Investigating the microfoundation of dynamic managerial and organizational capabilities for business model innovation: a mediated-moderated model of artificial intelligence

Abstract

In rapidly evolving digital environments, organizations must develop dynamic capabilities to drive business model innovation (BMI). This study examines how dynamic managerial capabilities (DMCs), including internal advice-seeking, external advice-seeking, and collaborative sensemaking, translate into BMI through two AI-driven organizational capabilities: artificial intelligence integration capability (AIIC) and AI-augmented coordination flexibility capability. Guided by the dynamic capabilities view (DCV), we construct and empirically test a multilevel model using survey data from 517 managers purposively sampled from Chinese firms undergoing digital transformation. Results from partial least squares structural equation modeling indicate that internal advice-seeking has the strongest direct effect on BMI. Conversely, external advice-seeking most strongly influences AIIC. AI integration enhances coordination flexibility, which in turn drives BMI. The findings support a serial mediation model in which DMCs influence BMI through AI integration and coordination flexibility. Moreover, a data-driven decision-making orientation significantly strengthens the relationship between coordination flexibility and BMI, underscoring the importance of organizational mindset. This study advances the DCV by introducing and validating two AI-driven capabilities that link sensing, seizing, and transforming in strategic renewal. Managers who actively seek external insights, foster collaborative sensemaking, and lead data-driven teams are best positioned to leverage AI for sustainable BMI.

Introduction

In today’s ambiguous, complex, uncertain, and volatile business environment, artificial intelligence (AI) has become central to how organizations pursue strategic transformation (Al-khatib & Ramayah, 2025; Ben Jabeur et al., 2023; Gama & Magistretti, 2025; Sjödin et al., 2021). While organizations increasingly invest in AI adoption and digital infrastructure, business model innovation (BMI), defined as “a change in a firm’s value offering, its value-creation architecture, or its revenue model logic” (Spieth & Schneider, 2016), has emerged as a key outcome of AI-driven transformation (Climent et al., 2024). Unlike incremental digital upgrades or process automation, BMI entails structural changes that introduce novel value logics and reposition firms within markets (Jabeur et al., 2024). Thus, BMI is not merely a byproduct of digitalization but a strategic outcome shaped by how organizational capabilities are developed and deployed within organizations (Jorzik et al., 2024; Martinez, 2022).

Although scholars acknowledge that successful BMI implementation is not purely technological, it remains a capability-driven and socially embedded process (Ding, 2026; Renfei & Zhongwen, 2026; Sjödin et al., 2021). To understand how organizations adapt disruptive technologies such as AI, the dynamic capabilities view (DCV) provides a compelling theoretical lens (Teece, 2018). The DCV posits that organizations must continuously sense opportunities, allocate resources to exploit them, and reconfigure internal processes to sustain competitive advantage. While conventional DCV research emphasizes firm-level capabilities (e.g., agility, absorptive capacity, and innovation capabilities), growing interest highlights their managerial microfoundations—namely, the cognitive, relational, and interpretive behaviors enacted by individual decision-makers (Durán & Aguado, 2022; Nicolai Foss & Mazzelli, 2025). This perspective underscores how managerial actions shape strategic adaptation in uncertain environments (Hock-Doepgen et al., 2025).

Building on these microfoundations, dynamic managerial capabilities (DMCs) have gained attention as key drivers of innovation-driven transformation (Panda, 2025). DMCs reflect managers’ abilities to develop, integrate, and reconfigure organizational resources (Helfat & Martin, 2014). Although prior research has examined DMCs extensively (Helfat & Martin, 2014; Hock-Doepgen et al., 2025), few studies have operationalized their specific behavioral dimensions in volatile digital contexts (Chedrawi et al., 2025). Drawing on the managerial learning and cognition literature, we focus on three actionable DMCs: internal advice-seeking capabilities (IASCs), external advice-seeking capabilities (EASCs), and collaborative sensemaking capabilities (CSCs).

IASC refers to a manager’s propensity to seek advice from colleagues and subordinates to acquire knowledge and address uncertainty. EASC captures manager engagement with external networks (e.g., partners, consultants, and industry experts) to access knowledge beyond firm boundaries (Adomako et al., 2022; Ma et al., 2019). Alternatively, CSC reflects higher-order social-cognitive processes through which managers collectively construct meaning around novel or complex events (Prior et al., 2018). Although each capability is valuable, their combined role remains underexplored, particularly in AI-driven innovation contexts. Moreover, CSC remains underdeveloped in DCV literature despite its importance in orchestrating organizational learning under technological uncertainty (Engström et al., 2024).

To examine how DMCs translate into organizational outcomes, we propose a multilevel framework in which DMCs give rise to two complementary dynamic capabilities: AI integration capability (AIIC) and AI-augmented coordination flexibility capability (AICFC). AIIC refers to the organization’s ability to integrate AI systems into resource allocation and decision-making processes, moving beyond superficial adoption toward strategic integration (Kayser & Gradtke, 2024; Wodecki et al., 2019). AICFC captures the organization’s ability to flexibly reconfigure and coordinate activities in response to AI-enabled complexity and insights (Hartati et al., 2025; Rehman & Jajja, 2023). These constructs are conceptually distinct from absorptive capacity, information technology (IT) capability, or generic coordination flexibility. AIIC emphasizes embedding algorithmic logic into core processes, whereas AICFC reflects cross-functional orchestration and human–machine collaboration under dynamic conditions. Together, they represent AI-specific dynamic capabilities that support sensing, seizing, and transforming within the DCV framework (Teece, 2018; Teece et al., 1997).

We further argue that the relationship between coordination flexibility and BMI is not automatic but contingent on the organization’s data-driven decision-making orientation (DDDMO). DDDMO represents a strategic orientation (Szukits, 2022) emphasizing analytical rigor, evidence-based decisions, and transparency across managerial levels (Garcia & Adams, 2023). It reflects a firm’s commitment to translating data into decisions and has been linked to innovation consistency, learning speed, and resource utilization (Gade, 2021; Karaboğa et al., 2019). However, its moderating role between capability and innovation remains underexplored. Accordingly, we position DDDMO as a moderator that strengthens the effect of AICFC on BMI, suggesting that capabilities require an appropriate mindset to generate innovation.

This study develops a conceptual model that follows a micro–to–macro sequence: managerial-level DMCs shape organizational-level AIIC and AICFC, which ultimately drive BMI. DDDMO is modeled as a moderator in the final stage, influencing how coordination capabilities translate into strategic outcomes. This structure reflects both the DCV logic (sensing, seizing, transforming) and the multi-actor nature of digital transformation. Based on the DCV of (Teece, 2018), this study addresses how DMCs evolve into firm-level practices that enable BMI. Specifically, we examine:

• •

  Do DMCs (IASC, EASC, and CSC) influence AIIC and BMI?

• •

  Do AIIC and AICFC mediate independently and sequentially mediate the relationship between DMCs and BMI?

• •

  Does DDDMO moderate the relationship between AICFC and BMI?

Although the DCV has advanced understanding of firm adaptation, key gaps remain in explaining how managerial actions translate into strategic transformation (Durán & Aguado, 2022). Existing research has highlighted the importance of DMCs (Nicolai Foss & Mazzelli, 2025; Hossain et al., 2025), yet they lack a clear pathway linking individual capabilities to organizational innovation (Wang et al., 2026). This study makes three contributions. First, it clarifies the sequential pathway through which managerial capabilities enable transformation via AI integration and coordination (Al-khatib & Ramayah, 2025; Renfei & Zhongwen, 2026). Second, it introduces and validates two AI-based dynamic capabilities (AIIC and AICFC) that connect microfoundations to firm-level innovation, extending the DCV. Third, it identifies the moderating role of organizational mindset, demonstrating that even advanced coordination systems depend on a data-driven culture to achieve strategic renewal (Tran-Dang et al., 2025). Practically, this study offers a conceptual model (see Fig. 1) to guide firms in navigating digital transformation by identifying key managerial behaviors and capabilities that enable AI-driven innovation.

Fig. 1.

Proposed model.

Theoretical background and hypotheses formulation

Microfoundations of DMCs and BMI

BMI requires managers not only to adopt new technologies but also to redefine firm logistics and reconfigure resources to create and capture new forms of value (Spieth & Schneider, 2016). The DCV posits that successful strategic renewal is grounded in three interrelated processes—sensing, seizing, and transforming—driven by managerial action and cognition (Teece, 2018). These processes operate through DMCs, which represent behavioral and cognitive microfoundations that enable organizations to sense opportunities, mobilize resources, and implement change (Durán & Aguado, 2022; Helfat & Martin, 2014). In AI-enabled contexts, DMCs are especially critical because technological opportunities are abundant yet ambiguous, requiring managers to interpret weak signals and translate data-driven insights into coherent strategic actions (Jorzik et al., 2024). Among managerial capabilities, three core DMCs align with the DCV processes: IASC as sensing, EASC as seizing, and CSC as transforming. These capabilities shape how managers access, evaluate, and respond to diverse information sources to reshape firm value logic (Hock-Doepgen et al., 2025).

IASC involves managers leveraging intra-organizational knowledge by seeking advice from peers, subordinates, and internal experts (Hock-Doepgen et al., 2025). This internal focus enables managers to identify bottlenecks, uncover latent capabilities, and detect process inefficiencies, generating localized insights critical for innovation (Kubíček et al., 2021; Piening & Salge, 2015). Through dialogue and reflection, managers surface underutilized knowledge and refine routines—core sensing activities that underpin BMI (Ghinoi & Di Toma, 2022). Most recently, Hock-Doepgen et al. (2025) reported the impact of IASC on BMI among German mid-size firms. Thus, we hypothesize that

H1a: IASC has a significant influence on BMI.

EASC refers to managers’ ability to access opportunities through engagement with external stakeholders, including customers, suppliers, consultants, and professional networks (Ma et al., 2019; Yitshaki, 2025). These external ties expose organizations to technological frontiers, regulatory shifts, and market evolution, enabling the discovery of AI-driven opportunities and new value-creation logics (Clausen & Molden, 2024; van doorn et al., 2017; Yitshaki, 2025). From a DCV perspective, EASC reflects the seizing phase by facilitating external knowledge acquisition and collaboration that supports new products and ecosystems (Ma et al., 2023). Most recently, Hock-Doepgen et al. (2025) reported the impact of IASC on BMI among German mid-size firms. Thus, we hypothesize that

H1b: EASC has a significant influence on BMI.

Finally, CSC underpins the transforming phase of the DCV by enabling managers to co-create meaning under conditions of ambiguity (Tessema & Reilly, 2025). According to Prior et al. (2018) it defines the capacity of the manager to co-create shared understanding and coordinate adaptive action among heterogeneous stakeholders. In AI-intensive contexts, CSC allows managers to challenge mental models and experiment with new value creation and capture mechanisms (Gattringer et al., 2021; Kayser & Gradtke, 2024). It supports the interpretation of data, alignment of stakeholders, and framing of strategic issues through continuous interaction (Keding, 2021). CSC thus facilitates the reframing of business model assumptions and encourages experimentation and adaptation. Therefore, we hypothesize the following:

H1c: CSC has a significant influence on BMI.

Microfoundations of DMCs and AIIC

In the DCV, managerial action is posited as the foundation for the evolution of firm-level dynamic capabilities (Helfat & Martin, 2014; Teece, 2018). An emerging capability within this tradition is AIIC, which refers to a firm’s ability to integrate AI tools and systems into strategic and operational processes to improve decision-making, catalyze innovation, and enhance coordination. AIIC extends beyond traditional absorptive capacity or IT integration by emphasizing not only the deployment of AI systems but also their alignment with strategic objectives and integration into value-creation processes (Hossain et al., 2025; Mikalef et al., 2021). See Appendix A for more information. Unlike generic IT capabilities focused on interoperability or infrastructure, AIIC represents an agile, strategic competency shaped by managerial communication, cognition, and cross-unit cohesion. We posit that DMCs (IASC, EASC, and CSC) serve as microfoundational mechanisms that support the emergence of AIIC by enabling organizations to sense and exploit AI opportunities and by facilitating the interpretive and relational processes required for effective AI assimilation.

Managers who seek internal and external advice are better able to identify AI applications, understand implementation challenges, and align technology with business objectives (Burström et al., 2021; Keding, 2021; Rane et al., 2024). Managers with strong IASC actively engage with technical teams, operational staff, and mid-level managers to assess digital readiness, identify inefficiencies, and evaluate future AI applications (Li & Huang, 2024; Tuncer & Varoglu, 2025). These interactions help uncover process-level challenges, clarify system integration points, and reveal opportunities for intelligent automation (Ahmadi, 2023; Stoykova & Shakev, 2023). Such internally generated insights are critical for tailoring AI tools to existing workflows, representing a key early stage in developing AIIC (Baines et al., 2024). Thus, we hypothesize that

H2a: IASC has a significant influence on AIIC.

EASC provides access to external AI trends, benchmark practices, digital strategies, and vendor solutions (Arroyabe et al., 2024). Managers leveraging external networks, including consultants, cross-industry peers, and suppliers, are better positioned to seize emerging AI opportunities and acquire complementary expertise. This external knowledge enables firms to assess, adapt, and embed AI solutions that are both contextually relevant and globally informed (Croitoru et al., 2025; Hamilton et al., 2024). Accordingly, the seizing function of EASC shapes the scope and quality of AIIC development.

H2b: EASC has a significant influence on AIIC.

Beyond knowledge acquisition, CSC enables managers to interpret complex AI information and collaboratively develop a shared vision for digital integration (Engström et al., 2024). Because AI adoption involves technical complexity, risk, and potential resistance, managers must engage in cross-functional dialogue to align goals and support strategic implementation (Hangl et al., 2023; Rane et al., 2024). Collaborative sensemaking fosters alignment, reduces misunderstanding, and facilitates coordinated responses to AI adoption. In this way, CSC strengthens the transformative dimension of the DCV and positions AIIC as a socially embedded and interpretively unified capability. Therefore, we propose that

H2c: CSC has a significant influence on AIIC.

AIIC and AICFC

Grounded in the DCV, the integration of AI tools across an organization’s operational and strategic layers extends beyond technological adoption, representing a sensing and seizing capability that underpins more adaptive and transformative capabilities (Al Dhaheri et al., 2024; Mikalef et al., 2021; Teece, 2018). As noted, AIIC refers to a firm’s ability to integrate AI systems into decision-making, operational processes, and workflows, positioning AI as a key mediator of insight generation and resource coordination. However, translating these technological assets into organizational flexibility requires higher-order capabilities such as AICFC (Ning et al., 2019). This capability reflects the organization’s ability to flexibly reconfigure coordination structures using AI-driven insights to respond to environmental changes and interdependent demands (Qalati & Siddiqui, 2026; Sjödin et al., 2021; Yesufu & Alajlani, 2025).

AIIC enables AICFC through the integration of intelligent systems that detect interdependencies, anticipate disruptions, and recommend work configurations (Belhadi et al., 2024). For example, AI-driven platforms can forecast demand, automate resource planning, and algorithmically schedule tasks, reducing delays and improving responsiveness (Nweje & Taiwo, 2025). When deeply embedded, these systems generate high-quality data streams that support decentralized, real-time, and predictive coordination—key features of flexible execution aligned with BMI (Tien, 2017). As AI becomes more integrated, it evolves into a coordination backbone that reduces silos and enables distributed action aligned with strategic goals (Wang et al., 2026; Zou & Yang, 2026).

Moreover, AICFC represents a dynamic capability because it transforms not only organizational structures but also the underlying logic of coordination. Unlike static or rule-based processes, AI-enabled flexibility involves continuous learning, iterative adjustment, and autonomous task orchestration, fostering innovation and experimentation (Badmus et al., 2024; Sjödin et al., 2020). This transformative capacity is critical for aligning internal actions with BMI objectives, particularly in dynamic digital environments where coordination speed and precision drive strategic renewal (Brock & von Wangenheim, 2019; Ghosh, 2025). Therefore, we hypothesize that

H3: AIIC has a significant influence on AICFC.

AICFC and BMI

BMI is a multifaceted, iterative process involving the redesign of an organization’s value creation, capture, and delivery mechanisms to meet evolving demands (Climent et al., 2024; Jorzik et al., 2024). From a DCV perspective, this redesign is not only about identifying opportunities but also about reconfiguring firm resources, roles, and routines to respond effectively (Teece, 2018). AICFC refers to a dynamic organizational capability that enables real-time adjustment of interdependent resource flows, tasks, and workflows through AI-driven systems (Brock & von Wangenheim, 2019; Raisch & Krakowski, 2021). This capability aligns with the transforming function of the DCV, enabling firms to modify internal processes and structures to support strategic change (Teece, 2018).

AICFC extends beyond traditional coordination by embedding AI into day–to–day decision-making, using intelligent algorithms to enable continuous synchronization across functions (Sjödin et al., 2020). Unlike hierarchical coordination systems, AI-driven coordination leverages real-time data to redistribute resources, anticipate disruptions, and support decentralized yet integrated decision-making (Yesufu & Alajlani, 2025). This enables firms to respond quickly to external changes and enhances their ability to implement innovative business models, such as servitization, platformization, or outcome-based pricing (Sjödin et al., 2020; Zott & Amit, 2010).

For example, monitoring real-time indicators or forecasting demand changes allows firms to dynamically reallocate resources, facilitating shifts in business models, such as transitions from product-based to service-oriented offerings or data-driven platforms (Zott & Amit, 2010). This reduces the time from idea to implementation and makes renewal more feasible and repeatable ( Nicolai Foss & Saebi, 2016).

Additionally, coordination flexibility supports the experimentation and iteration central to BMI (Snihur & Eisenhardt, 2022). AI-driven coordination enables process reconfiguration with lower switching costs and faster learning, allowing firms to prototype, test, and refine new business models more efficiently (Farayola et al., 2023). Prior research shows that BMI relies on trial-and-error, resource recombination, and cross-functional collaboration rather than purely planned change (Hock-Doepgen et al., 2025). Firms with high AI coordination flexibility can manage these processes more effectively and implement changes more rapidly (Shmatko & Ivchyk, 2024). Most recently, Zou and Yang (2026) and Cassânego et al. (2025) showed that AI capabilities enhance firm, process, and technological innovation. Furthermore, Renfei and Zhongwen (2026) and Qalati and Siddiqui (2026) showed that sustainable AI capabilities support business innovation Additionally, Ding’s (2026) study on Chinese firms reports a positive relationship between digital capabilities and BMI. Therefore, we posit

H4:AICFC positively influences BMI.

AIIC and aicfc as a mediator

While DMCs, including advice-seeking and collaborative sensemaking, are essential for recognizing opportunities and shaping strategic vision, their influence on BMI is largely indirect (Helfat & Martin, 2014). Rather, these microfoundations trigger organizational processes that evolve into higher-order capabilities (Teece et al., 1997). Drawing on the DCV, we argue that DMCs influence BMI through staged capability development (Teece et al., 1997). In this framework, AIIC represents the “seizing” phase, where firms leverage AI for strategic decision-making, resource allocation, and process optimization (Brock & von Wangenheim, 2019; Mikalef et al., 2021). Managers with strong DMCs—particularly in internal and external advice-seeking and collaborative sensemaking—are better positioned to evaluate, adopt, and champion AI aligned with strategic objectives (Helfat & Martin, 2014; Teece, 2016).

AI integration serves as a central mediating capability, reflecting how AI systems are embedded in strategic and operational activities (Brock & von Wangenheim, 2019). DMCs enable managers to identify suitable AI applications, promote adoption, and facilitate implementation through knowledge networks and interpretive alignment (Helfat & Martin, 2014; Teece, 2016). This process allows firms to seize opportunities by translating managerial insight into scalable AI-enabled routines (Erik & Andrew, 2017). Once integrated, AI forms the technological foundation for developing more advanced capabilities.

Following integration, AIIC enables the emergence of AICFC, categorized by Yesufu and Alajlani (2025) as the “transforming” phase, in which firms reconfigure tasks, roles, and workflows based on AI-driven insights (Volberda et al., 2010). This capability allows firms to reallocate resources, rebalance interdependencies, and optimize value chains, all of which are critical for achieving BMI (Zott & Amit, 2010). However, AICFC does not arise directly from managerial actions; it depends on established AI systems and infrastructure (Brock & von Wangenheim, 2019). Thus, AI integration acts as a prerequisite for coordination flexibility, forming an adaptive, AI-enabled coordination layer.

Together, these capabilities create a serial mediation pathway: DMCs drive AI integration, which enhances coordination flexibility, ultimately leading to BMI. This sequence reflects the DCV logic of sensing, seizing, and transforming (Hock-Doepgen et al., 2025; Teece, 2018). AI integration translates managerial cognition into technological capability, while coordination flexibility converts these capabilities into organizational transformation that supports BMI. We propose:

H5a: AIIC mediates the relationship between DMCs and AICFC.

H5b: AICFC mediates the relationship between AIIC and BMI

H5c: AIIC and AICFC jointly mediate the relationship between DMCs and BMI in a serial mediation pathway.

DDDMO as moderator

Although AICFC provides organizations with dynamic flexibility, its effectiveness in fostering BMI is not universal. Rather, it depends on the organization’s strategic mindset and cultural orientation toward data-driven decision-making. We conceptualize DDDMO as an organization-level cultural capability that systematically integrates algorithms, analytics, and data-driven reasoning into managerial and operational decisions (Davenport & Harris, 2017). Grounded in the DCV, DDDMO strengthens the transformation phase by ensuring that AI-enabled coordination translates into effective adaptation and renewal (Brynjolfsson & McElheran, 2016; Choudhury et al., 2021). While AICFC enables flexible task reconfiguration and integration, without strong DDDMO, such flexibility may not result in consistent or strategic business model change. For instance, firms with low data orientation may fail to leverage advanced coordination, relying instead on intuition, experience, or organizational politics, which can lead to delays, errors, or resistance to AI-driven insights (Ransbotham et al., 2017).

DDDMO acts as a behavioral and normative amplifier of AI-supported coordination by fostering trust in real-time data, enabling cross-functional knowledge sharing, and reinforcing a performance logic centered on experimentation and iteration—core elements of BMI (Wamba et al., 2017). In data-intensive organizations, executives are more likely to emphasize rapid adaptation, risk-taking, and opportunities revealed by digital technologies, thereby strengthening the impact of AICFC on innovation outcomes.

Conversely, weak analytical orientation reduces coordination effectiveness and limits its contribution to BMI. Firms lacking strong data-driven capabilities often struggle to align AI tools with organizational strategy, constraining the development of transformative business models (Wamba et al., 2017). Therefore, we hypothesize:

H6: A DDDMO positively moderates the relationship between AICFC and BMI, such that the relationship is stronger when DDDMO is high.

Methodology

Research context and sample

To examine the relationships above, we employed a quantitative, cross-sectional research design guided by deductive reasoning, consistent with the theory-driven model. Our empirical context included Chinese firms across multiple sectors, as China is a global leader in AI-driven digital transformation. Government strategies and the emphasis on BMI under the dual-circulation policy make this setting particularly appropriate (Zhang, 2025). Rapid digitalization in China offers valuable insights into how managerial actions shape firm-level capabilities (Cao et al., 2023). Among leading digital economies such as the US, South Korea, and Germany, China’s combination of strong policy support, manufacturing digitization, and large-scale AI adoption provides a unique, yet underexplored, context for examining how managerial activity influences dynamic capabilities and innovation outcomes (Liu et al., 2025; Shan, 2023; Wu et al., 2020).

To collect data, we used purposive (judgmental) sampling, targeting mid- to senior-level executives involved in digital strategy, innovation, or technology implementation (Kanski & Pizon, 2023). These respondents were well positioned to provide insights into AI integration, coordination flexibility, and business model changes (Sjödin et al., 2023). A screening question ensured that only individuals actively engaged in digital transformation or innovation participated. Although this non-probability sampling approach may limit generalizability and introduce selection bias, we mitigated these concerns by selecting information-rich respondents to support analytical validity (Etikan et al., 2016). The objective was not statistical representativeness but theoretical generalizability across AI-intensive, transformation-driven firms.

Data were collected between January and April 2025 through online survey platforms and institutional research networks. After excluding incomplete and low-quality responses, the final sample comprised 517 valid observations. Table 1 indicates that males (60.3%) outnumber females (39.7%). Most firms were 11–20 years old (35.6%) or over 20 years (29.6%), with primary representation in manufacturing (29.6%) and IT/digital services (25.0%). Senior-level employees accounted for 58.8% of respondents, with functional areas including strategy/planning (26.3%), IT/digital transformation (22.2%), and operations/supply chain (21.9%). Ownership was predominantly non-state/private (45.1%) or state-owned (39.1%). Overall, the sample reflects an experienced workforce across sectors, largely concentrated in the private sector.

Table 1.

Respondent profile and firm-level characteristics.

Construct	Characteristics	Frequency	%
Gender
	Male	312	60.3
	Female	205	39.7
Managerial-level
	Senior-level	304	58.8
	Middle level	213	41.2
Department
	Strategy/planning	136	26.3
	Information technology (IT)/digital transformation	115	22.2
	Operations/supply chain	113	21.9
	Innovation/research and development	108	20.9
	Others	45	8.7
Industry sector
	Finance and insurance	86	16.6
	IT and digital service providers	129	25.0
	Manufacturing	153	29.6
	Retail and e-commerce	48	9.3
	Logistics	101	19.5
Firm age (years)
	<5	41	7.9
	5–10	139	26.9
	11–20	184	35.6
	Over 20	153	29.6
Firm ownership
	Joint ventures or foreign-owned	82	15.9
	State-owned	202	39.1
	Non-state/privately owned	233	45.1

Data collection procedure and common method variance (CMV)

To ensure data quality, we employed a two-step data collection process. First, a pilot study with 30 managers was conducted to improve item clarity and reliability. The final survey was then administered through professional networks and research platforms, ensuring respondent anonymity.

To mitigate CMV, we applied both procedural and statistical controls following (Podsakoff et al., 2003). Procedurally, we ensured anonymity and confidentiality, randomized item order across variables, used mixed-scale formats (e.g., reverse-coded items), and applied distinct labels for adjacent questions. Statistically, we conducted several CMV tests. Harman’s single-factor test indicated that the first factor explained 29.8% of the variance (below the 50% threshold). All variance inflation factors (VIFs) were below 3.3 (Kock, 2015), suggesting no multicollinearity concerns (see Table 2). Latent factor analysis showed no significant inflation across variables. We also applied the correlational marker technique (Lindell & Whitney, 2001), using respondents’ views on ecological issues as a theoretically unrelated marker variable. This variable did not significantly affect relationships among the study constructs, confirming discriminant validity. Although interaction terms (e.g., moderation by DDDMO) are inherently robust to CMV (Siemsen et al., 2009)), we acknowledge that no method can fully eliminate CMV risk.

Table 2.

Assessment of reliability and validity.

Construct	Items	Loading	CR	AVE	VIF
Internal advice-seeking capabilities (IASC)	IASC1	0.899	0.933	0.831	2.119
	IASC2	0.918
	IASC3	0.934
External advice-seeking capabilities (EASC)	EASC1	0.904	0.926	0.767	1.758
	EASC2	0.875
	EASC3	0.890
Collaboration sensemaking capability (CSC)	CSC1	0.874	0.941	0.769	2.363
	CSC2	0.853
	CSC3	0.904
	CSC4	0.889
	CSC5	0.876
	CSC6	0.864
AI-integrated capability (AIIC)	AIIC1	0.788	0.924	0.684	1.00
	AIIC2	0.831
	AIIC3	0.859
	AIIC4	0.840
	AIIC5	0.843
	AIIC6	0.812
	AIIC7	0.813
AI-augmented coordination flexibility capability (AICFC)	AICFC1	0.885	0.930	0.738	1.874
	AICFC2	0.881
	AICFC3	0.868
	AICFC4	0.886
	AICFC5	0.868
	AICFC6	0.760
Data-driven decision-making orientation (DDDMO)	DDDMO1	0.878	0.910	0.687	1.086
	DDDMO2	0.856
	DDDMO3	0.853
	DDDMO4	0.856
	DDDMO5	0.758
	DDDMO6	0.764
Business model innovation (BMI)	BMI1	0.807	0.929	0.636
	BMI2	0.812
	BMI3	0.810
	BMI4	0.832
	BMI5	0.756
	BMI6	0.811
	BMI7	0.800
	BMI8	0.798
	BMI9	0.752

Measures

All constructs were measured using established and validated scales adapted to the AI-driven business context in China. Items were measured on five-point Likert scales (see Appendix B). DMCs, specifically internal and external advice-seeking, were measured using six items (three each) adapted from Hock-Doepgen et al. (2025). Collaborative sensemaking was measured using six items adapted from Akgün et al. (2012). AIIC was measured using seven items adapted from Chatterjee et al. (2024)), while AICFC was assessed using six items adapted from Hock-Doepgen et al. (2025). DDDMO was measured using six items adapted from Dogan and Demirbolat (2021). Finally, the dependent construct BMI was measured using nine items adapted from Spieth and Schneider (2016).

To ensure cross-cultural validity, all items underwent back-translation (Brislin, 1970). The translation process—from English to Mandarin and back to English—was conducted by two bilingual professors. The items were further validated through a pilot test to confirm cultural appropriateness in the Chinese context. Control variables included firm age, industry sector, and ownership type, capturing structural and institutional factors that may influence digital transformation and innovation.

Analytical strategy

To analyze the proposed model, we used partial least squares (PLS) structural equation modeling (SEM) in SmartPLS 4.2. This approach is well suited for theory-driven model development, particularly for complex models involving higher-order constructs, non-normal data, and mediation and moderation effects (Becker et al., 2012; Hair et al., 2019). It enables robust estimation of direct, mediating, moderating, and combined effects (Hair et al., 2017). The model was estimated using the standard path algorithm for all first-order reflective constructs, with bootstrapping (5000 resamples) applied for path estimation, mediation, and interaction analysis (Hair et al., 2017).

Results

Measurement model assessment

To assess the reliability and validity of the measurement model, we followed established procedures (Hair et al., 2019). Internal consistency was evaluated using indicator reliability and composite reliability (CR). Table 2 shows that all factor loadings exceed 0.70, indicating satisfactory item reliability; therefore, no items were removed. CR values also exceed the recommended threshold of 0.70 (Hair et al., 2019).

Convergent validity was confirmed through average variance extracted (AVE) values above the 0.5 threshold (Hair et al., 2019). Multicollinearity was assessed using VIF, which ranged from 1.086 to 2.363, indicating that the constructs are distinct and free from redundancy (Hair et al., 2019).

The correlation matrix reveals meaningful relationships among the constructs, with all values below the discriminant validity thresholds (HTMT < 0.85 and 0.90) (Hair et al., 2019; Henseler et al., 2015). AICFC shows moderate to strong correlations with AIIC (r = 0.611), CSC (r = 0.587), and IASC (r = 0.589), indicating conceptual alignment. AIIC exhibits its strongest associations with EASC (r = 0.708) and CSC (r = 0.664), reinforcing its role in knowledge-driven processes (see Table 3).

Table 3.

Assessment of discriminant validity.

	AICFC	AIIC	BMI	CSC	DDDMO	EASC	IASC
AI-AICFC	1
AI-integrated capability	0.611	1
BMI	0.550	0.503	1
Collaboration sensemaking capability (CSC)	0.587	0.664	0.539	1
DDDMO	0.287	0.259	0.504	0.241	1
EASC	0.527	0.708	0.524	0.702	0.278	1
IASC	0.589	0.61	0.534	0.535	0.331	0.479	1

BMI demonstrates moderate correlations across constructs, with the strongest relationship with DDDMO (r = 0.504), underscoring the importance of data-driven orientation in innovation. In contrast, DDDMO shows weaker correlations with other variables (all r < 0.35), indicating its distinctiveness as a standalone construct. The absence of excessively high correlations (all r < 0.708), together with satisfaction of the Fornell-Larcker criterion (diagonal AVE√ > inter-construct correlations), confirms discriminant validity and supports the robustness of the measurement model for further analysis (Fornell & Larcker, 1981).

Structural model assessment

The structural model demonstrates strong predictive power and satisfactory fit. R² values for endogenous constructs exceed the acceptable threshold of 0.10 (Falk & Miller, 1992), with particularly strong explanatory power for AICFC (R² = 0.617) and BMI (R² = 0.518) (see Fig. 2), indicating substantial explained variance in key outcomes. Model fit indices meet recommended benchmarks, with SRMR (0.057) below the 0.08 cutoff (Hu & Bentler, 1999) and all Q² values above zero, confirming predictive relevance (Hair et al., 2019).

Fig. 2.

Structural equating model.

All path coefficients are statistically significant (p < 0.05), supporting the proposed relationships. Notably, AIIC has a strong effect on AICFC (β = 0.611, p < 0.001), and DDDMO exhibits a significant moderating effect (β = 0.058, p < 0.01). Effect sizes (f²) range from small to large (Cohen, 1988), with particularly large effects for AIIC → AICFC (f² = 0.595) and EASC → AIIC (f² = 0.222) (see Table 4).

Table 4.

Assessment of structural model and model fit.

Hypothesis	Paths	Beta (S.D)	t-value	Confidence interval (LL–UL)	f^2	Supported
Direct effect
H1a	IASC → BMI	0.165^⁎⁎⁎(0.047)	3.483		0.032	Yes
H1b	EASC → BMI	0.150^⁎⁎(0.052)	2.913		0.022	Yes
H1c	CSC → BMI	0.156^⁎⁎(0.056)	2.808		0.021	Yes
H2a	IASC → AIIC	0.297^⁎⁎⁎(0.040)	7.411		0.159	Yes
H2b	EASC → AIIC	0.415^⁎⁎⁎(0.044)	9.493		0.222
H2c	CSC → AIIC	0.214^⁎⁎⁎(0.039)	5.557		0.055
H3	AIIC → AICFC	0.611^⁎⁎⁎(0.030)	20.071		0.595
H4	AICFC → BMI	0.199^⁎⁎⁎(0.051)	3.873		0.044
Indirect effect
H5a	IASC → AIIC → AICFC	0.181^⁎⁎⁎(0.027)	6.710	[0.130–0.234]		Yes
	EASC → AIIC → AICFC	0.254^⁎⁎⁎(0.029)	8.894	[0.197–0.309]		Yes
	CSC → AIIC → AICFC	0.131^⁎⁎⁎(0.025)	5.192	[0.080–0.180]		Yes
H5b	AIIC → AICFC → BMI	0.121^⁎⁎⁎(0.032)	3.816	[0.061–0.185]
H5c	IASC → AIIC → AICFC → BMI	0.036^⁎⁎(0.011)	3.413	[0.018–0.059]
	EASC → AIIC → AICFC → BMI	0.050^⁎⁎⁎(0.014)	3.549	[0.025–0.081]
	CSC → AIIC → AICFC → BMI	0.026^⁎⁎(0.009)	3.016	[0.012–0.046]
Moderation effect
H6	DDDMO x AICFC → BMI	0.058*(0.029)	2.015	[0.003–0.116]		Yes

Note: ^⁎⁎⁎P < 0.001, ^⁎⁎P < 0.01, *P < 0.05,.

IASC = internal advice-seeking capabilities; EASC = external advice-seeking capabilities; CSC = collaboration sensemaking capability; AIIC = AI-integrated capability; AICFC = AI-augmented coordination flexibility capability; BMI = business model innovation; DDDMO = data-driven decision-making orientation; LL = lower limit; UL = lower limit; S.D = standard deviation.

R ²_AIIC = 0.617; R ²_AICFC = 0.373; R ²_BMI = 0.518.

R ²_BMI = 0.440 (without moderation effect).

Q² _AIIC = 0.417; Q²_AICFC = 0.273; Q²_BMI = 0.326.

SRMR = 0.057.

Mediation effect assessment

Table 4 demonstrates significant mediation effects using established statistical approaches. Following the bootstrap confidence interval method (Preacher & Hayes, 2008), all indirect pathways are statistically significant, as their 95% confidence intervals exclude zero. For example, the sequential mediation from IASC through AIIC to AICFC (β = 0.036, CI [0.018–0.059]) is confirmed. Similarly, the indirect effect of AIIC on BMI through AICFC (β = 0.121, CI [0.061–0.185]) is significant.

To assess mediation strength, the variance accounted for (VAF) approach was applied (Hair, 2014)). Results show that AIIC partially mediates the relationship between EASC and AICFC (VAF > 50%), while AICFC partially mediates the link between AIIC and BMI (VAF ≈ 37.8%), supporting H5a and H5b. These findings indicate that AIIC plays a central role in translating knowledge-seeking behaviors into coordination flexibility, which in turn contributes to BMI.

Additionally, the three-stage mediation pathway shows that DMCs (e.g., EASC) influence BMI sequentially through AIIC and AICFC (β = 0.050, CI [0.025–0.081]), supporting H5c and reinforcing the phased development of organizational capabilities.

Moderation effect assessment

The results show a significant interaction effect of DDDMO (β = 0.058, p = 0.044), supporting the moderating hypothesis (H6) (see Table 4). To further assess its impact, we compared model explanatory power with and without DDDMO. The results indicate that BMI explanatory power increased from 0.440 to 0.518. We then applied Cohen’s formula to compute the effect size (f2).

This value falls within the medium effect range (0.02 = small, 0.15–0.35 = medium, and >0.35 = large) (Cohen, 1988), indicating that DDDMO meaningfully improves the model’s predictive validity. The interaction slope further shows that when DDDMO is high (+1 SD), the relationship between AICFC and BMI strengthens, whereas it weakens when DDDMO is low (−1 SD) (see Fig. 3). This suggests that firms with strong data-driven orientations more effectively leverage coordination flexibility to drive BMI.

Fig. 3.

The moderating effect of DDDMO on the relationship between AICFC and BMI.

These findings imply that firms with a strong data-driven mindset are better positioned to capitalize on coordination flexibility for innovation. Given that DDDMO functions as a cultural and strategic moderator, China’s expanding data infrastructure and state-led AI policies may further amplify its effect. However, caution is warranted in generalizing these findings, and future research should examine this relationship in other digital economies, such as South Korea and the US.

Discussion and conclusion

This study contributes by explaining how specific managerial actions—internal advice-seeking, external advice-seeking, and collaborative sensemaking—serve as microfoundations of BMI through AI integration. Grounded in the DCV (Teece, 2018), the findings clarify how organizations progress through sensing, seizing, and transforming phases in digital environments.

The results show that DMCs are not homogeneous but consist of distinct components that influence BMI through different mechanisms (Durán & Aguado, 2022; Jorzik et al., 2024; Sjödin et al., 2021). IASC builds knowledge of internal routines and constraints, whereas EASC introduces external technological insights and opportunities (Hock-Doepgen et al., 2025). The stronger role of EASC in shaping AIIC highlights the importance of external networks, particularly in competitive digital contexts such as China, for advancing AI adoption (Gama & Magistretti, 2025; Liu et al., 2025). This supports the DCV perspective that environmental conditions shape the relevance of specific DMCs, reinforcing a context-sensitive approach to capability development (Teece, 2010; Teece et al., 1997).

CSC emerges as a critical yet underexplored dimension of sensemaking, linking interpretive alignment with strategic flexibility. Although prior research has examined CSC in relation to team cohesion or innovation framing (Akgün et al., 2012; Bellis & Verganti, 2021; Sheng, 2017), it is also connected to broader perspectives on collective cognition and absorptive capacity, which emphasize the updating of shared mental models in dynamic environments (Wang et al., 2026). CSC appears to amplify an organization’s ability to reframe assumptions and adapt to technological change (Drazin et al., 1999; Jantunen & Koivisto, 2016).

This study also advances understanding of AIIC and AICFC as distinct but interrelated dynamic capabilities (Madanaguli et al., 2024; Toth et al., 2025). Unlike absorptive capacity or traditional IT capabilities (Abou-Foul et al., 2023; Wamba-Taguimdje et al., 2020), AIIC focuses on embedding AI into organizational processes, while AICFC captures the dynamic reconfiguration of workflows and decision rights based on AI insights (Kayser & Gradtke, 2024; Zou & Yang, 2026). This distinction clarifies AI’s role within the DCV as both an enabler and outcome of capability development (Duong, 2025; Wamba-Taguimdje et al., 2020).

The findings further show that DMCs do not directly generate coordination agility but operate through AI integration to produce such outcomes (Abourokbah et al., 2023; Lee et al., 2021). AIIC enhances coordination flexibility, enabling more adaptive and responsive organizational processes that support BMI (Burström et al., 2021; Ghosh, 2025). The validated serial mediation model provides a structured view of how managerial capabilities translate into system-level transformation (Heubeck, 2024). At the same time, evidence of partial mediation suggests that capability development may occur through both direct and staged pathways, consistent with extensions of DCV theory that emphasize non-linear configurations (Eisenhardt & Martin, 2017)).

Finally, the moderating role of DDDMO highlights the contingent nature of dynamic capabilities. While AICFC contributes to BMI, its impact is amplified by a strong data-driven mindset (Teece, 2018; Tran-Dang et al., 2025). This illustrates the importance of cognitive and cultural infrastructure in enabling technology-driven capabilities (Kayser & Gradtke, 2024). Without such alignment, even well-integrated AI systems may fail to produce meaningful business model transformation (Jorzik et al., 2024; Sjödin et al., 2021).

Implications

Theoretical implications

This research contributes to the DCV by proposing a multilevel explanation of how DMCs (i.e., internal and external advice-seeking and collaborative sensemaking) serve as microfoundations that translate into firm-level outcomes such as BMI (Durán & Aguado, 2022). This addresses calls to examine the human origins of sensing, seizing, and transforming processes (Nicolai Foss & Mazzelli, 2025; Hossain et al., 2025; Kryeziu et al., 2024).

Second, this study contributes by empirically validating two types of dynamic capabilities: AIIC and AICFC (Al-khatib & Ramayah, 2025; Gao et al., 2025). While traditional IT capabilities emphasize system use and absorptive capacity focuses on knowledge acquisition, AIIC captures the integration of AI into strategic workflows. In contrast, AICFC extends coordination by enabling flexible task reconfiguration through AI analytics, representing a critical transformation layer within the DCV framework (Hartati et al., 2025; Rehman & Jajja, 2023).

Finally, introducing DDDMO as a moderator highlights the reinforcing role of cultural and cognitive infrastructures in strengthening the link between coordination flexibility and innovation (Tran-Dang et al., 2025). This extends the DCV by showing that capability deployment is both structural and conditional, depending on how firms interpret and use data.

Practical implications

Managerially, the findings suggest that firms pursuing AI-driven business model transformation must address the human factors that enable AI to create value. This is particularly important in dynamic environments where interpretation and alignment are critical. Firms should integrate AI capabilities across operations, ensuring that AI functions not only as a technical tool but also as a mechanism for organizational alignment. This requires establishing agile project structures and governance systems that support AI-enabled decision-making.

Moreover, organizations must cultivate a data-driven mindset at all levels. This extends beyond technical training to embedding decision-making practices grounded in evidence and analytical reasoning. While data-driven approaches enhance alignment and flexibility, firms must also remain attentive to risks associated with algorithmic bias, rigidity, and over-automation.

Limitations and future directions

Like any study, this research has limitations. First, as a cross-sectional design, it limits strong causal inference. Despite procedural and statistical controls for CMV, issues of simultaneity and endogeneity may persist. Future studies could address these concerns using time-lagged surveys, experimental or quasi-experimental designs, or longitudinal data. The use of instrumental variables or panel data could further strengthen causal identification and reduce endogeneity concerns.

Second, the focus on the Chinese context may limit generalizability, although it is appropriate given the rapid pace of AI-driven digitalization. China’s institutional features—including state-led AI strategies, digital transformation subsidies, and a dynamic market environment—may shape how capabilities are developed and applied. Future research could compare different institutional contexts (e.g., Germany, South Korea, and the US) to assess boundary conditions and examine cultural or policy-driven differences.

Third, while this study examines three core DMCs—internal advice-seeking, external advice-seeking, and collaborative sensemaking—other capabilities may also influence BMI. These may include digital literacy, stress management, and risk-taking, which could be conceptualized as socio-emotional or entrepreneurial dynamic capabilities relevant to AI-driven transformation.

Finally, future research should explore potential reverse causality, such as whether successful BMI enhances the development of DMCs or strengthens AI-related capabilities. This feedback loop remains underexplored but is critical for understanding organizational learning and adaptation following innovation. Additional research could also examine unexplored relationships, including the direct effect of DMCs on AICFC or the potential for AIIC to directly influence BMI under specific conditions (e.g., digital readiness). Potential moderators such as top management support, learning climate, and environmental uncertainty also warrant further investigation.

Funding

This study is supported by the 14th Five-Year Plan for National Business Education and Research: Exploration of integrating ideological and political education into higher vocational financial management courses based on the OBE concept [SKJYKT-220634], National Natural Science Foundation of China [72302120], and the Natural Science Foundation of Jiangsu Province [BK20230365]

Ethical statement

This research was conducted in full accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Declaration of Helsinki and its later amendments, or comparable ethical standards. The Ethics Committee of Nanjing University of Posts and Telecommunications granted ethical approval for this study under the approval number NUPT-ECSOM-2024–006. All subjects provided their informed consent before their inclusion in the study. Where applicable, confidentiality and anonymity were rigorously protected.

Informed consent

Informed consent was obtained from all participants and/or their legal guardians for participation in the study.

Data availability statement

Data will be made available at the request of the authors

CRediT authorship contribution statement

Ji Li: Writing – review & editing, Validation, Methodology, Formal analysis, Conceptualization. Haikuo Yu: Writing – original draft, Visualization, Investigation, Data curation. Hui Zhang: Writing – review & editing, Resources, Project administration, Methodology, Formal analysis.

Acknowledgement

The author used AI-assisted tools, including Grammarly, to support language editing and clarity improvements during the manuscript preparation process.