The Persistent Innovation Effect of Platform Ecosystem Embeddedness

Abstract

In the context of the digital economy, corporate innovation is shifting from an internally driven, linear model to a multilateral, collaborative mechanism enabled by platform ecosystems. This study integrates persistent innovation into the analytical framework of platform ecosystem embeddedness and evaluates its impact from two dimensions: innovation input and innovation output. Using panel data from Shanghai and Shenzhen A-share listed companies between 2012 and 2023, we apply fixed effects models and mediation analysis to empirically examine the mechanisms through which platform ecosystem embeddedness affects innovation. The results show that platform ecosystem embeddedness significantly enhances companies’ persistent innovation. Specifically, it promotes innovation input by reducing supply–demand coordination costs and improving the operational efficiency of the company, and it boosts innovation output by facilitating knowledge flow and knowledge creation. Furthermore, the effects are more pronounced in industries characterized by high competition, rich regional resources, high-tech orientation, and low environmental pollution, where both input and output are strengthened. In contrast, in less competitive, resource-constrained, non-high-tech, and heavily polluting industries, the impact is mainly reflected in innovation output, with limited influence on input.

1. Introduction

Against the backdrop of accelerated digital economic development, the paradigm of enterprise innovation is gradually shifting from an internally driven, linear process toward a platform-based ecosystem characterized by multilateral collaboration and continuous iteration [1,2,3]. While technological advances have fueled waves of innovation [4,5], maintaining long-term innovation capacity in dynamic environments remains a critical challenge. In the case of Chinese enterprises, data from 2012 to 2021 show that only 30.28% of companies listed on the Shanghai and Shenzhen A-share markets exhibited persistent innovation capabilities. In contrast, nearly 70% suffered from the structural issue of “frequent initiation, limited continuity” in their innovation activities [6]. This highlights that beyond achieving technological breakthroughs, sustaining the continuity of innovation paths and the stability of resource commitment has become a key bottleneck to high-quality development.

As a core infrastructure of the digital economy, platform ecosystem embeddedness has reshaped the logic of resource allocation and coordination mechanisms among enterprises. The rapid rise of industrial internet platforms—such as Haier’s COSMOPlat and Huawei’s OceanConnect IoT—has established multilateral resource networks and modularized architectures that serve as critical vehicles for enterprise digital transformation and technological upgrading [7,8] (National Industrial Internet Platform Application Data Map, 2023). Existing studies have extensively examined how platforms contribute to enhancing immediate innovation performance, optimizing product portfolios, and enabling cross-domain collaboration [5,9,10,11,12,13], and have underscored the role of platform governance structures and technical interface standards in facilitating innovation behavior [8,14].

However, whether platform ecosystems can support enterprises in sustaining long-term technological accumulation and knowledge renewal remains underexplored in both theory and empirical evidence. Within the innovation literature, persistent innovation is commonly defined as a company’s ability to consistently allocate resources and steadily generate innovation output over an extended period [15,16,17]. Most of the underlying mechanisms have focused on internal company-level resource endowments, explaining persistent innovation through the lenses of the “success-breeds-success” effect [18], sunk cost commitment [15], and increasing returns to innovation [19]. These perspectives, however, typically treat companies as isolated and autonomous entities, overlooking the increasingly embedded reality of companies operating within platform structures. In the digital era, companies’ innovation capabilities rely increasingly on access to heterogeneous resources, dynamic knowledge networks, and technical toolkits provided by platforms, rather than on internal accumulation alone. Understanding how platform ecosystems influence the persistence of innovation thus represents a critical theoretical gap. Although some studies have begun to explore how platforms enable innovation from an organizational perspective, the literature still primarily concentrates on the immediate effects of platforms on breakthrough innovation, green innovation, or business process innovation [3,9,20]. To date, little is known about whether, and through which mechanisms, platform embeddedness helps enterprises maintain stable resource commitment on the input side and sustained knowledge renewal on the output side. A structured and mechanism-oriented analytical framework remains absent from current scholarship.

This study seeks to address these gaps by focusing on how platform ecosystem embeddedness affects companies’ persistent innovation. Using panel data of Chinese A-share listed companies from 2012 to 2023, we construct a fixed effects model to empirically investigate the mechanism through which platform ecosystem embeddedness affects persistent innovation. Specifically, we aim to answer three key questions: (1) Does platform ecosystem embeddedness significantly promote persistent innovation? (2) Do the mechanisms through which platform ecosystem embeddedness influences innovation differ between the input and output dimensions? (3) Is the enabling effect of platform ecosystem embeddedness contingent on contextual factors?

By addressing these questions, the study seeks to reveal how platform ecosystems enable the “structural continuity” of innovation across the input–output process and offers theoretical insights into long-term innovation mechanisms under platform-dominated environments. The potential marginal contributions of this research are three-fold:

First, theoretical extension: This study introduces the concept of persistent innovation into the research framework of platform ecosystems, operationalized through a dual-dimensional measurement of innovation input and innovation output, thereby expanding the current understanding of platform-enabled innovation.

Second, mechanism identification: This study constructs an integrated analytical framework of “embeddedness-synergy-persistent innovation,” revealing that platform ecosystem embeddedness primarily drives innovation input through resource synergy, while promoting innovation output through knowledge synergy, thus enriching the understanding of structural evolution in innovation processes within platform environments.

Third, contextual identification: This study identifies the contextual dependence of platform-enabled innovation, uncovering significant variations in its effects across dimensions such as industry competition intensity, regional development levels, and company characteristics. These findings contribute to differentiated innovation policy design and strategic platform development.

2. Theoretical Analysis and Research Hypotheses

2.1. Platform Ecosystem Embeddedness and Persistent Innovation

Platform ecosystem embeddedness refers to a stable yet dynamic organizational connection between companies and platform ecosystems. This connection is established through technological interfaces, data connectivity, and rule-based coordination mechanisms [4]. It facilitates interaction among multiple stakeholders [21] and extends beyond basic technological interoperability. It also enables resource sharing, strategic alignment, and the co-embedding of knowledge. In contrast to traditional social embeddedness, which is typically built on dyadic transactional relationships [22], platform ecosystem embeddedness features stronger multilateral coordination, modular openness, and structural flexibility [4,23]. For instance, Apple developers operate under platform-defined rules while retaining the flexibility to use ecosystem toolkits and API modules. This setup allows them to balance platform control with individual autonomy. Such embeddedness, shaped by technological architecture and institutional design, disrupts the lock-in effects typically caused by path dependence and asset specificity in traditional supply chains. As a result, it fundamentally reshapes how companies engage in innovation networks.

Persistent innovation refers to a company’s sustained innovation efforts over time, characterized by continuous resource investment and knowledge accumulation [15,16,17]. It includes activities such as technological iteration, product refinement, and process optimization. In contrast to breakthrough innovation—which is high-risk and disruptive—persistent innovation emphasizes steady progress along a defined trajectory. Two key factors underpin this process: on the resource side, companies must maintain stable R&D investment, technical personnel, and experimental infrastructure [24]; on the knowledge side, continuous experiential learning and organizational memory are essential for preserving technological continuity and enhancing the value of knowledge assets [25]. Within traditional linear supply chains, companies often encounter challenges such as limited access to resources, weak knowledge diffusion, and inefficient collaboration—all of which constrain persistent innovation. In contrast, the multilateral openness and programmable architecture of platform ecosystems enable a novel embedded innovation mechanism that differs fundamentally from conventional supply network logic.

From the perspective of persistent innovation motivation, platform ecosystem embeddedness can enhance companies’ incentives for long-term innovation. Platform owners typically exercise significant architectural control and may impose strategic alignment pressure on embedded companies through rule-setting mechanisms. For example, industrial internet platforms often mandate regular interface upgrades by equipment manufacturers to ensure technological compatibility, thereby encouraging continuous technical iteration [21]. Gawer and Cusumano [4] emphasize that platform owners use tools such as API version control and application update protocols to synchronize complementors with the platform’s evolution, maintaining the momentum of technological and product advancement. In addition, the long-term partnerships and stable transaction expectations fostered by platform ecosystem embeddedness help mitigate short-term opportunism and reduce the perceived risks associated with innovation projects [26]. For instance, small and medium-sized component suppliers embedded in new energy vehicle platforms benefit from stable order flows, which enable more confident investments in R&D related to new materials and energy technologies. Furthermore, intense competition within platform ecosystems—such as the App Store or Amazon—reinforces internal incentives for continuous innovation. Ranking systems and user feedback mechanisms create a “Red Queen effect,” whereby companies must consistently update and improve their products to maintain their competitive position within the ecosystem [27,28].

From the perspective of innovation capacity, platform ecosystem embeddedness enhances companies’ ability to accumulate and integrate resources. First, the extensive industry datasets aggregated by platforms serve as critical inputs for product iteration [29]. For example, Haier’s COSMOPlat allows home appliance companies to refine product design and functionality based on user behavior feedback derived from open data and operational logs, thereby supporting continuous iterative innovation. Second, open digital tools and API interfaces provided by platforms lower the trial-and-error costs associated with technological innovation. Companies embedded in Alibaba Cloud, for instance, can reconfigure their products using shared modules such as AI algorithm libraries and cloud computing services. According to the Ministry of Industry and Information Technology (National Industrial Internet Platform Application Data Map, 2023), applications of industrial internet platforms have improved average company productivity from 15% to 30%. Third, platform ecosystems facilitate knowledge spillovers through technical communities and collaborative R&D initiatives. Jacobides et al. [23] note that the tacit knowledge accumulated by platform owners—such as expertise in digital transformation—can be transferred across organizational boundaries, helping embedded companies overcome path dependence. For example, the NVIDIA CUDA Developer Program enables efficient knowledge sharing in AI model development across the ecosystem, significantly reducing product development cycles for partner companies. In contrast to traditional supply chains, resource sharing within platform ecosystems is non-zero-sum. Embedded companies can drive innovation and create value without requiring exclusive access to resources [30], thus enabling sustained and collaborative innovation among multiple stakeholders.

From the perspective of the innovation environment, platform ecosystem embeddedness offers a supportive context for sustaining innovation. First, risk-sharing mechanisms commonly found in platform ecosystems—such as “equity-holding without control” [4]—provide stable innovation capital while preserving companies’ decision-making autonomy. For instance, Tencent’s combination of equity investment and traffic support for JD.com offered both platform-based resources and financial buffers without interfering in its operations, thereby enhancing the JD.com capacity for persistent innovation. Second, digital governance tools like blockchain and smart contracts are increasingly used to ensure the traceability and automated protection of innovation outcomes. These tools help reduce the risks of intellectual property infringement and disputes, thus supporting the continuity of innovation efforts [31]. Finally, the multilateral network effects of platforms greatly enhance the speed and quality of market feedback mechanisms [12]. The diverse ecosystem of actors connected through platforms enables real-time validation and adjustment of innovation outcomes, providing companies with timely, high-quality feedback that fuels a continuous cycle of innovation, feedback, and re-innovation [32].

Based on the above analysis, we propose the following research hypothesis:

H1: 

Platform Ecosystem Embeddedness significantly promotes persistent innovation in companies.

2.2. The Impact of Platform Ecosystem Embeddedness on Persistent Innovation Input

In the digital economy, companies are shifting their innovation activities from a closed, internally driven R&D model to a platform-centered, open, and collaborative approach. Platform ecosystems, as structured organizational systems, provide critical functions such as data aggregation, resource coordination, and technological intermediation [33]. These functions help alleviate common barriers to persistent innovation input, including resource scarcity and inefficient allocation. In contrast to traditional R&D investments that rely heavily on internal resource accumulation, platform ecosystem embeddedness enables companies to reshape supply–demand coordination and reconfigure operational resources. This restructuring fosters resource synergy, thereby enhancing the firm’s ability to sustain innovation input over time.

First, platform ecosystem embeddedness significantly reduces the coordination costs of innovation input by optimizing the supply–demand matching mechanism. On the one hand, platforms aggregate user behavior data, industrial supply information, and technology-matching rules. These resources provide a foundation for companies to identify and access external innovation inputs more efficiently [29]. For instance, Alibaba Cloud’s ET Industrial Brain leverages a large-scale industrial cloud database—comprising manufacturing process data and standardized customer datasets—to enable precise supply–demand alignment. This capability substantially shortens the technology procurement cycle for manufacturing companies. On the other hand, platforms also establish unified technical standards, such as API protocols and data format specifications. These standards reduce technical adaptation costs in inter-company collaboration. By improving system interoperability and lowering dependence on proprietary solutions, platforms support the smooth flow and flexible recombination of innovation resources across ecosystem participants [34].

Second, platform ecosystem embeddedness enhances companies’ capacity to reinvest in innovation by unlocking operational efficiency dividends. On the one hand, companies embedded in platforms gain on-demand access to shared digital resources such as cloud computing, artificial intelligence, and design tools. This access helps convert traditionally high fixed costs into flexible, controllable variable costs [30], thereby improving the efficiency of innovation-related resource acquisition. For example, firms using Microsoft Azure AI services reported a 7% cost reduction by the third year, along with lower licensing, maintenance, and support expenses compared to legacy systems or third-party solutions [35]. On the other hand, platforms offer integrated solutions that support data-driven process reengineering across R&D, production, and sales. These solutions eliminate redundant operations and break down information silos, thereby improving companies’ overall operational efficiency [36,37].

Improvements in operational efficiency not only reduce companies’ marginal costs but also free up slack resources. These resources provide both financial and managerial capacity to support high-risk, long-cycle persistent innovation projects [38]. As Teece [39] notes, transforming efficiency gains into innovation capability requires companies to reconfigure and redeploy their internal resources. Platform ecosystems offer modular resources and digital infrastructure, enabling companies to redirect slack resources into R&D budgets, technology upgrades, or cross-functional innovation initiatives. For example, a small manufacturing company embedded in a platform ecosystem used capital saved through improved inventory turnover to establish an internal data analytics team. This reinvestment further advanced its transformation toward smart manufacturing. This intermediary mechanism—connecting efficiency gains with resource reallocation, capability restructuring, and ultimately with increased innovation input—serves as a key pathway through which platform ecosystem embeddedness promotes persistent innovation.

In summary, platform ecosystem embeddedness facilitates persistent innovation input by first lowering supply–demand coordination costs and then by improving companies’ operational efficiency to release reconfigurable resources. These mechanisms collectively provide institutional space and capacity foundations essential for sustaining innovation input. Based on this reasoning, we propose the following hypotheses:

H2a: 

Platform ecosystem embeddedness promotes companies’ persistent innovation input by reducing supply–demand coordination costs.

H2b: 

Platform ecosystem embeddedness promotes companies’ persistent innovation input by enhancing operational efficiency of the company.

2.3. The Impact of Platform Ecosystem Embeddedness on Persistent Innovation Output

In traditional closed innovation systems, companies often encounter several barriers to innovation output. These include limited access to external knowledge, technological path dependence, and slow diffusion of innovation outcomes [40]. Together, these factors lead to persistent innovation bottlenecks. The emergence of platform ecosystems creates new opportunities to overcome these challenges by reshaping the mechanisms that support persistent innovation output. As a key channel for connecting companies to digital knowledge networks, platform ecosystem embeddedness facilitates engagement in multilateral, data-driven knowledge synergy systems. Through this embeddedness, companies can enhance their innovation output capacity via two primary pathways: knowledge flow and knowledge creation.

On the one hand, platform ecosystems enhance inter-company knowledge flow through architectural standardization and intelligent coordination mechanisms. First, platforms impose standardized technological protocols—such as API interfaces and modular communication frameworks—that codify tacit experiential knowledge and enable its transfer across organizational boundaries [41]. Tiwana [34] argues that such interfaces improve the efficiency of external knowledge absorption and reduce friction during knowledge exchange. Second, most platform ecosystems exhibit a “small-world” network topology. In this structure, companies benefit from dense local ties combined with cross-domain connections, which accelerate knowledge diffusion throughout the ecosystem [42]. This configuration not only increases the internal connectivity of knowledge flows, but also enhances the likelihood of leveraging weak ties during knowledge search processes. Third, platforms utilize algorithmic recommendation systems and knowledge graphs to match knowledge supply with demand more precisely among companies [29,34]. These intelligent tools reduce search costs and eliminate redundant information. They also accelerate problem identification and resolution in the innovation process, thereby deepening inter-company knowledge collaboration and sharing. As platforms continuously learn from user behavior, transaction history, and semantic linkages, their adaptive matching capabilities improve over time. This fosters a dynamic, demand-driven system of knowledge flow that strengthens companies’ responsiveness to complex technological environments and supports persistent innovation.

On the other hand, platform ecosystem embeddedness enhances companies’ capacity for knowledge creation. It shifts innovation output from a model of “acquisition–absorption” to one of “recombination–breakthrough.” First, the modular architecture provided by platforms gives companies the flexibility to reconfigure their existing knowledge assets [43]. These modular combinations significantly broaden innovation trajectories. Companies can integrate diverse external modules for collaborative innovation or embed their own knowledge into broader innovation networks via open interfaces. This structure promotes high recombinability and substitutability of knowledge elements, enabling rapid restructuring of knowledge bases to meet changing market demands and generate novel solutions. Second, cloud-native development tools and low-cost testing environments—such as sandboxes and simulation interfaces—lower the barriers to innovation experimentation. In clinical research, Saville and Berry [44] note that platform-based trials enable more flexible and efficient testing, allowing multiple treatments to be evaluated with fewer resources and in less time. Similarly, in the corporate context, these environments support frequent prototype testing and iterative technology development under conditions of low risk and cost. This iterative experimentation generates experiential learning and supports new knowledge creation. Third, the openness and cross-domain connectivity of platform ecosystems facilitate the recombination of heterogeneous knowledge. This recombination drives non-linear innovation and breakthrough outcomes. According to Fleming’s [45] theory of “recombinant advantage,” integrating knowledge across domains increases the likelihood of breakthrough innovation by producing asymmetrical value. Platform ecosystem embeddedness provides both the organizational infrastructure and technological pathways needed to support such combinatorial innovation.

In summary, platform ecosystem embeddedness systematically enhances companies’ persistent innovation output by optimizing the mechanisms of knowledge flow and stimulating knowledge creation. This dual mechanism not only overcomes the limitations of traditional output models that rely solely on internal accumulation, but also establishes a dynamic and self-reinforcing innovation trajectory. It reflects the deep enabling role of platform ecosystems on the output dimension of persistent innovation.

Based on the above, we propose the following research hypotheses:

H3a: 

Platform ecosystem embeddedness promotes companies’ persistent innovation output by improving knowledge flow.

H3b: 

Platform ecosystem embeddedness promotes companies’ persistent innovation output by enhancing knowledge creation.

The theoretical framework of this study is illustrated in Figure 1.

3. Research Design

3.1. Data Sources

This study selects Chinese A-share listed companies from the Shanghai and Shenzhen stock exchanges over the period from 2012 to 2023 as the research sample. Companies designated as ST and *ST, as well as those with missing key information, were excluded from the analysis. To mitigate the influence of outliers, all continuous variables were winsorized at the 1st and 99th percentiles. The final dataset comprises 9279 company-year observations.

The data used in this study are sourced as follows: platform ecosystem embeddedness data are obtained through web scraping of companies’ annual reports; annual report data are retrieved from CNINFO (www.cninfo.com.cn); patent data are sourced from the CNRDS database; and all other company-level variables are obtained from the CSMAR database. Web scraping and textual frequency analysis are conducted using Python 3.13.5, and the econometric analysis is performed using STATA 16.0.

3.2. Variable Construction

• Independent Variable

Platform ecosystem embeddedness (PEE). Drawing on the methodology of Chen and Li [46], this study constructs the independent variable based on the frequency of keywords related to platform ecosystem embeddedness extracted from companies’ annual reports. As a key strategic orientation for traditional companies pursuing digital transformation and high-quality development, platform ecosystem embeddedness is typically disclosed in annual reports, which serve as programmatic and strategic documents in corporate communications.

Mentions of platform-related ecosystem concepts in these reports often reflect companies’ strategic positioning, future vision, and planned actions in areas such as platform collaboration, data integration, and ecosystem partnerships. Moreover, as formal disclosures aimed at investors and regulatory agencies, annual reports are generally characterized by their seriousness and forward-looking nature. Their contents not only convey the company’s external strategic commitments but also reflect, to some extent, its internal intent to implement the stated strategies.

Therefore, it is both feasible and scientifically robust to extract high-frequency platform-related keywords from annual reports to construct a company-level measure of platform ecosystem embeddedness. The total keyword frequency for each company is aggregated and then log-transformed to construct the final measure of platform ecosystem embeddedness. A detailed list of the relevant keywords is provided in

Table 1. Keyword statistics of platform ecosystem embeddedness.

Keyword Statistics

Platform Ecosystem Embeddedness

Internet solutions, Internet thinking, Internet actions, Internet strategy, Internet model, Internet business model, Internet, Internet+, online and offline, from online to offline, online and offline integration, O2O, Cloud computing, stream computing, graph computing, in-memory computing, secure multi-party computation, brain-inspired computing, green computing, cognitive computing, converged architecture, 100-million-level concurrency, EB-level storage, Internet of Things (IoT), cyber-physical systems, big data, data mining, text mining, data visualization, heterogeneous data, credit investigation, augmented reality (AR), mixed reality (MR), virtual reality (VR), Internet platforms, Internet technology, mobile Internet, Internet services, Internet applications, Internet, B2B, C2C, B2C, C2B, industrial Internet, industrial platform, Internet ecosystem

Source: refer to the paper of Chen and Li [46].

.

• Dependent Variables

Persistent innovation input (Inno_in) and persistent innovation output (Inno_out). In measuring persistent innovation, conventional approaches, such as patent continuity indices [47] or binary classification methods [48], fall short of capturing the nuanced and dynamic nature of innovation persistence in platform ecosystems. Following Yang and Yang [6], this study adopts ratio-based measures to reflect the continuity of both innovation input and output. Specifically, we construct two indicators based on the ratio of current-period R&D investment and patent applications to the sum of those in the preceding two periods. The calculation formulas are as follows:

$$\mathrm{I}\mathrm{n}\mathrm{n}{\mathrm{o}\_\mathrm{i}\mathrm{n}}_{\mathrm{i}\mathrm{t}}=\mathrm{l}\mathrm{n}\left[{\displaystyle \frac{\mathrm{r}{\mathrm{d}}_{\mathrm{i}\mathrm{t}}+\mathrm{r}{\mathrm{d}}_{\mathrm{i},\mathrm{t}-1}}{\mathrm{r}{\mathrm{d}}_{\mathrm{i}\mathrm{t}-1}+\mathrm{r}{\mathrm{d}}_{\mathrm{i},\mathrm{t}-2}}}\times \left(\mathrm{r}{\mathrm{d}}_{\mathrm{i}\mathrm{t}}+\mathrm{r}{\mathrm{d}}_{\mathrm{i},\mathrm{t}-1}\right)+1\right]$$

(1)

$$\mathrm{I}\mathrm{n}\mathrm{n}{\mathrm{o}\_\mathrm{o}\mathrm{u}\mathrm{t}}_{\mathrm{i}\mathrm{t}}=\mathrm{l}\mathrm{n}\left[{\displaystyle \frac{{\mathrm{p}\mathrm{t}}_{\mathrm{i}\mathrm{t}}+{\mathrm{p}\mathrm{t}}_{\mathrm{i},\mathrm{t}-1}}{\mathrm{p}{\mathrm{t}}_{\mathrm{i}\mathrm{t}-1}+\mathrm{p}{\mathrm{t}}_{\mathrm{i},\mathrm{t}-2}}}\times \left({\mathrm{p}\mathrm{t}}_{\mathrm{i}\mathrm{t}}+{\mathrm{p}\mathrm{t}}_{\mathrm{i},\mathrm{t}-1}\right)+1\right]$$

(2)

where $\mathrm{r}{\mathrm{d}}_{\mathrm{i}\mathrm{t}}$ denotes current R&D expenditure, while $\mathrm{r}{\mathrm{d}}_{\mathrm{i},\mathrm{t}-1}$ and $\mathrm{r}{\mathrm{d}}_{\mathrm{i},\mathrm{t}-2}$ represent R&D expenditure in the previous one and two years, respectively. Similarly, ${\mathrm{p}\mathrm{t}}_{\mathrm{i}\mathrm{t}}$ denotes the number of current-period patent applications, while ${\mathrm{p}\mathrm{t}}_{\mathrm{i},\mathrm{t}-1}$ and $\mathrm{p}{\mathrm{t}}_{\mathrm{i},\mathrm{t}-2}$ correspond to patent applications in the preceding one and two years.

• Control Variables

Following prior literature, this study includes a set of control variables that may influence persistent innovation. These variables are categorized into three dimensions: (1) Company characteristics: we control for Company Age (Age), measured by the number of years since the establishment; Company Size (Size), measured as the natural logarithm of total assets; and Proportion of Fixed Assets (FIXED), calculated as the ratio of fixed assets to total assets. (2) Financial indicators: we include Book-to-Market Ratio (BM); Leverage (Lev), calculated as the ratio of total liabilities to total assets; Cash Flow Ratio (Cashflow); and Return on Equity (ROE). (3) Internal governance: we control for Ownership Concentration (Shrcr1), measured by the shareholding ratio of the largest shareholder; and CEO Duality (Dual), a dummy variable equal to 1 if the CEO concurrently serves as the board chair, and 0 otherwise. The theoretical rationale for selecting the control variables is as follows:

Company Age (Age): This variable controls for differences in companies’ life cycle stages. Established companies typically accumulate more stable innovation resources and exhibit stronger path dependence, which can give them an advantage in achieving persistent innovation [49].

Company Size (Size): Measured by the logarithm of total assets, this variable accounts for the effect of economies of scale. Larger companies may enjoy advantages in resource acquisition, innovation investment, and platform access capability [50].

Fixed Asset Ratio (FIXED): This variable reflects how the structure of a company’s assets affects its innovation flexibility. Capital-intensive companies with higher fixed asset ratios may face constraints in dynamic adjustment and sustained innovation [51].

Book-to-Market Ratio (BM): This indicator reflects the relationship between a company’s market valuation and its fundamentals. Companies with lower BM ratios—often growth-oriented—are more likely to pursue persistent innovation to capture excess returns [52].

Leverage (Lev): This variable measures the level of financial leverage. Excessive debt may limit a company’s risk tolerance and willingness to invest in R&D, thus hindering its ability to sustain innovation [53].

Cash Flow Ratio (Cashflow): Measured as the ratio of operating cash flow to total assets, this variable controls for internal financing capacity. Companies with abundant cash flow are better positioned to support long-term R&D activities [54].

Return on Equity (ROE): This variable measures the net income generated per unit of equity capital, reflecting a company’s profitability and capital utilization efficiency. On the one hand, ROE is a key indicator of financial performance; on the other hand, prior research shows a significant relationship between R&D investment and ROE [55]. Therefore, controlling for ROE helps avoid potential confounding effects of financial performance on innovation behavior.

Shareholding Ratio of the Largest Shareholder (Shrcr1): This variable captures ownership concentration. A high level of controlling ownership may enhance monitoring efficiency and promote long-term strategic orientation, thereby facilitating persistent innovation [56]. However, it may also lead to a preference for low-risk investments and avoidance of high-failure-rate innovation [57].

CEO Duality (Dual): This variable indicates whether the CEO and the board chair positions are held by the same individual, reflecting the degree of power concentration. While duality may improve governance efficiency, it may also weaken oversight mechanisms, thereby influencing a company’s innovation strategy [58].

Taken together, it is necessary to include the above variables as controls to account for potential confounding effects of company characteristics, financial performance, and corporate governance on innovation behavior.

In addition, year fixed effects and industry fixed effects are included to account for unobservable temporal and sector-specific heterogeneity. Detailed definitions and construction methods for all variables are provided in

Table 2. Variable definitions.

Variable Type	Symbol	Variable Name	Definition
Independent Variable	PEE	Platform Ecosystem Embeddedness	See definition in previous section
Dependent Variables	Inno_in	Persistent Innovation Input	See definition in previous section
	Inno_out	Persistent Innovation Output	See definition in previous section
Control Variables	Age	Company Age	ln (Current Year−Year of Establishment + 1)
	Size	Company Size	Natural logarithm of total assets
	FIXED	Proportion of Fixed Assets	Net fixed assets/Total assets
	BM	Book-to-Market Ratio	Book value/Market value
	Lev	Leverage	Total liabilities at year-end/Total assets at year-end
	Cashflow	Cash Flow Ratio	Net cash flow from operating activities/Total assets
	ROE	Return on Equity	Net income/Equity
	Shrcr1	Ownership Concentration	Number of shares held by the largest shareholder/Total number of shares
	Dual	CEO Duality	Dummy variable: 1 if the Chairman also serves as CEO, 0 otherwise

Source: it is calculated based on the data from the CSMAR database.

.

3.3. Model Specification

To empirically examine the effect of platform ecosystem embeddedness on the level of persistent innovation in manufacturing companies, we construct the following multivariate regression models, as shown in Equations (3) and (4):

$${Inno\_in}_{i,t}={\alpha }_0+{\alpha }_1{PEE}_{i,t-1}+{\alpha }_2{Controls}_{i,t}+{\lambda }_t{+\delta }_j+{\epsilon }_{i,t}$$

(3)

$${Inno\_out}_{i,t}={\beta }_0+{\beta }_1{PEE}_{i,t-1}+{\beta }_2{Controls}_{i,t}+{\lambda }_t{+\delta }_j+{\epsilon }_{i,t}$$

(4)

In the above equations, ${\mathrm{P}\mathrm{E}\mathrm{E}}_{\mathrm{i},\mathrm{t}-1}$ denotes the platform ecosystem embeddedness of company i in year t − 1; ${\mathrm{I}\mathrm{n}\mathrm{n}\mathrm{o}\_\mathrm{i}\mathrm{n}}_{\mathrm{i},\mathrm{t}}$ and ${\mathrm{I}\mathrm{n}\mathrm{n}\mathrm{o}\_\mathrm{o}\mathrm{u}\mathrm{t}}_{\mathrm{i},\mathrm{t}}$ denote the persistent innovation input and persistent innovation output of company i in year t, respectively. ${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}\mathrm{s}}_{\mathrm{i},\mathrm{t}}$ denotes the set of control variables defined in Table 2. ${\mathsf{\lambda }}_{\mathrm{t}}$ and ${\mathsf{\delta }}_{\mathrm{j}}$ represent industry and year fixed effects, respectively; ${\mathsf{\epsilon }}_{\mathrm{i},\mathrm{t}}$ is the error term.

4. Empirical Analysis

4.1. Descriptive Statistics

The descriptive statistics of the key variables used in this study are presented in

Table 3. Descriptive statistics.

Variable	Obs	Mean	Std. Dev.	Min	Max
Inno_in	9279	19.572	1.346	12.690	25.488
Inno_out	9279	3.202	1.379	0.077	9.460
l.PEE	9279	1.646	1.401	0.000	6.290
Age	9279	3.002	0.274	1.792	4.220
Size	9279	22.499	1.250	19.857	28.636
FIXED	9279	0.206	0.133	0.001	0.876
BM	9279	0.592	0.250	0.044	1.559
Lev	9279	0.417	0.184	0.014	0.979
Cashflow	9279	0.056	0.065	−0.313	0.839
ROE	9279	0.067	0.147	−4.320	1.442
Shrcr1	9279	31.894	14.206	3.003	86.347
Dual	9279	0.294	0.456	0.000	1.000

Source: it is calculated based on the data from the CNINFO, CNRDS database, and CSMAR database.

. Regarding persistent innovation input (Inno_in), the mean value is 19.572 with a standard deviation of 1.346. The values range from 12.690 to 25.488, indicating that companies, on average, maintain a relatively high level of innovation input, albeit with notable variation. This distribution suggests heterogeneity in companies’ allocation of innovation resources, potentially influenced by factors such as company size, capital capacity, or access to external resources.

For persistent innovation output (Inno_out), the mean is 3.202, with a standard deviation of 1.379. The minimum and maximum values are 0.077 and 9.460, respectively, reflecting significant heterogeneity in innovation output. Some companies exhibit limited output despite sustained input, possibly indicating low innovation efficiency. In contrast, other companies demonstrate strong capabilities in knowledge conversion, suggesting structural differences in the efficiency of innovation resource utilization.

The core explanatory variable, Lagged platform ecosystem embeddedness (l.PEE), has a mean of 1.646 and a standard deviation of 1.401, with values ranging from 0 to 6.290. The distribution is right-skewed, reflecting considerable variation in companies’ level of embeddedness within digital platforms. Most companies remain at a relatively low level of engagement, while only a few display strong platform participation or digital ecosystem integration capabilities. This variance provides both theoretical and empirical justification for examining the potential impact of platform ecosystem embeddedness on companies’ innovation behavior.

4.2. Multicollinearity Test

To examine the potential issue of multicollinearity in the regression model, we conducted a Variance Inflation Factor (VIF) test on both the key explanatory variable and the control variables (see

Table 4. Multicollinearity test results.

Variable	VIF	1/VIF
l.PEE	1.180	0.846
Age	1.090	0.921
Size	1.820	0.550
FIXED	1.230	0.811
BM	1.530	0.652
Lev	1.540	0.649
Cashflow	1.310	0.761
ROE	1.310	0.766
Shrcr1	1.070	0.936
Dual	1.040	0.966

Source: it is calculated based on the data from the CNINFO, CNRDS database, and CSMAR database.

). The results indicate that all VIF values are well below the commonly accepted threshold of 10. The highest VIF observed is 1.820 for company size (Size), while the remaining variables fall between 1.0 and 1.5. In addition, the corresponding 1/VIF values are generally greater than 0.5, suggesting weak correlations among the variables. Overall, the model does not suffer from serious multicollinearity, and the regression estimates are stable and reliable.

4.3. Baseline Regression

Table 5. Baseline regression.

	(1)	(2)	(3)	(4)	(5)	(6)
	Inno_in	Inno_in	Inno_in	Inno_out	Inno_out	Inno_out
l.PEE	0.185 ***	0.094 ***	0.038 ***	0.168 ***	0.125 ***	0.105 ***
	(18.92)	(14.93)	(5.81)	(16.72)	(12.62)	(8.83)
Age		0.064 **	–0.067 **		–0.182 ***	0.064
		(2.08)	–2.36)		(–3.76)	(1.25)
Size		0.920 ***	0.992 ***		0.504 ***	0.561 ***
		(104.59)	(124.63)		(36.57)	(38.93)
FIXED		–0.804 ***	–0.328 ***		–0.366 ***	–0.282 **
		(–11.82)	(–4.92)		(–3.44)	(–2.34)
BM		–0.593 ***	–0.668 ***		–0.568 ***	–0.658 ***
		(–14.67)	(–17.58)		(–8.95)	(–9.58)
Lev		–0.082	–0.115 **		0.113	–0.050
		(–1.49)	(–2.40)		(1.31)	(–0.58)
Cashflow		0.305 **	0.380 ***		–0.346	–0.206
		(2.12)	(3.13)		(–1.53)	(–0.94)
ROE		0.184 ***	0.212 ***		0.254 **	0.223 **
		(2.90)	(3.97)		(2.55)	(2.32)
Shrcr1		–0.005 ***	–0.001 **		–0.000	0.001
		(–9.14)	(–2.32)		(–0.07)	(0.65)
Dual		0.071 ***	0.004		0.056 *	0.054 *
		(3.88)	(0.29)		(1.95)	(1.96)
_cons	19.267 ***	–0.802 ***	–2.095 ***	2.925 ***	–7.450 ***	–9.343 ***
	(910.57)	(–4.21)	(–11.83)	(134.44)	(–24.97)	(–29.18)
Year	NO	NO	YES	NO	NO	YES
Industry	NO	NO	YES	NO	NO	YES
N	9279	9279	9276	9279	9279	9276
R2	0.037	0.660	0.772	0.029	0.203	0.290
Adj. R2	0.037	0.660	0.770	0.029	0.203	0.284
F	358.140	1798.833	2420.434	279.601	236.592	247.883

Note: *** p < 0.01, ** p < 0.05, * p < 0.1. Source: it is calculated based on the data from the CNINFO, CNRDS database, and CSMAR database.

reports the fixed effects regression results of lagged platform ecosystem embeddedness (l.PEE) on companies’ persistent innovation input (Inno_in) and persistent innovation output (Inno_out). Columns (1) and (4) present the baseline models without any control variables. Columns (2) and (5) include the control variables listed in Table 2 but do not control for industry and year fixed effects. Columns (3) and (6) report the full models, which incorporate all control variables as well as industry and year fixed effects.

Across all model specifications, platform ecosystem embeddedness demonstrates a significantly positive effect on both innovation input and output. The results suggest that embedding in platform ecosystems significantly enhances companies’ persistent innovation activities. This positive effect remains robust under various model configurations and control settings, thereby supporting the hypothesis that platform ecosystem embeddedness promotes persistent innovation. Hypothesis H1 is supported.

4.4. Endogeneity Test

Given potential endogeneity issues caused by omitted variables or measurement errors due to sample selection limitations, we employ the instrumental variables method (IV) and the propensity score matching method (PSM) to validate the model. The results are presented in

Table 6. Endogeneity test.

	(1)	(2)	(3)	(4)	(5)
	l.PEE	Inno_in	Inno_out	Inno_in	Inno_out
Avg_PEE_City	0.506 ***
	(30.71)
l.PEE		0.222 ***	0.109 ***	0.0395 ***	0.114 ***
		(9.95)	(2.82)	(5.02)	(7.47)
Controls	YES	YES	YES	YES	YES
Year	NO	NO	NO	YES	YES
Industry	NO	NO	NO	YES	YES
N	9279	9279	9279	6196	5755
R2	0.517	0.753	0.290	0.783	0.324
Adj. R2	0.513	0.745	0.274	0.781	0.315
F	124.95	364.88	46.67	1703.000	177.000

Note: *** p < 0.01. Source: it is calculated based on the data from the CNINFO, CNRDS database, and CSMAR database.

.

First, for the instrumental variables test, following the approach of Mao et al. [59], we use the annual city-level average of platform ecosystem embeddedness (Avg_PEE_City) as the instrumental variable. This variable captures the overall level of platform ecosystem embeddedness in a given city, while remaining unrelated to the persistent innovation performance of individual companies, thereby satisfying both the relevance and exogeneity conditions of a valid instrument. As shown in columns (1) through (3) of Table 6, the two-stage regression results indicate that the positive effect of platform ecosystem embeddedness on persistent innovation remains empirically supported after addressing endogeneity bias.

Second, for the propensity score matching method, we divide the sample companies into two groups: those with platform ecosystem embeddedness (l.PEE) above the sample mean are assigned a value of 1 for the variable type, and those below the mean are assigned a value of 0. We then estimate the propensity scores based on all control variables using a logit model. After obtaining the propensity scores for all companies in the sample, we apply a 1:5 nearest-neighbor matching approach to minimize the mean squared error. This yields a matched sample of 8180 companies, including 3638 in the treatment group and 4542 in the control group. Finally, we rerun the regression using the matched sample. The results, presented in columns (4) and (5) of Table 6, further concompany the significant positive impact of platform ecosystem embeddedness on persistent innovation.

4.5. Robustness Test

To ensure the robustness of our findings, we conduct a series of tests by adjusting variable measurement, sample scope, and fixed effects. First, we test the robustness by replacing the key independent variable with alternative measures. The results are presented in

Table 7. Robustness test using alternative measures of the independent variable.

	(1)	(2)	(3)	(4)	(5)	(6)
	Inno_in	Inno_out	Inno_in	Inno_out	Inno_in	Inno_out
l.Platform	0.090 ***	0.155 ***
	(5.50)	(5.24)
l.Plat_number			0.118 ***	0.075 ***
			(12.00)	(4.20)
l.Plat_trade					0.106 ***	0.041 ***
					(12.45)	(2.65)
Controls	YES	YES	YES	YES	YES	YES
N	9276	9276	7861	7861	7861	7861
R2	0.772	0.286	0.771	0.285	0.771	0.284
Adj. R2	0.770	0.280	0.769	0.278	0.769	0.277
F	2419.163	241.522	2011.963	203.891	2015.855	202.552

Note: *** p < 0.01. Source: it is calculated based on the data from the CNINFO, CNRDS database, and CSMAR database.

.

The first alternative measure uses a binary variable indicating whether a company has joined a platform (l.Platform). If the annual report of the company in the previous year contains keywords such as “internet platform,” “digital platform,” “data platform,” “artificial intelligence platform,” or “industrial internet platform,” the company is considered to have engaged with a platform, and the variable is coded as 1; otherwise, it is coded as 0. The results are reported in columns (1) and (2) of Table 7.

The second alternative follows the approach of Feng and Wu [60], in which we proxy platform ecosystem embeddedness using regional platform economy development indicators. Specifically, we use the lagged number of companies involved in e-commerce activities (l.Plat_number) and the lagged regional e-commerce transaction volume (l.Plat_trade) to represent the level of local platform economy development. Both variables are log-transformed. The number of e-commerce companies reflects platform development at the company level, while transaction volume captures the scale of platform economic activity. As shown in columns (3) to (6) of Table 7, the results remain significantly positive, concompanying the robustness of our findings even when alternative measures of platform ecosystem embeddedness are used.

Additional robustness test results are reported in

Table 8. Additional robustness regression results.

	(1)	(2)	(3)	(4)	(5)
	persistence	Inno_in	Inno_out	Inno_in	Inno_out
l.PEE	0.820 ***	0.037 ***	0.097 ***	0.033 ***	0.110 ***
	(37.07)	(3.55)	(5.42)	(2.92)	(6.28)
Controls	YES	YES	YES	−0.025	0.026
Year	YES	YES	YES	YES	YES
Year	YES	YES	YES	YES	YES
N	9276	4317	4317	9248	9248
R2	0.592	0.741	0.294	0.818	0.376
Adj. R2	0.588	0.737	0.283	0.811	0.352
F	266.066	893.567	112.300	576.575	64.104

Note: *** p < 0.01. Source: it is calculated based on the data from the CNINFO, CNRDS database, and CSMAR database.

.

We first replace the dependent variable. Specifically, we use the persistence of management’s digital innovation orientation (persistence) as an alternative measure of the company’s persistent innovation. The result is presented in column (1) of Table 8.

Next, we narrow the sample period. Considering the substantial disruption caused by the COVID-19 pandemic to production and work routines of Chinese companies between 2020 and 2022, we restrict the sample to the years 2012–2019. The results of this subsample analysis are shown in columns (2) and (3).

We also adopt higher-order fixed effects. To address potential correlations across cities and industries, and to mitigate heteroscedasticity concerns, we include triple fixed effects at the city–industry–year level. In addition, standard errors are clustered at the city level to enhance the robustness and reliability of the estimates. This specification helps isolate the influence of complex macroeconomic and spatial factors on persistent innovation. The results are reported in columns (4) and (5).

Overall, the robustness tests concompany the validity of the baseline results. Platform ecosystem embeddedness remains a significant and positive driver of persistent innovation across multiple model specifications.

4.6. Heterogeneity Analysis

4.6.1. External Environment of Companies

Table 9. Heterogeneity analysis: external environment of companies.

	Inno_in		Inno_out		Inno_in			Inno_out
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	H_C	L_C	H_C	L_C	E	C	W	E	C	W
l.PEE	0.070 ***	0.014	0.120 ***	0.088 ***	0.043 ***	0.039 **	0.030	0.135 ***	0.031	0.073 *
	(7.22)	(1.55)	(6.81)	(5.47)	(6.63)	(2.01)	(0.90)	(10.18)	(0.94)	(1.77)
Controls	YES	YES	YES	YES	YES	YES	YES	YES	YES	YES
Year	YES	YES	YES	YES	YES	YES	YES	YES	YES	YES
Industry	YES	YES	YES	YES	YES	YES	YES	YES	YES	YES
N	4634	4641	4634	4641	6908	1419	940	6908	1419	940
R2	0.785	0.768	0.312	0.272	0.823	0.751	0.673	0.308	0.372	0.384
Adj. R2	0.782	0.765	0.302	0.262	0.822	0.741	0.654	0.300	0.346	0.349
F	1247.854	1218.056	144.295	104.857	2463.195	258.247	111.002	209.024	32.683	25.230

Note: *** p < 0.01, ** p < 0.05, * p < 0.1. H_C represents High competition, L_C represents Low competition, E represents the Eastern region, C represents the Central region, and W represents the Western region. Source: it is calculated based on the data from the CNINFO, CNRDS database, and CSMAR database.

presents the heterogeneity analysis results of how platform ecosystem embeddedness influences companies’ persistent innovation under different external environmental conditions. Specifically, we conduct subgroup analyses based on two external dimensions: industry competition and geographic location.

First, we use the industry Lerner index to capture the degree of industry competition. The sample is split into two groups based on the mean value of the index. Companies in industries with a Lerner index above the sample mean are classified as operating in highly competitive industries. As shown in Columns (1) to (4), platform ecosystem embeddedness exerts a significantly positive effect on both persistent innovation input and output in highly competitive industries. This suggests that under intense competitive pressure, companies are more likely to leverage platform ecosystem resources to enhance their digital coordination capabilities and responsiveness to innovation, thereby achieving differentiated competitive advantages.

In contrast, in industries with lower levels of competition, the effect of platform ecosystem embeddedness on innovation input is not statistically significant, while its effect on innovation output remains positively significant. This indicates that even in less competitive environments, platform resources can still facilitate the generation of innovation outcomes. However, the incentive effect on resource allocation and R&D investment may be relatively limited in these settings.

Second, we examine geographic heterogeneity by grouping the sample according to companies’ locations in eastern, central, and western regions of China. As reported in Columns (5) to (10), platform ecosystem embeddedness has a significantly positive effect on both persistent innovation input and output in the eastern region, suggesting that in areas with more advanced digital infrastructure and greater platform resource concentration, companies are better positioned to embed within platform ecosystems. This allows them to benefit substantially from resource acquisition, collaborative innovation, and technology diffusion, thereby enhancing both the stability of innovation input and the efficiency of innovation output.

In the central region, platform ecosystem embeddedness shows a significantly positive effect on innovation input, but its impact on innovation output is not significant. This implies that although companies in the central region have developed a certain degree of platform participation, they still face constraints in efficiently converting innovation investment into tangible outcomes, possibly due to limitations in transformation efficiency and weaker technological collaboration.

In comparison, in the western region, platform ecosystem embeddedness does not significantly influence innovation input, but has a marginally significant positive effect on innovation output. This may reflect the fact that although digital resource accessibility and platform integration depth remain limited in the western region, platforms may still contribute to the introduction of external knowledge or provide output-oriented support, which can, under specific conditions, stimulate the generation of innovation outcomes.

4.6.2. Internal Company Characteristics

Table 10. Heterogeneity analysis: internal company characteristics.

	Inno_in		Inno_out		Inno_in		Inno_out
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	H_T	N_H_T	H_T	N_H_T	H_P	N_H_P	H_P	N_H_P
l.PEE	0.043 ***	0.014	0.113 ***	0.071 ***	–0.015	0.049 ***	0.101 ***	0.109 ***
	(6.65)	(0.77)	(8.35)	(2.90)	(–0.77)	(7.37)	(3.72)	(8.23)
Controls	YES	YES	YES	YES	YES	YES	YES	YES
Year	YES	YES	YES	YES	YES	YES	YES	YES
Industry	YES	YES	YES	YES	YES	YES	YES	YES
N	6874	2402	6874	2402	2013	7263	2013	7263
R2	0.809	0.713	0.290	0.323	0.716	0.796	0.340	0.281
Adj. R2	0.808	0.705	0.287	0.306	0.712	0.795	0.329	0.275
F	2529.093	319.838	221.752	40.464	281.165	2362.744	39.129	213.580

Note: *** p < 0.01. H_T represents High Technology, N_H_T represents Non-High Technology, H_P represents Heavy Pollution, and N_H_P represents Non-Heavy Pollution. Source: it is calculated based on the data from the CNINFO, CNRDS database, and CSMAR database.

presents the heterogeneity analysis results of the impact of platform ecosystem embeddedness on persistent innovation under different internal company characteristics. Specifically, we conduct subgroup analyses based on two dimensions: technological intensity and environmental compliance burden.

First, following the classification criteria in Yang and Zhou [61], we divide the sample into high-tech companies and non-high-tech companies. As shown in Columns (1)–(4), platform ecosystem embeddedness exhibits a significantly positive effect on both persistent innovation input and output among high-tech companies. This suggests that high-tech companies possess greater digital sensitivity and absorptive capacity, enabling them to more effectively leverage knowledge externalities and technological spillovers from platform ecosystems to enhance innovation activities.

By contrast, for non-high-tech companies, while platform ecosystem embeddedness continues to have a significantly positive impact on innovation output, its effect on innovation input is not statistically significant. This indicates that the innovation incentive effect of platform ecosystems may be more limited for non-high-tech companies, potentially due to weaker internal capabilities or insufficient motivation for digital transformation.

Second, in line with the classification of Pan et al. [62], we group companies into heavily polluting and non-heavily polluting categories. As reported in Columns (5)–(8), platform ecosystem embeddedness does not have a significant impact on innovation input among heavily polluting companies. These companies face stricter environmental regulations and often divert substantial resources toward pollution treatment and compliance, potentially crowding out resources for innovation investment. Moreover, the technological paths in high-pollution industries are typically more rigid, and companies may have lower demand for innovation enablement via platform ecosystems, thereby weakening the embedding effect on innovation input.

Nonetheless, innovation output in heavily polluting companies is still positively influenced by platform embedding. This may be attributed to process optimizations supported by platform ecosystems, such as intelligent management systems that reduce energy consumption and improve resource utilization efficiency, which indirectly contribute to increased innovation output.

In contrast, non-heavily polluting companies show significantly positive effects of platform ecosystem embeddedness on both innovation input and output. These companies face relatively fewer environmental constraints and can allocate more platform-enabled resources toward technological R&D and innovation activities. Moreover, the faster pace of technological change in these industries increases companies’ demand for platform-enabled knowledge integration and technology sharing, thereby enhancing innovation input. With higher innovation efficiency, these companies are also better positioned to convert input into outcomes, and the technological synergies offered by platforms directly contribute to improved innovation output.

5. Further Analysis

5.1. Mechanism Analysis of Persistent Innovation Input

5.1.1. Supply–Demand Coordination Costs

Supply–demand coordination costs in supply chain collaboration are difficult to measure directly. Therefore, this study follows the approach of Cachon et al. [63] and adopts a deviation-based approach to quantify the precision of supply–demand matching in the supply chain. Specifically, we construct a Supply–Demand Deviation (DSDD) index to capture companies’ supply–demand coordination costs, which is defined as follows:

$$\mathrm{D}\mathrm{S}\mathrm{D}{\mathrm{D}}_{\mathrm{i}\mathrm{t}}={\displaystyle \frac{\mathsf{\sigma }\left(\mathrm{P}\mathrm{r}\mathrm{o}\mathrm{d}\mathrm{u}\mathrm{c}\mathrm{t}\mathrm{i}\mathrm{o}{\mathrm{n}}_{\mathrm{i}\mathrm{t}}\right)}{\mathsf{\sigma }\left(\mathrm{D}\mathrm{e}\mathrm{m}\mathrm{a}\mathrm{n}{\mathrm{d}}_{\mathrm{i}\mathrm{t}}\right)}}-1$$

(5)

$$\begin{array}{c}{\mathrm{Production}}_{\mathrm{it}}={\mathrm{Cost}}_{\mathrm{it}}+{\mathrm{Inv}}_{\mathrm{it}}-{\mathrm{Inv}}_{\mathrm{it}-1}\end{array}$$

(6)

Here, $\mathsf{\sigma }(·)$ denotes the standard deviation of the corresponding variable. The numerator and denominator measure the volatility of company production and demand, respectively. Company production ($\mathrm{P}\mathrm{r}\mathrm{o}\mathrm{d}\mathrm{u}\mathrm{c}\mathrm{t}\mathrm{i}\mathrm{o}{\mathrm{n}}_{\mathrm{i}\mathrm{t}}$) is calculated based on Equation (6), with $\mathrm{C}\mathrm{o}\mathrm{s}{\mathrm{t}}_{\mathrm{i}\mathrm{t}}$ representing the company’s operating costs and $\mathrm{I}\mathrm{n}{\mathrm{v}}_{\mathrm{i}\mathrm{t}}$ referring to the year-end net value of inventories. Demand ($\mathrm{D}\mathrm{e}\mathrm{m}\mathrm{a}\mathrm{n}{\mathrm{d}}_{\mathrm{i}\mathrm{t}}$) is proxied by the company’s operating costs. A higher DSDD indicates a greater mismatch between production and demand, implying lower supply–demand matching accuracy and, consequently, higher supply–demand coordination costs.

As this is a company-level indicator, to reduce endogeneity concerns, we use the first lag of DSDD as the proxy for supply–demand coordination costs. Following the mediation analysis procedures used in prior studies [64,65], we test whether this mechanism mediates the relationship between platform ecosystem embeddedness and persistent innovation input.

The regression results are presented in

Table 11. Mechanism test: supply–demand coordination costs.

	(1)	(2)	(3)	(4)
	Inno_in	l.DSDD	Inno_in	Inno_in
l2.PEE	0.037 ***	–0.004 ***		0.037 ***
	(5.09)	(–2.73)		(5.03)
l.DSDD			–0.180 ***	–0.175 ***
			(–3.57)	(–3.16)
Controls	YES	YES	YES	YES
Year	YES	YES	YES	YES
Industry	YES	YES	YES	YES
N	7152	7120	9269	7120
R2	0.785	0.074	0.772	0.785
Adj. R2	0.782	0.064	0.770	0.783
F	2005.210	5.393	2406.956	1815.627

Note: *** p < 0.01. Source: it is calculated based on the data from the CNINFO, CNRDS database, and CSMAR database. The calculation of DSDD refers to the approach of Cachon et al. [63].

. Platform ecosystem embeddedness is found to significantly enhance companies’ persistent innovation input, with supply–demand coordination costs playing a partial mediating role. This provides empirical support for Hypothesis H2a. The second lag of platform embedding (l2.PEE) has a significantly positive effect on persistent innovation input (Inno_in), indicating that digital platform engagement directly promotes sustained R&D investment and outcome realization.

Further analysis shows that platform ecosystem embeddedness significantly reduces supply–demand deviation (l.DSDD), suggesting that embedded companies can better mitigate mismatches in their supply chains and lower their coordination costs. Moreover, when DSDD is included in the regression model, it exhibits a significantly negative effect on innovation input—indicating that smoother supply–demand coordination enhances companies’ ability to maintain sustained investment in innovation.

These findings suggest that platform ecosystem embeddedness lowers frictional losses in supply–demand matching, thus providing both the resource base and operational stability needed for persistent R&D investment. In traditional supply chains, companies often face challenges such as information asymmetry, lack of technological interoperability, and delays in resource allocation, leading to imbalanced production–demand relationships, resource waste, and shortages that hinder the long-term allocation of innovation resources.

By contrast, platform ecosystems enable companies to establish standardized data interfaces, adopt uniform coordination protocols, and participate in multilateral trading mechanisms. These systems improve companies’ sensitivity to market demand changes and responsiveness on the supply side, enhancing the synchronicity of supply–demand fluctuations and reducing coordination costs arising from mismatches.

From the perspective of resource synergy, platform ecosystem embeddedness enables the integration of data flows, technology flows, and logistics, forming a collaborative mechanism for resource allocation. Embedded companies can use shared data to monitor real-time changes in upstream and downstream dynamics, predict demand more accurately, and quickly dispatch resources. This reduces production volatility caused by forecasting errors or coordination delays. As a result, companies can lower their need for redundant buffer resources under uncertainty, freeing up capacity to be stably allocated to R&D activities, thereby enhancing the sustainability of innovation input.

Furthermore, platform-enabled recommendation algorithms and coordinated production scheduling systems improve both the efficiency and precision of resource allocation. These digital infrastructures offer structural support for companies to build “low-friction, high-responsiveness” innovation input systems.

5.1.2. Operational Efficiency

To further examine the mechanism through which platform ecosystem embeddedness promotes companies’ persistent innovation input, this study follows the approach of Li et al. [66] and adopts total asset turnover (asset_turn) as a proxy for operational efficiency. A mediation model is constructed to test whether platform embedding indirectly facilitates sustained R&D investment by improving companies’ operational efficiency. Operational efficiency reflects a company’s ability to convert assets into sales revenue. Higher efficiency implies more effective internal resource utilization, thereby providing greater room for innovation investment.

To mitigate potential endogeneity concerns, the one-period lag of operational efficiency (l.asset_turn) is used as the mediator, while the core explanatory variable, platform ecosystem embeddedness, is lagged by two periods (l2.PEE).

Regression results are reported in

Table 12. Mechanism test: operational efficiency.

	(1)	(2)	(3)	(4)
	Inno_in	l.asset_turn	Inno_in	Inno_in
l2.PEE	0.037 ***	0.012 ***		0.031 ***
	(5.09)	(4.07)		(4.29)
l.asset_turn			0.620 ***	0.589 ***
			(24.36)	(20.59)
Controls	YES	YES	YES	YES
Year	YES	YES	YES	YES
Industry	YES	YES	YES	YES
N	7152	7125	9276	7125
R2	0.785	0.386	0.785	0.797
Adj. R2	0.782	0.380	0.784	0.795
F	2005.210	107.741	2622.973	1965.149

Note: *** p < 0.01. Source: it is calculated based on the data from the CNINFO, CNRDS database, and CSMAR database.

. The results show that platform ecosystem embeddedness significantly enhances companies’ persistent innovation input, and that operational efficiency plays a partial mediating role in this relationship, thereby providing empirical support for Hypothesis H2b. Further analysis reveals that platform embedding significantly improves operational efficiency, suggesting that companies optimize their business processes by leveraging the resource allocation, process coordination, and information integration capabilities embedded within platform ecosystems. At the same time, operational efficiency positively and significantly affects innovation input, indicating that more efficient companies are better positioned to invest in continuous innovation while maintaining stable operations.

From the perspective of resource synergy, platform embedding helps companies overcome traditional organizational boundaries and capability constraints by leveraging shared resources and services offered by the platform to optimize business processes and enhance efficiency. Through data middleware, algorithmic support, and collaborative operational mechanisms, platforms enable end-to-end resource reconfiguration across procurement, production, and sales. This improves asset utilization efficiency and reduces excessive inventory and redundant resource usage.

An increase in asset turnover not only signals improved operational efficiency, but also indicates that each unit of resource is participating in the value creation process more frequently. This releases a greater volume of reconfigurable resources, thereby enhancing the company’s capacity for sustained R&D investment.

In addition, service modules and operational tools embedded in the platform ecosystem, such as cloud computing, intelligent scheduling, and digital marketing, enable companies to shift parts of capital-intensive business activities toward platform-based and service-oriented configurations. This reduces fixed cost burdens and improves the flexibility of resource use. Such a “low-redundancy, high-utilization” model empowers companies to allocate resources toward innovation more consistently while maintaining operational stability.

The efficiency dividends generated by platform ecosystems create a positive internal cycle of “operational savings → resource reallocation → reinvestment in innovation,” thereby providing both institutional and capability foundations for sustained innovation input.

5.2. Mechanism Analysis of Persistent Innovation Output

5.2.1. Knowledge Flow

To further uncover the mechanism through which platform ecosystem embeddedness enhances companies’ persistent innovation output, this study introduces knowledge flow efficiency (Kflow) as a mediating variable, following the research framework proposed by Fung and Chow (2002) [67]. We construct a mediation model to investigate whether platform ecosystem embeddedness indirectly promotes persistent innovation output by improving companies’ knowledge flow efficiency.

Specifically, knowledge flow efficiency is measured as the natural logarithm of one plus the average number of non-self citations per patent application in the following year, which reflects the company’s knowledge diffusion capability within the technological network. To mitigate potential endogeneity concerns, the first lag of knowledge flow efficiency (l.Kflow) is used as the mediator, and platform ecosystem embeddedness is measured with a two-period lag (l2.PEE).

The regression results are presented in

Table 13. Mechanism test: knowledge flow.

	(5)	(6)	(7)	(8)
	Inno_out	l.Kflow	Inno_out	Inno_out
l2.PEE	0.098 ***	0.008 ***		0.094 ***
	(7.16)	(4.73)		(6.86)
l.Kflow			0.415 ***	0.409 ***
			(5.26)	(4.24)
Controls	YES	YES	YES	YES
Year	YES	YES	YES	YES
Industry	YES	YES	YES	YES
N	7152	7122	9251	7122
R2	0.308	0.709	0.286	0.310
Adj. R2	0.301	0.706	0.280	0.302
F	206.080	22.940	240.211	188.062

Note: *** p < 0.01. Source: it is calculated based on the data from the CNINFO, CNRDS database, and CSMAR database.

. We find that platform ecosystem embeddedness significantly promotes persistent innovation output, and this effect is partially mediated by knowledge flow. Without controlling for the mediator, platform ecosystem embeddedness has a significantly positive effect on innovation output, concompanying its role in enhancing companies’ innovation continuity. When the mediator is introduced, platform ecosystem embeddedness continues to exhibit a significantly positive effect on knowledge flow, and knowledge flow itself positively and significantly affects persistent innovation output.

In the full mediation model—where both platform embedding and knowledge flow are included as explanatory variables—the coefficient of platform ecosystem embeddedness slightly decreases from 0.098 to 0.094, while both coefficients remain statistically significant. This result indicates that knowledge flow plays a partial mediating role in the relationship, thereby providing empirical support for Hypothesis H3a.

From the perspective of knowledge synergy, platform ecosystem embeddedness enhances the company’s structural position and connectivity in the knowledge network, facilitating greater fluidity and frequency of knowledge interaction. This, in turn, supports knowledge recombination and iterative application along existing innovation trajectories, thereby improving the continuity of innovation output. Through standardized interfaces, platforms reduce the costs of inter-organizational knowledge transfer, enabling the codification and dissemination of tacit technical experience and methods. Additionally, multilateral network structures and algorithmic matching mechanisms improve the efficiency of knowledge matching and increase companies’ capacity for knowledge absorption and diffusion.

This synergistic environment not only reinforces organizational knowledge accumulation, but also accelerates feedback loops and reuse of innovation outcomes, helping companies maintain innovation momentum in dynamic environments. Overall, by enhancing knowledge flow efficiency, platform ecosystem embeddedness significantly strengthens companies’ capacity for persistent innovation output, validating the critical mediating role of knowledge flow in the platform-enabled innovation process.

5.2.2. Knowledge Creation

To further explore the mechanism by which platform ecosystem embeddedness facilitates companies’ persistent innovation output, this study follows the approach of Aghion et al. [68] and Akcigit et al. [69] by introducing patent knowledge breadth (Kcreat) as a proxy for knowledge creation. A mediation model is constructed to test whether platform embedding indirectly promotes persistent innovation by expanding the company’s foundational knowledge base.

Patent knowledge breadth measures the complexity and diversity of technological knowledge embedded in a company’s patent portfolio. A broader spread across multiple major IPC categories indicates deeper and more diverse knowledge creation. To improve causal identification, the model uses platform ecosystem embeddedness lagged by two periods as the independent variable (l2.PEE), and the one-period lag of knowledge breadth (l.Kcreat) as the mediator.

The regression results are presented in

Table 14. Mechanism test: knowledge creation.

	(5)	(6)	(7)	(8)
	Inno_out	l.Kcreat	Inno_out	Inno_out
l2.PEE	0.098 ***	0.007 ***		0.087 ***
	(7.16)	(6.68)		(6.39)
l.Kcreat			1.497 ***	1.384 ***
			(12.41)	(9.32)
Controls	YES	YES	YES	YES
Year	YES	YES	YES	YES
Industry	YES	YES	YES	YES
N	7152	7125	9276	7125
R2	0.308	0.145	0.295	0.316
Adj. R2	0.301	0.136	0.290	0.309
F	206.080	63.168	257.470	196.207

Note: *** p < 0.01. Source: it is calculated based on the data from the CNINFO, CNRDS database, and CSMAR database.

. Platform ecosystem embeddedness has a significantly positive impact on persistent innovation output, and this effect is partially mediated by knowledge creation. In the baseline model without the mediator, platform embedding shows a significantly positive effect on innovation output, indicating that platform engagement helps companies maintain continuity in innovation performance over time.

When knowledge creation is introduced as the dependent variable, platform ecosystem embeddedness continues to show a significantly positive coefficient, suggesting that digital platform participation enhances the company’s capacity to generate diverse and complex knowledge. In the full mediation model, where both platform embedding and knowledge creation are included as explanatory variables, both coefficients remain significantly positive, while the coefficient for platform embedding slightly decreases relative to the baseline model. These results indicate that knowledge creation partially mediates the relationship between platform ecosystem embeddedness and persistent innovation output, providing empirical support for Hypothesis H3b.

Platform ecosystems offer embedded companies access to abundant external knowledge resources and technological modules. The modular nature of platform architecture enables flexible recombination of heterogeneous resources, fostering combinatorial innovation and expanding the boundaries of knowledge exploration and technological recombination. Additionally, the involvement of diverse participants and developer communities in platform ecosystems provides fertile ground for acquiring tacit knowledge and experiential diversity across industries and domains. This helps companies break through existing technological path dependencies and enhance the complexity and breadth of their knowledge base, driving substantial upgrades in their knowledge creation capacity.

From the perspective of knowledge synergy, platform embedding facilitates efficient recombination of knowledge elements across organizational boundaries. Companies can collaborate with various ecosystem participants to conduct joint R&D and application-driven innovation, fostering knowledge fusion and the generation of new knowledge. Compared to closed innovation systems, the platform model offers not only a broader range of knowledge access channels, but also enables higher interoperability and reconfigurability of knowledge components through standardized protocols, technical interfaces, and platform governance rules. These features lay a strong knowledge foundation for achieving persistent innovation output.

Therefore, by enhancing the company’s knowledge creation capability, platform ecosystem embeddedness plays a deep and sustained enabling role in long-term innovation performance, highlighting the structural contribution of platform ecosystems to output-side innovation persistence.

6. Discussion

In the existing literature, substantial attention has been paid to the role of platform ecosystems in promoting company innovation performance. Prior studies have primarily focused on how platform structures stimulate breakthrough innovation through functional mechanisms [4,5], or explored the causal relationship between platforms and innovation output from the perspective of technological modules and interactions with complementary products [23,34]. However, there remains a lack of in-depth exploration into how platform ecosystems continuously incentivize companies to engage in stable technological accumulation and iterative innovation over time—particularly in terms of generating sustained effects on both innovation input and innovation output.

This study extends the current research in three key ways:

First, it introduces the concept of persistent innovation into the platform ecosystem discourse, moving beyond the traditional emphasis on one-off breakthroughs or short-term outcomes. By foregrounding innovation continuity as a critical dimension of platform empowerment, this study provides a more comprehensive understanding of long-term innovation dynamics.

Second, it constructs a theoretical framework of “embedding–synergy–persistent innovation”, identifying a dual mechanism through which platform ecosystem embeddedness operates: by promoting persistent innovation input via resource synergy, and persistent innovation output via knowledge synergy. This enriches the meso-level process logic of how platform ecosystems empower innovation.

Third, through heterogeneity analysis, this study reveals that the persistent innovation effects of platform ecosystem embeddedness are systematically moderated by industry competition, regional resource endowments, and company-specific attributes. These findings broaden the theoretical boundary of contextual adaptability in platform embedding.

To more intuitively present the correspondence between the empirical findings and the research hypotheses,

Table 15. Summary of hypotheses and empirical results.

Hypothesis	Hypothesis Statement	Mechanism Type	Empirical Result
H1	Platform ecosystem embeddedness significantly promotes persistent innovation.	Overall Effect	Supported H1
H2a	Platform ecosystem embeddedness promotes persistent innovation input by reducing coordination costs.	Resource Synergy Mechanism	Supported H2a
H2b	Platform ecosystem embeddedness promotes persistent innovation input by improving operational efficiency.	Resource Synergy Mechanism	Supported H2b
H3a	Platform ecosystem embeddedness promotes persistent innovation output by enhancing knowledge flow.	Knowledge Synergy Mechanism	Supported H3a
H3b	Platform ecosystem embeddedness promotes persistent innovation output by enhancing knowledge creation.	Knowledge Synergy Mechanism	Supported H3b

Source: compiled by the author.

summarizes each hypothesis, its associated pathway type, and the corresponding test results.

Overall, this study responds to the call by Cusumano et al. [70] for deeper exploration of the long-term growth mechanisms enabled by platform economies. Building upon prior research that primarily emphasizes platform-driven breakthrough innovation or short-term performance improvements [14,20], we introduce the concept of persistent innovation into the platform ecosystem literature, thereby extending the theoretical boundary of platform empowerment.

At the mechanism level, we develop a dual-pathway framework based on resource synergy and knowledge synergy, which reveals how platform ecosystem embeddedness stimulates the “persistent momentum” of innovation under varying organizational and environmental conditions. This identification of mechanisms aligns with the approach of Jun et al. [20], who emphasize the mediating role of organizational readiness, and echoes the findings of Yu and Chen [9], who show that the empowering effects of industrial internet platforms depend critically on the innovation capability of companies.

7. Conclusions and Implications

7.1. Research Conclusions

Based on panel data of A-share listed companies in China from 2012 to 2023, we empirically examine the impact of platform ecosystem embeddedness on companies’ persistent innovation, including its underlying mechanisms and contextual conditions. The results show that platform ecosystem embeddedness significantly enhances companies’ performance in both dimensions of persistent innovation—manifested in the continuity of R&D activities and sustained output of innovation results. This effect remains robust after controlling for endogeneity and conducting multiple robustness checks, indicating that platform structures play a critical role in supporting long-term technological accumulation for companies.

The mechanism analysis further reveals that platform ecosystem embeddedness influences persistent innovation through two distinct paths of synergy. On the input side, platforms reduce supply–demand coordination costs and improve operational efficiency, enabling companies to sustain innovation input. On the output side, platforms promote persistent innovation output by enhancing knowledge flow and knowledge creation in complex environments.

In addition, the innovation effect of platform ecosystems exhibits strong contextual dependence. Companies operating in highly competitive industries or resource-abundant regions are more likely to benefit from full-chain innovation synergy—from input to output—through platforms. In contrast, under conditions of limited resources or insufficient competition, the effect of platform embeddedness is more concentrated on innovation output. The heterogeneity analysis of internal company characteristics also shows that, in high-tech and non-polluting companies, platform ecosystem embeddedness drives both innovation input and output; however, in non-high-tech or heavily polluting companies, the platform effect is primarily reflected in output stimulation, with delayed responses on the input side.

In summary, this study not only concompanies the positive impact of platform ecosystem embeddedness on persistent innovation, but also clarifies the dual-path mechanisms and contextual boundaries through which it operates. These findings provide empirical evidence and theoretical support for companies developing persistent innovation strategies, platform managers optimizing embeddedness mechanisms, and policymakers formulating differentiated platform governance policies.

7.2. Managerial Implications and Policy Recommendations

This study enhances the theoretical understanding of how platform ecosystems influence the continuity of corporate innovation and provides empirical and policy-oriented insights for improving digital transformation outcomes and promoting innovation governance based on platform empowerment.

• Managerial Implications for Companies

First, companies should prioritize “asset-light embeddedness” strategies—leveraging platform resources—over “asset-heavy acquisitions” to access innovation inputs. Companies are encouraged to actively develop strategies for platform ecosystem embeddedness to achieve efficient resource acquisition and dynamic capability reconfiguration. Our findings show that platform ecosystem embeddedness significantly improves both dimensions of persistent innovation, enabling companies to address challenges such as uneven resource allocation and insufficient capability accumulation in innovation processes. Practical approaches may include joining industry platforms, building interface-compatible systems, and participating in platform standard-setting processes to enhance linkages to external resources and capabilities.

Second, companies should focus on achieving dynamic alignment with the platform’s evolving ecosystem and improving their coordination and adaptation capabilities. This may involve establishing dedicated roles such as “platform liaison officers” or forming “digital collaboration teams” to enhance response efficiency in areas such as data interface management, platform-based project participation, and knowledge transformation. These efforts are particularly important in industries with rapid technological iteration and strong feedback mechanisms. Additionally, companies in resource-constrained regions or non-high-tech sectors should actively use modular platform resources and algorithmic recommendation systems to compensate for internal innovation limitations and enhance the quality and continuity of their innovation output.

Third, platform enterprises should strengthen support for persistent innovation among embedded companies through thoughtful architectural and governance design. On the technical side, improving API standards to enhance compatibility and reusability across modules can lower the entry barriers for member companies integrating platform resources. Providing standardized, well-documented, and continuously updated APIs, as well as supporting multi-language and multi-protocol development environments, enable companies to rapidly access core resources, reduce product iteration cycles, and enhance the responsiveness of persistent innovation. On the governance side, platforms should establish incentive mechanisms specifically designed to promote long-term innovation rather than rewarding only short-term breakthroughs. Examples include implementing “long-term collaboration scorecards,” “continuous improvement leaderboards,” or “stable iteration reward programs” that assess member companies based on technical update frequency, product optimization continuity, and active knowledge sharing. In turn, platforms can offer priority access to data resources, API privileges, or funding recommendations to reinforce persistent innovation behavior.

• Policy Recommendations for Governments

First, policymakers should recognize platform ecosystems as critical infrastructure for supporting persistent innovation. Our findings suggest that platform ecosystem embeddedness promotes innovation input and output through resource synergy and knowledge synergy, indicating that platforms are not only carriers of digital transformation but also institutional foundations for long-term innovation. Therefore, governments should incorporate platform development into national or regional innovation infrastructure strategies, offering comprehensive support through fiscal incentives, tax breaks, land use policies, and technology certification—similar to support mechanisms for industrial parks or technology incubators.

Second, differentiated platform embeddedness policies should be developed based on enterprise type and regional characteristics. Our heterogeneity analysis shows that platform ecosystem embeddedness is more effective for high-tech companies, non-polluting enterprises, and companies located in economically advanced eastern regions. As such, targeted incentives can be introduced. For example, subsidies can be provided to polluting enterprises to support platform integration for technological upgrading; high-potential companies—such as green manufacturing and high-tech companies—can be encouraged to deeply integrate with platform ecosystems to build regional collaborative innovation networks. These networks can foster frequent cooperation, knowledge sharing, and outsourced R&D. In addition, in central and western China, pilot projects combining “platform + industrial chain” can be implemented to fill regional gaps in platform ecosystems and promote efficient resource flow and knowledge exchange.

Third, platform governance mechanisms should transition toward a “persistent innovation orientation.” Given the long-term guiding effect of platform ecosystem embeddedness on company innovation behavior, governments should promote institutional innovation in platform governance. On the one hand, platform companies should be guided to optimize API standards, open data interfaces, and adopt continuous version updates to technically support ongoing R&D by embedded companies. On the other hand, regulatory frameworks can incentivize platforms to incorporate innovation inputs and outputs into their governance systems. Examples include establishing innovation performance-based rankings, providing tax rebates, and other measures that help unlock the institutional externalities of platforms in fostering persistent innovation.

7.3. Research Limitations and Directions for Future Research

Although this study develops a theoretical framework and conducts empirical testing to examine the impact of platform ecosystem embeddedness on companies’ persistent innovation and its underlying mechanisms, several limitations remain and call for further research.

First, this study is based on data from listed companies in China’s Shanghai and Shenzhen stock markets. Due to the specific institutional context, platform governance structures, and corporate disclosure practices, the generalizability of the findings to other countries or settings with different platform models requires caution. In particular, the applicability of the conclusions may vary in environments where platform governance is more decentralized or where innovation incentive systems differ.

Second, the measurement of platform ecosystem embeddedness relies on the frequency of platform-related keywords in annual reports of listed companies. While this approach offers explanatory power and accessibility of text-based data, it may still be influenced by subjective factors such as disclosure preferences and narrative style. Future research is encouraged to incorporate external data sources—such as platform participation status, the number of API calls, or the degree of digital interface integration—to improve the objectivity and robustness of this indicator.

Third, the measurement of innovation output primarily relies on the number of patent applications. While this captures aspects of technological output, it does not fully reflect non-patent innovation activities such as product development, collaborative innovation, or business process optimization. Future research could expand the scope of innovation output by including data on product launches, customer feedback, or case-based project documentation to better capture the multifaceted nature of innovation performance.

In light of these limitations, this study suggests several directions for future research. First, cross-country comparative studies using multinational company samples could further explore how institutional environments, governance regimes, and innovation cultures affect the relationship between platform ecosystem embeddedness and innovation performance. Second, future work could examine the heterogeneity of platform governance structures by analyzing how platform openness, exclusivity of rules, and incentive design moderate the effects of embeddedness on innovation outcomes. Such work would help identify optimal response strategies for companies operating under different platform architectures. Third, case studies of bilateral embeddedness between platforms and companies could shed light on the interaction patterns in collaboration mechanisms, behavioral strategies, and resource coordination. This approach may reveal the dynamic logic of innovation empowerment within platform ecosystems and contribute to a more comprehensive understanding of platform-governed innovation networks.