AI technologies affording the orchestration of ecosystem-based business models: the moderating role of AI knowledge spillover
Data collection and sample
To test the proposed hypotheses, we collected an initial sample of A-share listed firms in the Shanghai and Shenzhen Stock Exchange from 2014 to 2021. Artificial intelligence probably began to be gradually disclosed in China’s corporate annual reports in 2014, signifying AI’s rapid emergence. Therefore, we initialized 2014 as the starting year of our study. After obtaining the initial database, we applied specific criteria to filter out irrelevant firms as follows. Firstly, we excluded firms with special treatment, labeled as ST or *STFootnote 1 in our obtained data. Secondly, we removed firms that were marked as suspended or terminated. Thirdly, we deleted firms whose annual reports did not contain the data required for our research framework. Finally, a total of 3632 pieces of yearly data with complete information were obtained from 454 companies in the years from 2014 to 2021.
We utilized the China Stock Market and Accounting Research (CSMAR) database to obtain crucial information such as financial performance and supply chain data. We used the dataset of corporation social responsibility (CSR) from Hexun Score (HXS) to measure part of the ecosystem-based business model. The Hexun Score is a commonly used dataset in China that provides an intuitive understanding of the performance of each company with its various stakeholders, in line with our measurement of an ecosystem-based business model.
Measurement
Dependent variable: EBM
EBM is generated by the entropy method of calculating four indicators in the CSR data of the Hexun Score and nine financial indicators of the company. The innovation of BM is mainly measured from three dimensions value creation, value delivery, and value acquisition (Clauss, 2017; Teece 2010). Value creation refers to a series of business activities and the cost structure of the enterprise to produce and supply products or services to meet the needs of target customers, including the core competencies of daily operations, capital situation, and other internal elements, which mainly include the ability to utilize funds and the ability to pay off debts. Hence, the current ratio (X1), the debt coverage ratio (X2), and the capitalization ratio (X3) are selected to measure value creation. Value delivery dimension refers to the way and means by which consumers receive products or services, and how to establish sustainable long-term consumption relationships with consumers, where operational capability is the key element, so inventory turnover ratio (X4), receivables turnover ratio (X5), and total asset turnover ratio (X6) are selected to measure value delivery. Value capture refers to the way a company controls and reduces costs to create more profit points and profitability. Profitability and growth ability are key factors thus the increasing rate of main business revenue (X7), net profit growth rate (X8), and main business profit margin (x9) are selected to measure value acquisition.
The above measurements are extended by combining the definitions of EBM, which refers to an organic, dynamic, and environmentally friendly open system in which a large number of stakeholders from the main focal firm and other related nested business models play different roles and share their outcomes such as species in a system (Chin et al. 2022; Konietzko et al. 2020). According to this logic and based on the measurement of basic BM, this study extended measures EBM in three other dimensions – profit, people, and the planet. The CSR scored by the Hexun Score is assessed in terms of four indicators: shareholders’ responsibility, employees’ responsibility, suppliers’, customers’ and consumers’ rights and interests’ responsibility, and environmental responsibility with the first three (X10X11X12) reflecting the people dimension, and the fourth one (X13) reflecting the planet dimension. Another dimension of EBM is profit, and the measurement of value capture mentioned above is actually from the perspective of economic efficiency. Summarily, our study utilizes these 13 indicators and applies the entropy method to calculate a new value to comprehensively measure EBM from the perspectives of basic BM, economic benefits, stakeholders, and earth responsibilities.
The entropy method is a scientific and objective way of determining weight judgments based on the amount of information contained in the data. The greater the dispersion of the data, the greater the impact of the indicator on the overall evaluation. The following are the main steps of this method:
-
(1)
Standardize the collected data. Since different indicators have different scales and units, they need to be standardized, and we choose the min-max method to standardize the raw data. In Eqs. (1) and (2), minXtij and maxXtij are the minimum and maximum observed values, Xtij is the value of the indicator j in the t year of the company i and Ztij is the standardized result with the range of values 0 to 1.
Positive indicator standardization:
$$Z_tij=\fracX_tij-\min X_tij\max X_tij-\min X_tij$$
(1)
Negative indicator standardization
$$Z_tij=\frac\max X_tij-X_tij\max X_tij-\min X_tij$$
(2)
-
(2)
Calculate the entropy value of the collected data. The entropy value of the data can be calculated by equations. In Eqs. (3), (4), and (5), Ptij is the percentage of standardized data Ztij, and Ej is the entropy value of indicator j.
$$P_tij=\fracZ_tij\sum _t=1^m\sum _i=1^kZ_tij(\rmt=1,2,\cdots \rmm;i=1,2,\cdots \rmk)$$
(3)
$$k_1=\frac1Ln\left. (m\times k\right)$$
(4)
$$E_j=-k_1\mathop\sum \limits_t=1^m\mathop\sum \limits_i=1^kP_tij\,LnP_tij$$
(5)
-
(3)
Calculate the entropy weights of the collected data. After deriving the entropy value Ej of indicator j, the entropy weight of indicator j can be obtained. From Eq. (6), the indicator Dj is the utility value and the entropy weight Wj can be solved by Eq. (7).
$$D_j=1-E_j$$
(6)
$$W_j=\fracD_j\sum _j=1^nD_j$$
(7)
-
(4)
Calculate the EBM score. Using the weighting method, the EBM score of company i in year t, i.e., EBMti, is calculated from Eq. (8).
$$EBM_ti=P_{{tij}}\times W_j$$
(8)
Independent variable: the degree of AI
AI, as one of the leading technologies of the enterprise, will be disclosed in the annual report with a summary and guidance, reflecting the strategic characteristics and prospects of the enterprise. Some scholars (Verhoef et al. 2021; Zhai et al. 2022) used the word frequency of “ABCD” technologies such as AI, blockchain, cloud computing, and big data to measure the degree of digital transformation. Therefore, we believe that it is feasible and scientific to measure the degree of AI from the perspective of word frequency statistics related to AI technologies in the annual report of listed enterprises. The AI data utilized comes from CSMAR, where previous researchers extracted AI-related datasets such as Artificial Intelligence, Business Intelligence, Image Understanding, Intelligent Data Analytics, Intelligent Robotics, Machine Learning, Deep Learning, Semantic Search, Biometrics, Face Recognition, Speech Recognition, Identity Authentication, Autonomous Driving, and Natural Language Processing from the annual reports by using the Pathon Crawler function (Wu et al. 2021). A review by experts and scholars in this relevant domain was also initiated, which approved the reliability of the measurement.
Moderating variables: direct knowledge spillover and indirect knowledge spillover
Direct knowledge spillover
We utilized R&D stockt to examine direct knowledge spillover, as previously done by scholars Serrano-Domingo and Cabrer-Borrás (2017). Direct knowledge spillover is primarily caused by communication on specific R&D projects. The investment in cooperative innovation determines the number of projects or topics that cooperate on, the frequency of the resulting exchanges, and the direct knowledge spillover. According to Xu et al. (2023), we use the perpetual inventory method to measure the R&D stock. The specific formula is as follows:
$$R\& D\,stock_t=\left(1-\rm\delta \right)R\& D\,stock_t-1+RI_t$$
(9)
where R&D stockt is the value of phase t, R&D stockt−1 is the R & D stock value of phase t − 1, RIt is the R & D expense of phase t, δ represents R&D depreciation rate. Xu et al. (2023) estimated the profitability brought by corporate R&D and also verified the reasonableness of the estimated R&D depreciation rate of Chinese firms from 1990 to 2021 through domestic and international comparisons with the same industry as well as a robustness analysis, and the final measurements show that the R&D depreciation rate of corporates in the Chinese market is 26.37%. Therefore, we take 26.37% as the R&D depreciation rate of our research enterprises.
Indirect knowledge spillover
We use the management cost ratio in revenue to test the indirect knowledge spillover by following the footsteps provided by Singh (2005) and Serrano-Domingo and Cabrer-Borrás (2017). The company’s operations play a crucial role in facilitating knowledge transfer. While communication among managers, employees, and other external organizations may not result in direct knowledge spillover, it can still have indirect effects. In other words, an invisible form of knowledge floats inside the organization and is acquired by people through better communication, thus increasing the organization’s overall knowledge base.
Control variables
To ensure that an organization’s EBM is not affected by other variables, we control several firm-level variables (i.e., firm size, intangible asset ratio, and supply chain concentration) and individual-level variables (i.e., CEO’s age and overseas background) as shown in Table 1.
Firm size
The size of an enterprise reflects the scale of its operations and resource inputs. Larger firms usually have more human, financial, and technological capabilities, which can influence their ability to innovate their business models.
Intangible asset ratio
Intangible assets are non-material assets owned by the enterprise, such as intellectual property rights, brand value, innovation capability, goodwill, etc., which will influence the innovation capability of firms. Thus, the intangible asset ratio is used as one of the control variables.
Supply chain concentration
According to Osterwalder and Pigneur’s business model canvas (2010), the formation of EBM is influenced by both customers and suppliers. Following this idea, we chose supply chain concentration as a control variable by calculating the average of the total purchase ratio for the top five suppliers and the total sales ratio for the top five sellers, as shown in formula (5).
$$Supply\,Chain\,Concentration=\frac12\left(\fractop\,five\,suppliers\text’\,purchase\,total\,purchases+\fractop\,five\,sellers\text’\,salestotal\,sales\right)$$
(10)
CEO age
CEO age was used as a control variable to assess the influence of top managers’ personalities and attitudes on a company’s strategic decisions. Young CEOs are likelier to adopt BM innovation than CEOs with a conventional mindset (Eugenio and Sicilia 2020).
CEO’s overseas background
As mentioned, firm CEOs with overseas backgrounds may be more inclined toward adopting EBMs. Therefore, the CEO’s overseas background was used as a control variable. We set it as a dummy variable stating that if the CEO has overseas work experience or educational background, the value is 1; otherwise, it is 0.
The model
Our study tests the theoretical model and hypotheses using panel data regression. Panel data contains data in both the time dimension and the firm dimension, so the heterogeneity between firms and the variation in the time dimension need to be taken into account in the analysis. The dependent variable AI in this study varies significantly in terms of firm heterogeneity and time dimension, and the use of fixed-effect regression can improve the accuracy and reliability of the analysis. We also performed Hausman’s test (χ2 = 147.63, Prob > χ2 = 0.0000), and which results showed that fixed-effect regression is fit for our research.
To test hypothesis H1, the regression equation has been established as the following formula (11):
$$EBM_i,t=\alpha _0+\alpha _1AI_i,t+\alpha _2AI2_i,t+\alpha _3\mathop\sum \limits_t^i\delta \cdot Controls_i,t+\rm\mu _i+\rm\lambda _t+\varepsilon _i,t$$
(11)
In formula (11), AI is squared to generate the quadratic term AI2, which is used to test whether there is a U-shaped relationship between AI and EBM. Controlsi,t represents the control variables. μi, λt represent firm-fixed and time-fixed effects, respectively. εi,t represents the random disturbance term. i represents the firm, and t represents the year.
To test hypothesis H2a, which explores the moderating effect of direct knowledge spillover (DKS) on the relationship between the main variables, we added DKS, the interaction items AI*DKS and AI2*DKS to the regression equation. The regression equation has been established as the following formula (12):
$$\beginarraylEBM_i,t=\gamma _0+\gamma _1AI_i,t+\gamma _2AI2_i,t+\gamma _3AI_i,tDKS_i,t+\gamma _4AI2_i,tDKS_i,t\\\qquad\qquad\;\;+\,\gamma _5DKS_i,t+\gamma _6\mathop\sum \limits_t^i\delta \cdot Controls_i,t+\rm\mu _i+\rm\lambda _t+\varepsilon _i,t\endarray$$
(12)
To test hypothesis H2b, which examines the moderating effect of indirect knowledge spillover (IDKS) on the relationship between the main variables, we added IDKS, the interaction items AI*IDKS and AI2*IDKS to the regression equation. The regression equation has been established as the following formula (13):
$$\beginarraylEBM_i,t=\gamma _0+\gamma _1AI_i,t+\gamma _2AI2_i,t+\gamma _3AI_i,tIDKS_i,t+\gamma _4AI2_i,tIDKS_i,t\\\qquad\qquad\;\;+\,\gamma _5IDKS_i,t+\gamma _6\mathop\sum \limits_t^i\delta \cdot Controls_i,t+\rm\mu _i+\rm\lambda _t+\rm\varepsilon _i,t\endarray$$
(13)
link
