. An empirical study using DML based on neural network algorithms is then conducted to address the following questions:

(1)

How to develop a theoretical framework for analyzing SCS in AI-powered SE.

(2)

How to accurately assess the impacts and mechanisms of AI on SCS in SE.

(3)

How to optimize the path for enhancing SCS in SE.

This study makes the following contributions: First, it strengthens the theoretical framework of SCS in the context of intelligent transformation by establishing a clear causal relationship between AI development and SCS in SE. Second, it offers new insights into the relationship between AI and SCS in SE by identifying logistics efficiency and employee quality as key mechanisms driving AI’s impact on SCS, and by uncovering the heterogeneity based on firm type and profitability status. Finally, we apply a DML that effectively controls for multidimensional variables and excludes confounding factors, offering reliable insights into the impact of AI development on SCS in SE.

Our study is structured as follows: Section 2 presents the literature review and research hypotheses, Section 3 outlines the research design, Section 4 provides the empirical analysis, Section 5 discusses the implications of our findings, Section 6 concludes with findings and policy recommendations, and Section 7 discusses the study’s limitations.

2. Literature Review and Research Hypothesis

2.1. Literature Review

Few studies have explored the impact of AI applications on the stability of corporate supply chains, particularly in SE. Existing literature primarily focuses on sectors such as energy [13], electric power [14], agribusiness [15], manufacturing [16], healthcare, and retail [17]. These studies predominantly address enterprises in developed countries, including the United States and France, as well as emerging markets such as China and India. Together, these studies highlight the critical role of AI in influencing supply chain disruption risk, security, quality, and resilience. Additionally, while some studies have explored the impact of AI applications on supply chain structure and supply chain resilience in SE [6,18], they have yet to examine how AI influences the stability of their supply chains.

Research has widely acknowledged AI as a means to add value to supply chain processes [19], including supply chain network design, supplier selection, inventory planning, demand planning, and green supply chain management. For instance, Radanliev et al. [20] demonstrated that AI applications provide real-time intelligent support for predictive cyber risk analysis, helping companies reduce uncertainty by generating insights to enhance supply chain visibility. Similarly, Chai et al. [21] found that AI assists companies in designing KPIs and adjusting search parameters or fine-tuning systems to evaluate supplier suitability. Zdravkovic et al. [22] found that AI enhances decision-making and automates inventory and logistics, as well as improves production planning and scheduling through the application of machine learning models or logic-based systems. Krishnamurthy et al. [23] analyzed regional energy demand in South Africa and demonstrated that AI-based modeling of generation potential and load demand forecasting enables intelligent management and planning of the energy supply chain.

Based on the existing literature, research on how the adoption of AI affects the supply chains of SE remains limited, and the evidence to date is inconclusive [24]. Wang et al. [25] employed a structural equation modeling approach and, using survey data from 570 Chinese SEs, found that the adoption of AI and digital technologies helps enhance customer responsiveness and value creation, thereby accelerating the integration of internal and external supply chains. However, when confronted with inventory delays and lost sales arising from seasonal demand fluctuations, the digital supply chain technologies currently used by sports enterprises are often insufficient to respond effectively. Although Qiu et al. [26] showed that emerging digital technologies—such as probabilistic linguistic term sets (PLTS)—can facilitate data-driven supplier selection and risk-mitigation decisions, thereby helping sustain SCS, SEs operate within highly dynamic and interconnected supply networks where both customer- and supplier-side risks remain persistently uncertain [27,28]. Accordingly, it is imperative to clarify, at the theoretical level, how AI influences SCS in SE.

Amid increasing challenges to the stability and security of global supply chains, issues such as weak foundations, inefficient factor flows, and insufficient innovation have intensified the vulnerability of supply chains in the sports sector. In practice, AI offers significant support for SE in maintaining SCS. For instance, Adidas has integrated AI and IoT technologies into its supply chain to achieve intelligent management and automated control. Similarly, companies like Anta and Li-Ning leverage intelligent big data to precisely identify consumer preferences and provide customized services to enhance customer retention. However, despite extensive research on the impact of AI on supply chains, the existing literature has largely focused on non-SE and has paid limited attention to how AI affects SCS in SE. Moreover, this stream of research has yet to develop a coherent analytical framework or reach a consensus on the empirical conclusions (See

Table 1. Summary of research on AI and supply chains.

Research Samples	Research Subjects	Empowerment Target	Potential Risks
Energy companies	Supply chain disruption risk	Supply chain network design	Inventory delays
Electric power companies	Supply chain security	Supplier selection	Sales losses
Agricultural companies	Supply chain quality	Inventory planning	Customer risk
Manufacturing companies	Supply chain resilience	Demand planning	Supplier risk
Healthcare companies		Green supply chain management
Retail companies

). Here, we address the following two core research questions:

(1)

Can AI enhance SCS in SE?

(2)

What are the underlying mechanisms?

We focus on Chinese SE and, drawing on systems theory and supply chain management theory, develop an integrated theoretical framework to elucidate how AI affects SCS in SE. We further employ a DML to empirically test our hypotheses. Our findings extend the literature linking AI adoption to SCS in SE and are critical for deepening practitioners’ understanding of how SE can strategically leverage AI to sustain stable supply chain foundations.

2.2. Research Hypothesis

Supply chain management theory posits that a sports business’s supply chain comprises multiple upstream and downstream entities. Effective management requires considering the flow of goods, services, information, and capital along the chain, along with the associated risks, challenges, and opportunities. Supply chains function most efficiently when they are structured as an organized network of interconnected entities [29]. On the one hand, AI enables SE to predict demand fluctuations using historical data and market information. By integrating and analyzing data from multiple sources, enterprises can assess supplier performance, optimize purchasing decisions, and reduce reliance on a single supplier, thereby enhancing the stability of upstream suppliers [30]. On the other hand, AI enables SE to analyze multidimensional data, including consumer behavior, historical sales, and market trends, to provide accurate sales forecasts [31]. This capability helps downstream SEs reduce inventory backlogs, avoid stockouts, and swiftly adjust sales strategies in response to demand changes, thereby maintaining downstream customer stability.

However, for SE, the application of AI may introduce cybersecurity risks and algorithmic biases [32]. The processing and analysis of vast amounts of data may lead to “data breaches” and “ethical risks” [33]. Moreover, since algorithms are developed by humans, they may reflect inherent biases in procurement and sales decisions, resulting in uncertainties in supplier and customer selection. Additionally, the technical complexity of AI applications may increase labor costs and energy consumption for SE [34]. This is because SEs need to hire additional highly skilled personnel to leverage the advantages of algorithms, and the training and operation of AI consume significant amounts of electricity [35]. As a result, SEs may offset these costs by increasing the unit prices of their products, a process that could lead to the loss of suppliers and customers.

Accordingly, we propose the following hypotheses:

H1a. 

AI significantly enhances the upstream SCS of SE.

H1b. 

AI significantly enhances the downstream customer stability of SE.

H2a. 

AI significantly reduces the upstream SCS of SE.

H2b. 

AI significantly reduces the downstream customer stability of SE.

From the perspective of systems theory, the supply chain of SE is viewed as a complex dynamic system, where coordination and cooperation among internal components (e.g., purchasing, production, distribution, and sales) are essential to maintaining overall stability [36]. Logistics encompasses the flow of goods from suppliers to consumers across the entire supply chain, making logistics efficiency a direct determinant of SCS. AI enables SE to swiftly respond to fluctuations in market demand and changes in supplier performance by automatically adjusting logistics strategies, optimizing inventory allocation, and ensuring coordinated operation across the supply chain [37], thereby enhancing SCS. Additionally, the application of AI in data analytics, automation, and decision optimization enhances efficiency and fosters innovation, which in turn attracts top talent and facilitates the flow of knowledge throughout the supply chain, thereby improving the SCS [38]. Accordingly, we propose the following hypotheses:

H3a. 

AI significantly enhances the SCS of SE by improving logistics efficiency.

H3b. 

AI significantly strengthens the SCS of SE by attracting high-quality talent.

Theoretical mechanisms are shown in Figure 2.

3. Methodology

3.1. Variable Selection

Explained variable: SCS (stability). We categorize SCS into two components: supplier stability and customer stability. Supplier stability is measured by the proportion of the top five suppliers that remain among the top five from the previous year, while customer stability is measured by the proportion of the top five customers that are also among the top five from the previous year [3]. The reason is twofold: On the one hand, listed companies disclose the aforementioned indicators in their annual reports, providing a solid data foundation for our empirical research. On the other hand, the increase in the number of repeat suppliers signifies enhanced consistency in supplier relationships. A higher proportion of repeat customers in the annual report indicates greater stability in customer relationships [39,40]. Jin’s [41] study suggests that the aforementioned indicators can also serve as proxies for upstream and downstream supply chain volatility. It is important to note that these two indicators, to some extent, fail to capture the overall characteristics of the supply chain status in SE, thus exhibiting certain limitations. However, supported by a substantial body of existing research [39,40,41,42], and considering the data completeness required for empirical analysis, we have opted to use the aforementioned indicators as proxies for the stability of SE supply chains.

Explanatory variable: AI (IT). Building on the work of Ren et al. [43], we employ a DML to generate an AI lexicon, and then construct corporate AI indicators based on annual reports and patent texts from listed companies (see Figure 3).

Control variables: The supply chain encompasses R&D (Research and Development), production, logistics, information transmission, and end-sales [44]. To more accurately identify the impact of AI, we select 14 control variables based on these five links and the fundamental characteristics of the SE. See

Table 2. Descriptive statistics.

Dimension	Variable	Unit	Obs	Mean	Std. Dev.	Min	Max
R&D	Amount of R&D investment	ln	585	17.627	1.624	10.176	21.456
	R&D expense ratio	%	585	0.031	0.134	−0.244	1.537
	Number of R&D staff as a percentage	%	585	11.897	14.754	−67.560	111.580
Production	Depreciation of fixed assets	ln	585	17.341	1.403	12.477	21.766
	Net cash flows from operating activities	ln	585	19.092	1.572	13.178	23.347
	Long-term asset suitability ratio	%	585	9.134	31.621	−7.189	423.266
	Shareholders’ equity to fixed assets ratio	%	585	40.929	191.431	−532.797	2120.139
Information transmission	Net intangible assets	ln	585	18.149	2.341	7.134	23.408
	Intangible assets ratio	%	585	0.051	0.071	0.000	0.424
Sells	Cash received from sales of goods and services	ln	585	21.375	1.462	11.294	24.800
	Growth rate of selling expenses	%	585	1.582	31.086	−8.584	748.310
Enterprise Characteristics	Company size	ln	585	22.095	1.202	16.520	26.039
	Remuneration of the first member of senior management	ten thousand RMB	585	34.986	16.061	7.800	119.000
	Operating profit per share	RMB	585	0.280	0.727	−8.167	3.642

.

3.2. Model Setting

SCS is disrupted by multiple factors, and traditional linear models struggle to address the challenges posed by high-dimensional data, nonlinear relationships, and complex causal chains. DML, which leverages machine learning models to net out the influence of control variables on key explanatory variables, generate residuals for causal isolation, and mitigate overfitting via cross-fitting [45], is well-suited for identifying the impact of AI on SCS in SE. We set up the following DML:

$${stability}_{i,t}={\alpha }_1{IT}_{i,t}+g\left(X_{i,t}\right)+U_{i,t}$$

(1)

$$\mathrm{E}\left(U_{i,t}|{IT}_{i,t},X_{i,t}\right)=0$$

(2)

The subscripts i and t denote individual enterprises and years, respectively, while stability refers to the SCS of SE. IT denotes the level of AI. X comprises city fixed effects based on enterprise location, year fixed effects, and enterprise fixed effects, whose impact on AI is estimated using a machine learning model g. We employ neural networks as the specification for g. This is because neural network algorithms are well-suited for complex pattern recognition and prediction tasks, and effectively handle dynamic, changing environments, which aligns with the focus of our research [46]. U denotes the random error term.

Next, we estimate $\widehat{g}\left(X\right)=E\left[stability\right|X]$ using Equation (1) to further estimate the treatment effect $\widehat{{\beta }_0}$, and the treatment process is described in Equation (3):

$$IT-\widehat{g}\left(X\right)={\beta }_{0}stability+U,E\left[U|X,IT\right]=0$$

(3)

To obtain unbiased estimates of the disposition coefficients, we further remove the effect of X on the stability to eliminate the estimation bias of $\widehat{g}\left(X\right)$. We orthogonalize the stability to obtain the orthogonal regression quantity $\widehat{V}=stability-\widehat{h}(X)$. Combining Equation (3), we obtain Equation (4) for $\widehat{{\beta }_0}.$

$${\widehat{{\beta }_0}=({\displaystyle \frac{1}{n}}\sum _i{{\widehat{V}}_{i}stability}_i)}^{-1}{\displaystyle \frac{1}{n}}\sum _i{{\widehat{V}}_i(IT}_i-\widehat{h}(X_i))$$

(4)

Equation (4) can also be approximated as Equation (5):

$$\sqrt{n}\left(\widehat{{\beta }_0}-{\beta }_0\right)=[\mathrm{E}\left(V^2\right){]}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}\sum _i{V}_i{U}_i+[\mathrm{E}(V^2{)}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}\sum _i[\widehat{g}\left(X_i\right)-g\left(X_i\right)]]\left[\widehat{h}\left(X_i\right)-\mathrm{h}\left(X_i\right)\right]$$

(5)

$$[E\left(V^2\right){]}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}{\sum }_i{V}_i{U}_i~N(0,\sum )$$

(6)

In the case of satisfying Equation (6), $[E(V^2{)}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}{\sum }_i[\widehat{g}\left(X_i\right)-g\left(X_i\right)]]\left[\widehat{h}\left(X_i\right)-h\left(X_i\right)\right]$ has an upper limit of $\sqrt{n}{n}^{-({\gamma }_1+{\gamma }_2)}$, ${\gamma }_1$ and ${\gamma }_2$ denote the rate of convergence of $\widehat{g}\left(X_i\right)$ to $g\left(X_i\right)$ and $\widehat{h}\left(X_i\right)$ to $h\left(X_i\right)$, respectively, and can converge to zero. With the help of Equation (5), we are able to obtain unbiased estimates of the disposition coefficients (i.e., the coefficients of the AI on the SCS).

To mitigate potential overfitting of the machine learning model, we implement K-fold cross-validation to enhance the accuracy of coefficient estimates. The sample is split at a 1:4 ratio, with the random seed set to 42. The reason for selecting 5-fold cross-validation is that it helps reduce random biases caused by different data splits and allows for the evaluation of the model’s generalizability [47].

3.3. Data Sources and Processing

We select a sample of Chinese A-share listed SEs from 2011 to 2023, as 2011 marks the launch of China’s 12th Five-Year Plan for Sports Development and represents the earliest year with reliable data availability. We apply linear interpolation to address missing data and exclude enterprises with substantial data gaps, resulting in a final sample of 45 enterprises (See

Table 3. ID of SEs.

ID of SEs
000529	002105	002701	600060	600814
000558	002168	300005	600136	600826
000639	002181	300043	600158	601718
000652	002346	300133	600287	603000
000735	002395	300162	600358	603555
000796	002400	300232	600386
000811	002431	300291	600637
002035	002587	300397	600706
002081	002659	600018	600749
002098	002694	600052	600768

). The data are obtained from the Wind database, Sina Finance, and the annual reports of listed companies.

4. Results and Discussions

4.1. Baseline Results

We empirically examine the impact of AI on the SCS of SE using Equations (1) and (2), with the results reported in

Table 4. Baseline regression results.

	(2)	(3)
	Supplier	Customer
IT	0.615 ***	−0.305 ***
	(8.43)	(−3.19)
Control variables	Yes	Yes
Individual effect	Yes	Yes
Year effect	Yes	Yes
City effect	Yes	Yes
_cons	0.223 ***	−0.129 *
	(3.31)	(−1.67)
N	585	585

z statistics in parentheses. * p < 0.05, *** p < 0.001.

. The results indicate that AI significantly enhances supplier stability (coefficient = 0.615, z = 8.43) but significantly reduces customer stability (coefficient = −0.315, z = −3.19). These findings support H1a but contradict H1b. A plausible explanation is that AI facilitates more frequent product customization and inventory optimization, thereby enhancing supply chain efficiency for SE. However, this may simultaneously reduce product variety and personalization, ultimately undermining customer loyalty and satisfaction [7,48]. Meanwhile, intense competition in China’s sports industry—particularly in pricing, product customization, and service—may lead to excessive customer selectivity, increasing the likelihood of brand switching. This aligns with the findings of Walzel and Nowak [9].

4.2. Robustness Tests

We conduct four robustness tests. First, we apply a 1% winsorization to all variables and re-estimate the models to mitigate the influence of extreme outliers. The regression results are reported in Columns (1) and (2) of

Table 5. Robustness test results.

	(1)	(2)	(3)	(4)	(5)	(6)
	Supplier	Customer	Supplier	Customer	Supplier	Customer
AI	0.615 ***	−0.305 ***	0.141 *	−0.871 ***	0.091 *	−0.030 ***
	(8.43)	(−3.19)	(1.67)	(−4.23)	(1.86)	(−2.71)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Individual effect	Yes	Yes	Yes	Yes	Yes	Yes
Year effect	Yes	Yes	Yes	Yes	Yes	Yes
City effect	Yes	Yes	Yes	Yes	Yes	Yes
_cons	0.223 ***	−0.129	−0.0577	−0.0829	0.025	−0.713 ***
	(3.31)	(−1.67)	(−0.46)	(−1.05)	(1.60)	(−5.92)
N	585	585	585	585	585	585

z statistics in parentheses. * p < 0.05, *** p < 0.001.

. The coefficient estimates and significance levels of IT remain broadly consistent with the baseline regression results, indicating that they are not affected by extreme outliers. Second, during the sample period, China’s Ministry of Commerce and other agencies launched a pilot program on supply chain innovation and application in 2018. Pilot enterprises were expected to enhance supply chain management through digital technologies, thereby improving efficiency, reducing costs, and generating substantial value for both firms and customers. To account for the potential confounding effects of this policy, we include policy dummy variables and re-estimate the regressions. The corresponding results are reported in Columns (3) and (4) of Table 5. It can be found that the coefficient values for IT remain significant at least at the 10% level, indicating that our conclusions remain robust. Third, we replace the core explanatory variable with the number of AI patents held by SE and re-estimate the model. The results are presented in columns (5) and (6) of Table 5. It can be observed that the coefficient of AI remains significant at least at the 10% level, indicating that our conclusions remain robust.

Fourth, following the methodology of Chernozhukov et al. [45], we construct the following instrumental variable model to address potential endogeneity concerns:

$${stability}_{i,t}={\alpha }_1{IT}_{i,t}+g\left(X_{i,t}\right)+U_{i,t}$$

(7)

$${IV}_{i,t}=m\left(X_{i,t}\right)+V_{i,t}$$

(8)

where ${IV}_{i,t}$ serves as the instrumental variable for ${IT}_{i,t}$, proxied by the logarithm of total audit fees and the proportion of artificial intelligence investment to intangible asset investment. On the one hand, AI enhances corporate information transparency and internal control transparency, mitigating information asymmetry and improving audit quality, while also incurring significant training costs. On the other hand, the total audit fees and the proportion of AI investment to intangible asset investment meet the exogeneity condition for the selection of instrumental variables. The regression results are presented in Columns (1) to (4) of

Table 6. Endogeneity test results.

	(1)	(2)	(3)	(4)
	Supplier	Customer	Supplier	Customer
AI	2.165 ***	−1.184 ***	0.753 ***	−0.115 *
	(4.11)	(−8.98)	(40.56)	(−1.89)
Control variables	Yes	Yes	Yes	Yes
Individual effect	Yes	Yes	Yes	Yes
Year effect	Yes	Yes	Yes	Yes
City effect	Yes	Yes	Yes	Yes
_cons	0.357	−0.169 *	0.271 ***	−0.126
	(1.66)	(−1.92)	(3.77)	(−1.62)
N	585	585	585	585

z statistics in parentheses. * p < 0.05, *** p < 0.001.

. The second-stage coefficient estimates are all significant at the 1% level and consistent with the baseline regression results, confirming the robustness of our findings.

4.3. Mechanism Tests

Since our explanatory variables are not binary, the mediation mechanism cannot be tested using the medDML package in R (4.1.1 version). To test whether two mechanisms hold, we construct the following mechanism analysis model, drawing on the methodology of Chernozhukov et al. [45]:

$$M_{i,t}={\alpha }_1{IT}_{i,t}+g\left(X_{i,t}\right)+U_{i,t}$$

(9)

$M_{i,t}$ denotes the mechanism variable, where talent attraction is measured by the proportion of employees with postgraduate degrees, and logistics efficiency is proxied by inventory turnover. The mechanism test results are reported in

Table 7. Mechanism test results.

	(1)	(2)
	Talent Attraction	Logistics Efficiency
AI	15.23 ***	3022.356
	(12.06)	(0.93)
Control variables	Yes	Yes
Individual effect	Yes	Yes
Year effect	Yes	Yes
City effect	Yes	Yes
	0.474	26,831.0
	(0.01)	(0.99)
N	585	585

z statistics in parentheses. *** p < 0.001.

. The coefficients of IT in the talent attraction channel are all significantly positive at the 1% level (t = 12.06), indicating that AI effectively promotes talent attraction. However, when combined with the baseline regression results, talent attraction is found to reduce customer stability. Therefore, H2a is only partially supported. A possible explanation is that the current linkage between market demand and talent application in China’s sports industry has yet to form a virtuous cycle. Employee digital skills are considered a key pillar of AI. Given that SEs often lack staff with the capabilities to share, process, and manage data, both the organizations and their managers must invest long-term resources to realize the full value of AI applications [49]. On the one hand, talent attraction may result in technological over-concentration [50] while overlooking customer relationship management and failing to meet individualized customer needs, thereby undermining long-term customer stability. This interpretation is also supported by insights from customer relationship theory [51]. On the other hand, recruiting digital skills talent in SE increases labor costs. In the short term, to reduce inventory, transportation, and manufacturing costs, SE may implement discount campaigns to boost customer orders and balance operational costs. However, since these orders are not based on actual consumer demand, they may distort the “true” demand within the supply chain, triggering the bullwhip effect, which is detrimental to customer stability [49].

In the analysis of the logistics efficiency mechanism, the coefficient of IT is positive but statistically insignificant, suggesting that AI does not enhance logistics efficiency. Therefore, H2b is not supported. This finding contrasts with the results of Jahangir et al. [52] and Malhotra and Kharub [53]. A possible explanation is that the effectiveness of AI heavily relies on data integration and sharing. A possible explanation is that sports logistics involves multiple interconnected nodes, including logistics networks, distribution centers, and terminal outlets [54], which requires the integrated coordination of multiple supply chain stages such as procurement, production, waste management, recycling and reprocessing, and distribution. The effectiveness of AI is highly contingent on data integration and sharing. However, constrained by limited innovation capacity in the application of sports data, deficiencies in data completeness, and persistent data-sharing barriers in China’s sports market, SEs face an insufficient supply of high-quality data across their supply chains, resulting in the suboptimal realization of sports data value. When confronted with demand characteristics specific to the sports market—such as high demand volatility, short product life cycles, and strong requirements for product customization—SEs face substantial difficulties in leveraging AI to meet highly heterogeneous sports logistics demands under conditions of limited information sharing [55,56]. These demands include the need for differentiated transport capacities and storage facilities, as well as larger shipment volumes and higher usage frequency. Moreover, logistical requirements become particularly complex during major sporting events, where the coexistence of multiple stakeholders—such as athletes, event organizers, and logistics providers—makes it difficult to achieve an efficient trade-off between one-time utilization and low inventory costs [57,58].

4.4. Heterogeneity Tests

Significant differences in governance structure, resources, capabilities, and policy dependencies between SOEs and non-SOEs may result in varying impacts of AI [48]. Accordingly, we conduct a heterogeneity test between SOEs and non-SOEs, and the results are presented in

Table 8. Heterogeneity: enterprise type.

	(1) SOE	(2) Non-SOE	(3) SOE	(4) Non-SOE
	Supplier	Customer	Supplier	Customer
AI	0.0213	−0.150	−0.291 **	0.285 ***
	(0.22)	(−1.15)	(−2.09)	(4.33)
Control variables	Yes	Yes	Yes	Yes
Individual effect	Yes	Yes	Yes	Yes
Year effect	Yes	Yes	Yes	Yes
City effect	Yes	Yes	Yes	Yes
_cons	0.205	0.238	0.222	0.101
	(1.45)	(1.52)	−1.95	−0.95
N	230	230	355	355

z statistics in parentheses. ** p < 0.01, *** p < 0.001.

. The results indicate that AI has a more significant impact on the SCS of non-state-owned SEs compared to state-owned SEs, as evidenced by a significant reduction in supplier stability (coefficient = −0.291, z = −2.09) and a notable increase in customer stability (coefficient = 0.285, z = 4.33). Theoretically, on the one hand, non-state-owned SEs typically exhibit greater market adaptability and face higher cost pressures [59]. The introduction of AI encourages non-state-owned SEs to prioritize cost control and process optimization, which increases their focus on price and supplier substitutability in supplier selection, ultimately reducing supplier stability. On the other hand, non-state-owned SEs are more market-oriented [60], and AI enables these enterprises to optimize product customization, precision marketing, and customer service, thereby enhancing customer loyalty and improving customer stability.

The development, application, and commercialization of AI require continuous high investment [61], and variations in profitability may result in differing value realizations of AI applications across SEs. Thus, we examine the heterogeneity of the sample enterprises by categorizing them into three groups based on net profit: profit fluctuation, profit, and high profit. The results are presented in

Table 9. Heterogeneity: profitability.

	(1)	(2)	(3)	(4)	(5)	(6)
	Profit Fluctuation		Profit		High Profit
	Supplier	Customer	Supplier	Customer	Supplier	Customer
AI	−0.767 ***	−0.423	0.885	−1.300 *	0.0137 **	0.0838
	(−3.34)	(−1.63)	(1.07)	(−1.65)	(2.18)	(0.62)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Individual effect	Yes	Yes	Yes	Yes	Yes	Yes
Year effect	Yes	Yes	Yes	Yes	Yes	Yes
City effect	Yes	Yes	Yes	Yes	Yes	Yes
_cons	0.169	−0.433	−0.582	0.64	0.509 ***	−0.157
	(0.87)	(−1.72)	(−1.32)	(1.81)	(3.39)	(−0.44)
N	170	170	213	213	202	202

z statistics in parentheses. * p < 0.05, ** p < 0.01, *** p < 0.001.

. The results show that AI significantly weakens supplier stability in SE during periods of profit/loss fluctuations (coefficient = −0.767, z = −3.34). In a profitable state, AI significantly reduces customer stability (coefficient = −1.3, z = −1.65). However, in a state of high profitability, AI significantly enhances supplier stability (coefficient = 0.0137, z = 2.18). A possible explanation is that, under fluctuating profit and loss conditions, AI may induce a “bullwhip effect” in SE, leading to more frequent supplier changes in response to fluctuating demand and increased market uncertainty, thereby exacerbating supplier relationship instability [53]. Meanwhile, customer relationship theory suggests that SEs with low profitability may prioritize using AI for profit maximization at the expense of addressing customers’ long-term needs and personalized services, ultimately leading to a decline in customer stability [62]. Moreover, when highly profitable, SEs possess ample financial capacity to deploy AI for supply chain optimization, endowing them with greater resources and incentives to innovate and thereby strengthening customer retention [63].

5. Discussions and Case Studies

Our findings indicate that the application of AI significantly enhances the SCS of SE suppliers while notably reducing the stability of customers. The research findings corroborate certain conclusions presented in the fourth annual sports goods industry report, 2024 Global Sports Goods Industry Report: Taking Action, published by McKinsey and the World Federation of the Sporting Goods Industry (WFSGI). Specifically, executives in the sports industry have recognized that managing supply capabilities is often built on an effective Integrated Business Planning (IBP) system. Equipping the supply chain with advanced analytics and AI can help SE reduce general costs by 6%. The introduction of AI enables SE to reduce information asymmetry with suppliers, enhance the scientific accuracy of demand analysis, drive the transformation of efficient supply–demand matching models, and optimize inventory management.

For instance, Anta has introduced the sportswear industry’s first AI-driven design large-scale model, the Linglong Design Model. The model is trained on tens of millions of footwear and apparel data points accumulated by Anta over more than three decades, together with a foundational large-parameter model comprising hundreds of billions of parameters. This enables the firm to fully leverage AI’s advantages in large-scale data processing and efficiency enhancement. By timely transmitting market information and sales data, Anta has established efficient mechanisms for information sharing and mutually reinforcing R&D collaboration with its suppliers. As a result, the development cycle of blockbuster products has been reduced by 80%, the selection rate of design proposals has increased by 30%, and a mutually beneficial outcome for both the firm and its suppliers has been achieved. Meanwhile, Anta has launched the “AI 365” strategy, under which AI technologies are employed to analyze nationwide sports trends and social media signals in order to establish a sales forecasting mechanism. In parallel, the firm has introduced smart manufacturing and automated warehousing systems, enabling it to proactively plan supplier procurement strategies.

In contrast, the latter differs from the conclusions of Liang et al. [64] and Song et al. [65], as well as from the majority of academic findings. While the role of AI in mitigating supply chain risks for businesses has been largely validated, SEs still face barriers in their understanding and adoption of AI technology. The data analysis from PwC’s Global Sports Industry Survey-China Report reveals that 60% of respondents believe the sports industry has yet to fully embrace AI. The main obstacles identified include a lack of clear potential applications and relevance, as well as insufficient financial support. The underlying reason may lie in data sparsity, which hinders the value creation of AI for customer stability [66]. Data sparsity refers to noise, bias, missing values, and outdated information. In the Chinese market, advanced big data analytics for precise demand-side analysis have yet to be widely adopted in the sports industry. The inconsistent quality of existing data makes it challenging for market participants to accurately extract user demands, leading to a mismatch between production and demand.

For example, supported by digital technologies such as artificial intelligence, Li-Ning has introduced a “digital store manager” model that enables autonomous decision-making, real-time alerts, and voice-based interaction, thereby allowing the real-time collection of in-store customer shopping behavior data. However, Li-Ning’s R&D and design, manufacturing, and downstream retail segments remain relatively siloed, and the firm lacks a coordinated supply chain management system. As a consequence, data flows across these segments are slow and information sharing is limited, increasing the likelihood of strategic decision-making errors. As competition among sportswear brands in the Chinese market has increasingly incorporated technological capability as a key competitive dimension, Li-Ning has exhibited a lack of systematic and standardized management of its accumulated data assets and related performance metrics. This has been accompanied by persistently low R&D intensity, with R&D expenditure accounting for 1.8%, 2.1%, 2.2%, and 2.3% of revenue from 2021 to 2024, respectively. These figures are markedly lower than those of Anta (ranging from 2.3% to 2.8%) and substantially below the contemporaneous R&D intensity of Nike and Adidas, both of which fall within the 5–10% range. Insufficient investment in R&D, weak data governance effectiveness, and the limited realization of AI-driven value have also exerted a pronounced negative impact on Li-Ning’s customer stability. In 2024, the number of sales outlets declined by 1.1% year-on-year, average daily foot traffic decreased by 10–20%, revenue from directly operated channels fell by 0.35% year-on-year, and net profit declined by 5.5% year-on-year.

From the theoretical perspective of research on the relationship between AI and SCS, the resource-based view suggests that firms with advanced AI capabilities are better able to mobilize key resources such as capital and technology, thereby enhancing supply chain resilience and reducing the risk of disruptions [16]. Meanwhile, the dynamic capabilities theory posits that AI technology strengthens the ability to make more rational, realistic, and accurate decisions in circular supply chains [67]. Stakeholder theory suggests that under the behavioral pressures from consumers, social media, and investors, companies are inclined to leverage AI technologies and vast amounts of data (e.g., supplier evaluation reports) to identify patterns and errors, aiming to achieve higher supply chain compliance and enhance transparency [68]. However, these theories appear insufficient to explain our research findings at a theoretical level. A possible reason is that, during the development of SE, there is a need to simultaneously address the integrated demands of production, technology, and market. In this context, a flexible production model becomes more critical. Yet, challenges such as weak infrastructure, inefficient flow of resources, and insufficient innovation exacerbate the vulnerability, uncertainty, and risks within the sports manufacturing supply chain. Therefore, we argue that, in analyzing the impact of AI on the stability of SEs’ supply chains, the supply chain management theory’s focus on the flow of goods, information, and capital provides an effective framework for theoretically explaining the relationship between AI and SCS in SE.

6. Conclusions and Implications

Building on supply chain management and systems theories to examine AI’s effects on SEs’ SCS, we develop a DML model using panel data from 45 Chinese listed SEs over 2012–2023. Overall, artificial intelligence exerts heterogeneous effects on supply chain stability in sports enterprises (See

Table 10. Conclusions.

Key Findings	Hypothesis Verification Scenario	Theoretical Contributions
AI can significantly enhance supplier stability for SE, but it markedly reduces customer stability.	Partially supports hypothesis H1a and H2b	Validated the research framework integrating supply chain management theory and systems theory. The research findings provide valuable insights into the applicability of the Resource-Based View, Dynamic Capabilities Theory, and Stakeholder Theory. Expanded the application of customer relationship theory.
Attracting talent is the primary channel connecting AI with the stability of SEs’ supply chains.	Partially supports hypothesis H3b
AI has significantly enhanced supplier stability for SE operating at substantial profit margins while reducing supplier stability for non-state-owned SE experiencing fluctuating profitability or losses.
AI has significantly enhanced customer retention for non-state-owned SE while reducing customer retention for profitable SE.

). Specifically, AI significantly enhances supplier stability while simultaneously leading to a significant decline in customer stability. Talent attraction constitutes one of the primary mechanisms underlying this effect. More specifically, with respect to supplier stability, AI significantly enhances supplier stability among SEs with exceptionally high profitability while reducing supplier stability for non-state-owned firms and for those experiencing volatile performance, including transitions between profit and loss. With respect to customer stability, AI significantly enhances customer stability among non-state-owned SEs while reducing customer stability for firms operating in a profitable state. Our findings indicate that AI operates as a “double-edged sword.” To bolster data reliability and transparency in the sports industry, AI’s computational capacity and precision must be enhanced through continuous algorithmic refinement and data optimization, and an independent third-party audit body should be engaged to verify and oversee sports data. At the enterprise level, SE should adopt AI strategically rather than indiscriminately, emphasizing internal information sharing, the exchange of technical expertise among scientific and technological personnel, and collaborative innovation.

7. Limitations

While our study addresses the questions of ‘if’ and ‘how’ AI can impact the SCS of SE, it still has several limitations. First, our study relies on publicly available data from listed SEs in China, resulting in a relatively small sample size that limits the generalizability of our findings. Future research could expand the sample by incorporating non-listed SEs through field studies. Second, due to data acquisition challenges, we employed text analysis to assess AI application in SE, which provides an incomplete reflection of its extent. Future studies could explore more robust and scientifically rigorous measurement methods. Finally, while our study focuses on China and provides generalizable recommendations based on its findings, future research should include sports firms from other countries for a broader empirical analysis.