Leadership architectures for digital innovation: extending upper echelons theory through expert-coded evidence articulated abstract Open Access

Digital transformation has become a defining challenge for organizations, requiring leaders not only to adapt existing strategies but also to orchestrate innovation across technologies, functions, and markets. Firms are under pressure to continually integrate digital technologies into products, services, and processes, yet the ability to do so depends heavily on the configuration of top management leadership. While much is known about how organizational structures and strategies shape digital outcomes, the role of executive expertise and team dynamics remains less clearly specified (see Tables 3 and 5–7).

Upper Echelons Theory (UET) provides a powerful lens to examine this issue. UET posits that organizational outcomes are a reflection of the experiences, values, and cognitive frames of top managers (Hambrick and Mason, 1984; Hambrick, 2007). However, most empirical studies of UET continue to rely on demographic proxies such as age, tenure, or education, which offer only partial insight into the domain-specific expertise and integrative capacities needed to manage digital complexity. As recent reviews note, there is a need for context-sensitive measures that capture both substantive executive capabilities and the structural mechanisms through which diverse top management teams (TMTs) coordinate their efforts (Cortellazzo et al., 2019; Wesemann et al., 2024).

This study addresses that gap by focusing on chief executive officer (CEO) digital expertise (CDE), TMT diversity, and integration mechanisms as key elements of digital leadership architectures. Drawing on an expert-coded dataset of 12 firm-years from six global technology firms (Intel, NVIDIA, Huawei, Tencent, SAP, and ASML), we evaluate how these leadership attributes shape digital innovation outputs such as patents, product launches, and digital revenue share. Our approach allows us to move beyond demographic indicators and toward richer, evidence-based constructs that directly capture the cognitive and structural dimensions of digital leadership. As recent work emphasizes, context-specific studies provide unique theoretical leverage by uncovering dynamics that broader, generalized models often miss (Stremersch et al., 2023).

The contributions are threefold. First, we extend UET by demonstrating that domain-specific expertise and integration roles at the top of the firm condition the value of diversity for digital innovation. Second, we highlight a potential nonlinearity in this relationship: while CDE amplifies diversity’s benefits, excessive centralization may crowd out plurality and reduce innovative performance. Third, we introduce expert coding as a methodological contribution—a transparent, replicable approach for quantifying executive cognition and team structures when surveys or large-scale samples are infeasible.

In doing so, the study speaks to both theory and practice. For scholars, it refines UET by specifying the mechanisms through which leadership shapes digital outcomes. For practitioners, it highlights the importance of balancing CEO expertise with empowered integrators, and of coupling diversity with coordination, to ensure that leadership configurations enable—not hinder—digital transformation.

This review begins with UET, the dominant lens for understanding how leadership shapes organizational outcomes in digital contexts.

UET holds that organizational outcomes reflect the experiences, values, and cognitive bases of top executives (Hambrick and Mason, 1984; Hambrick, 2007). Since Hambrick and Mason’s formulation and Hambrick’s update, research has connected executive characteristics to strategic choices, innovation, and performance, while highlighting the role of managerial discretion in shaping observable outcomes (Carpenter et al., 2004; Finkelstein et al., 2009). A recurrent limitation, however, is the field’s reliance on demographic proxies (e.g. age, tenure, education) that say little about the domain-specific expertise now required to orchestrate digital transformation; recent syntheses urge context-sensitive measures of executive expertise and integrative team dynamics—particularly in turbulent digital environments where cross-functional cognition and coordination are pivotal (Bonelli, 2014; Cortellazzo et al., 2019; Yoo et al., 2012; Wesemann et al., 2024).

We use “digital” transformation to mean an enterprise-wide reconfiguration in which digital technologies reshape value creation, customer engagement, and operating models—going beyond isolated information technology (IT) or process digitization to coordinated, strategic change (Verhoef et al., 2021). In the same spirit, digital transformation is a process that enhances an entity by triggering significant changes to its properties through combinations of information, computing, communication, and connectivity technologies, clarifying that digital transformation (DT) is broader than single-function IT adoption (Vial, 2019). Consistent with this framing, successful transformation depends on the continual development of dynamic capabilities—sensing, seizing, and transforming—that enable strategic renewal under technological turbulence (Warner and Wäger, 2019).

Against that backdrop, “digital innovation” is iterative, boundary-spanning, and recombinative rather than linear and contained (Yoo et al., 2010; Lyytinen et al., 2016; Henfridsson et al., 2018). Innovation typically integrates technical and non-technical domains and continues well after initial deployment through updates and recombinations of components, data, and processes (Hanseth and Lyytinen, 2010; Baiyere et al., 2020). These properties elevate an information-processing challenge at the top of the firm: leaders must bridge silos, enable agility, and continually reframe strategy as technologies and ecosystems evolve (Tushman and Nadler, 1978; Volberda et al., 2021). Conceptual work links these demands directly to leadership configuration, emphasizing cross-functional collaboration, empowered integrative roles, and adaptive decision architectures that translate transformation intent into sustained digital innovation.

“Digital leadership” is increasingly framed as a distinctive capability that fuses technological literacy with strategic vision, coordination, and cultural change (Oberer and Erkollar, 2018; Zeike et al., 2019; Denning, 2020a, b). Empirical studies associate digital leadership with higher innovation outcomes and firm performance (Benitez et al., 2022; Henderikx and Stoffers, 2023), and information-systems research highlights the C-suite’s role—especially chief information officer (CIO)/chief digital officer (CDO)/chief digital officer (CTO) positions—in translating intent into execution (Chen et al., 2021; Firk et al., 2022; Bendig et al., 2023; Garms and Engelen, 2019). Recent work emphasizes the substantive—rather than solely demographic—nature of CDE, showing that CEOs with domain knowledge are better positioned to evaluate technological trajectories and guide transformation (Hambrick, 2007; Busenbark et al., 2016). Related evidence in SMEs indicates that digital leadership at the helm is linked to innovative performance and competitive advantage in resource-constrained contexts, underscoring generalizability beyond large tech firms (Zeike et al., 2019; Denning, 2020a, b).

Within UET, cognitive diversity in the TMT broadens information search, improves sensemaking, and enhances innovative problem solving (Bunderson and Sutcliffe, 2002; Certo et al., 2006; Carpenter et al., 2004). Heterogeneity across functions and demographic backgrounds tends to strengthen creative recombination and responsiveness under uncertainty—conditions that typify digital innovation (Dezsö and Ross, 2012; Buyl et al., 2011). Conceptual syntheses build a framework in which TMT plurality (size, functional heterogeneity, gender diversity) expands the team’s information-processing capacity needed to manage digital complexity, and empirical work links plurality to outcomes such as patenting, digital product launches, and process digitization (Bendig et al., 2023; Wesemann et al., 2024).

Diversity expands information processing but also raises coordination costs and the risk of subgroup fault-lines. To convert plurality into performance, firms need integration mechanisms that channel dispersed expertise into coherent action. In our framing, three elements do the work: (1) empowered boundary roles at the top table (CIO/CDO/CTO with clear authority and budget), (2) cross-functional councils with decision rights, cadence, and metrics, and (3) routinized ecosystem practices (e.g. standards, partner forums, chapters/guilds) that bridge technical and market domains. This emphasis on organization design parameters and empowered digital leadership roles is consistent with evidence that chief digital officers shape transformation outcomes through structural configuration and coordination authority (Singh, 2020). These mechanisms reduce translation costs, align priorities, and keep dissent productive rather than paralyzing. In UET terms, they supply the structural glue that enables a diverse TMT to realize its cognitive advantages without being undone by coordination frictions. Accordingly, our hypotheses treat integration as a mediating channel from TMT plurality to digital innovation (H3) and as a boundary condition that strengthens the CEO-expertise–diversity interplay (H4): when integrators are genuinely empowered, plurality is more likely to register as sustained digital outcomes rather than noise.

UET also cautions that concentrated power and overconfidence can reduce the benefits of plurality if the CEO anchors decisions too tightly to a single cognitive frame (Hambrick, 2007; Busenbark et al., 2016). Evidence in digital contexts shows precisely such nonlinearity: CDE strengthens the diversity–innovation link up to a point, beyond which centralization attenuates diversity’s contribution. The upshot is to balance CEO authority with distributed decision-making to preserve the value of TMT plurality.

Across these streams: (1) diverse TMTs process digital complexity more effectively; (2) CDE is valuable but must be balanced; and (3) integration roles translate plurality into digital innovation outcomes (DIO) (Carpenter et al., 2004; Firk et al., 2022; Bendig et al., 2023). Evidence from SMEs suggests this architecture travels beyond Big Tech (Zeike et al., 2019).

UET and information-processing views imply that diverse leadership teams—across functional, experiential, and gender lines—bring broader search, richer interpretation, and more robust solution generation, which are especially crucial for iterative, recombinative digital innovation (Bunderson and Sutcliffe, 2002; Certo et al., 2006; Yoo et al., 2010). Empirical work links TMT diversity to multiple innovation indicators across global tech firms (Bendig et al., 2023; Wesemann et al., 2024).

Subject-matter expertise at the apex helps evaluate technologies, set priorities, and orchestrate resources, thereby amplifying the benefits of TMT plurality (Benitez et al., 2022; Chen et al., 2021). Yet UET warns that excessive centralization can crowd out dissenting views and reduce the returns to diversity (Hambrick, 2007; Busenbark et al., 2016). Evidence supports an inverted-U moderation—moderate CEO expertise strengthens the diversity–innovation link, whereas very high expertise can attenuate it.

Diversity provides raw material; empowered integrators (e.g. CIO/CDO/CTO), cross-functional routines, and decision protocols are the machinery that converts plurality into action (Garms and Engelen, 2019; Firk et al., 2022; Bendig et al., 2023). Findings show partial mediation through integration roles, consistent with governance routines that operationalize this mechanism.

Where technology-oriented executives hold influence at the top table, CEO expertise and TMT plurality are more effectively combined: information is brokered across domains, translation costs fall, and alignment rises (Firk et al., 2022; Bendig et al., 2023). This boundary condition is reinforced by evidence that leadership-enabled integration translates digital intent into innovation and advantage across contexts (Zeike et al., 2019).

This section outlines the sample, expert-coding protocol, construct measures, reliability checks, and empirical strategy.

We assembled a 12–firm-year panel from six global technology companies—Intel, NVIDIA, Huawei, Tencent, SAP, and ASML—observed in 2023–2024 (two observations per firm). This scope gives cross-firm variation in leadership architectures while holding industry context relatively constant across a period of intense digital investment.

Three independent senior experts coded leadership constructs using a structured rubric and common evidence base:

1. Coder A: current CEO of a Shenzhen software firm.

2. Coder B: retired U.S. semiconductor CEO.

3. Coder C: European scholar familiar with the UET.

Each coder worked independently and entered scores in standardized sheets that also captured verbatim evidence anchors (URLs/excerpts) for auditability. Coders evaluated each firm-year on CDE, TMT plurality, and integration mechanisms (details below). We used coder-means to aggregate item scores and reduce idiosyncratic rater noise.

We define and justify each construct below—detailing item rubrics, scaling (1–4 unless noted), aggregation, and summary statistics—before turning to the control variables.

We operationalize CDE as a five-item rubric scored on a 1–4 scale (higher values denote more of the attribute). The items cover (1) human-capital depth in digital domains, (2) a demonstrated track record in leading digital initiatives, (3) boundary-spanning across technology–business ecosystems, (4) digital risk governance, and (5) the degree of centralization of digital decision rights (reverse-coded so that higher values consistently indicate greater expertise). The CDE index is the coder mean across items; in our sample its distribution centers at M = 2.65 with Standard Deviation (SD) = 0.11 (scale 1–4).

Top-management plurality is captured with three observable components: team size, functional heterogeneity, and gender diversity. Functional heterogeneity is computed via a Blau index across executive functional backgrounds, while gender diversity is the female share among TMT members. Each component is standardized, and their average forms the TMT Diversity Index, which by construction has a mean near zero (SD ≈ 0.61).

Integration reflects the structures that translate plurality into coordinated action. We assess the empowerment of integration roles (CIO/CDO/CTO), the presence and authority of a cross-functional integration council, and the routinization of ecosystem partnering. Each element is scored from 1 to 4, and their average yields the Integration Index (M = 2.28, SD = 0.37).

Innovation performance is tracked with four firm-year indicators: counts of digital patent families and digital product releases, the percentage of revenue attributable to digital offerings, and a 1–4 milestone score for process digitization. For analysis, we synthesize a DIO composite as the mean of standardized components—specifically: z[ln(1 + patents)], z[product releases], and z[digital revenue share].

Because the process-digitization score is constant at 3 across observations, we omit it from the composite to preserve variance. By construction, the composite has M ≈ 0 and SD ≈ 0.50.

All regressions include standard firm-year controls merged from public filings: firm size (log employees), firm age (years since founding), R&D intensity (R&D-to-sales), and year fixed effects (FE) for 2023 and 2024.

We assessed reliability using Krippendorff’s α (interval) for numeric items and quadratic-weighted Cohen’s κ for ordinal items, averaging κ across the three coder pairs. As anticipated, agreement was highest on objective indicators and more modest on judgment-intensive assessments. For the objective items, TMT size and functional heterogeneity showed perfect agreement (α = 1.00 for both), whereas gender share was lower (α = 0.39). For the CDE rubric (CDE1–CDE5), agreement at the item level was moderate (mean κ = 0.53; item means: 0.54, 0.56, 0.55, 0.33, 0.67). Integration-mechanism items were similar (mean κ = 0.52; roles 0.48; council 0.54; ecosystem 0.54). When the five CDE items are combined into an interval-scaled index, internal consistency is weak (α ≈ 0.11; Appendix B), consistent with heterogeneous implicit weightings across sub-items; accordingly, we rely on item-level κ and coder means. Accordingly, we treat the item-level κ statistics as the primary reliability evidence and aggregate by coder means at the item level rather than privileging the composite’s α. Given the small N and the mixture of ordinal and interval indicators, we rely on coder-averaged scores for all constructed leadership variables; the objective measures are identical across coders (i.e. 100% agreement).

Across 12 firm-years, firms average 42.8 digital patent families (SD = 31.5), 7.6 digital product releases (SD = 1.7), and derive 38.1% of revenue from digital offerings (SD = 35.3). The CDE index aligns strongly with patenting (r ≈ 0.78) and moderately with top-team integration (r ≈ 0.59), while integration relates positively to digital revenue share (r ≈ 0.67). Together, these patterns are consistent with the idea that substantive CEO expertise and integrative structures travel with innovation outputs, though with N = 12 they are best treated as descriptive associations rather than causal effects. Full descriptives and the correlation matrix appear in Table 1 and Appendix C; summary statistics for the core variables are reported in Table 2.

Our empirical strategy proceeds in four steps aligned with H1–H4. For H1 we estimate firm-level models using coder-averaged data (N = 12 firm-years) with firm FE and year FE. For H2–H4 we use a coder-stacked specification that preserves coder variance (N = 36 = 6 firms × 2 years × 3 coders) with coder FE and year FE. Firm FE are omitted in the stacked models because outcomes are identical across coders within a firm-year, which would otherwise saturate the model.

First, to assess the main effect of team plurality (H1), we model DIO for firm i in year t as a function of top-management diversity, standard controls, and fixed effects:

where μi captures time-invariant firm heterogeneity, τ_t absorbs year shocks, and β′Xit ​ includes firm size, firm age, R&D intensity. This baseline quantifies whether more diverse top teams are associated with stronger digital innovation, net of observables and unobserved firm constants.

Second, to test whether CDE conditions the diversity–innovation link in an inverted-U fashion (H2), we augment the model with both the linear and squared CEO terms and their interactions with diversity:

Evidence of an inverted-U moderation is the joint pattern β₃ > 0 and β₅​<0, indicating that CEO expertise initially amplifies the returns to diversity but that excessive centralization ultimately dampens them. We report the turning point of CEO expertise for the diversity slope as −β₃/(2β₅) with a confidence interval and confirm it lies within the observed CDE range.

Third, we evaluate integration as a mediating mechanism (H3) using a two-equation system and bootstrap inference for the indirect path:

This structure tests whether empowered integrative roles and routines translate cognitive plurality into concrete innovation outputs.

Finally, we probe a boundary condition (H4) by estimating a three-way interaction that asks whether the CEO-based moderation itself is stronger when integrators are more empowered:

A positive θ₃ implies that CIO/CDO/CTO authority at the top table strengthens the beneficial interplay between CEO expertise and team diversity.

All continuous predictors are mean-centered prior to forming interactions. Patent counts enter the DIO composite as In(1+patents). Interaction terms use the centered indices Div₍c₎, CDE₍c₎, Int₍c₎; nonlinear terms such as CDE₍c₎² are constructed from these centered variables. Unless otherwise noted, standard errors are HC3. For coder-stacked models we additionally report wild-cluster bootstrap p-values by firm (6 clusters; Rademacher weights; B = 2,000). We run leave-one-firm influence checks and summarize maximum Cook’s D and any sign flips in Appendix D. Given the small N (H1: 12 firm-years; H2–H4: 36 coder-stacked rows), inference emphasizes effect directions and pattern consistency over precise magnitudes. Estimation is implemented via standard ordinary least squares (OLS) with FE; de-identified materials and code sufficient to reproduce all analyses are publicly archived (Bonelli, 2025).

Results for H1 use the firm-level design (§4.3); results for H2–H4 use the coder-stacked design (§§4.4–4.6); diagnostics appear in Appendix D.

We designed the study to be fully auditable. Every coder worksheet preserves the evidence trail (URLs and verbatim excerpts) used to justify each score, enabling independent verification of coding decisions. The analysis pipeline aggregates coder-level inputs into firm-year constructs via a transparent coder-mean rule; outcome variables draw on objective sources and, by design, are identical across coders in the raw files. To facilitate replication, all coder-level datasets and accompanying materials are publicly deposited on Mendeley Data, DOI: 10.17632/gy58pzjgmp.1. The deposited archive, together with the evidence anchors embedded in the coding sheets, provides a complete audit trail from raw sources to final firm-year measures.

This section presents the empirical findings, beginning with reliability and measurement diagnostics, followed by descriptive patterns and the hypothesis-driven analyses.

Agreement on objective team attributes was extremely high. TMT size and functional heterogeneity reached perfect reliability (Krippendorff’s α = 1.00 for both), whereas gender diversity was lower (α = 0.39). For judgment-intensive constructs, coder-pair agreement was moderate: the five CDE items averaged κ ≈ 0.53 and the three integration items averaged κ ≈ 0.52. When the CDE items are combined into a single index, internal consistency is weak (Appendix B: α ≈ 0.11; 95% CI [0.06, 0.13]), reinforcing our choice to rely on item-level κ and coder-means for constructed variables. By contrast, the TMT Diversity composite shows strong internal consistency (α ≈ 0.89; 95% CI [0.77, 0.95]). Disagreement profiles indicate that coder B applied systematically more conservative thresholds, while coders A and C were closely aligned.

See Appendix A for coding instruments and protocol, and Appendix B (Table B1) for the full reliability statistics and disagreement profiles.

Across 12 firm-years (Intel, NVIDIA, Huawei, Tencent, SAP, ASML; 2023–2024), firms average 42.8 digital patent families (SD = 31.5), 7.6 digital product releases (SD = 1.7), and 38.1% digital revenue share (SD = 35.3). The CDE index centers at 2.65 (SD = 0.11); the Integration index averages 2.28 (SD = 0.37). Correlations align with the study’s conceptual architecture: CDE correlates strongly with patents (r ≈ 0.78; 95% CI [0.38, 0.93]) and moderately with integration (r ≈ 0.59; 95% CI [0.15, 0.84]); integration correlates positively with digital revenue share (r ≈ 0.67; 95% CI [0.40, 0.89]). TMT diversity is positively associated with integration (r ≈ 0.51) and with product releases (r ≈ 0.56). Given N = 12, these are descriptive patterns rather than definitive estimates, but they are directionally consistent with UET expectations about expertise, plurality, and integrative structures.

Together, these descriptive associations establish the empirical groundwork for testing H1–H4; extended descriptive statistics (Table C1) and the full correlation matrix with bootstrapped confidence intervals (Table C2) are reported in Appendix C.

We estimated OLS models with firm FE and year dummies; continuous predictors were mean-centered (Methods §3.6). Sensitivity results—varying the aggregation of the diversity index, reporting robust (heteroskedasticity-consistent estimator type 1 (HC1)) standard errors, and running leave-one-out checks—are summarized below and in Appendix D. Diversity is positively associated with product releases and, imprecisely, with digital revenue share; it is negatively associated with patents (log-transformed), conditional on CDE and integration. With N = 12, we emphasize effect directions rather than significance. The baseline, alternative aggregation, and robustness estimates are reported in Table 3.

Unless noted, H2–H4 estimates use the coder-stacked specification described in 3.6 (coder and year FE; HC3; wild cluster bootstrap by firm).

Estimating DIO on Diversity, CDE, CDE², and their interactions with coder and year FE yields the expected inverted-U moderation pattern, albeit imprecisely estimated in this sample. The Diversity × CDE and Diversity × CDE² coefficients are both negative with wild cluster p-values ≈0.50 (6 firm clusters; Rademacher weights; B = 2,000). The implied turning point of CDE for the Diversity slope is −0.239 in centered units (95% wild-bootstrap CI [−0.428, 0.210]), which lies within the observed range. Figure 1 plots predicted DIO across centered CDE at low (−1 SD) vs. high (+1 SD) Diversity with 95% HC3 intervals. We interpret H2 as pattern-consistent but not confirmed.

Turning point of CDE for the Diversity slope: −0.239 (95% CI [−0.428, 0.210]).

Figure 1 illustrates the predicted inverted-U moderation of CDE on the diversity–innovation relationship, showing differences at low vs. high TMT diversity.

Results support the mechanism descriptively. The estimated path coefficients and indirect effects are presented in Table 5. Diversity predicts Integration (a = 0.274, HC3 SE 0.101; t = 2.71), and Integration predicts DIO controlling for Diversity (b = 0.958, HC3 SE 0.101; t = 9.46). The indirect effect is a × b = 0.262, with a cluster-bootstrap 95% CI [−0.203, 0.789] (B = 2,000), which crosses zero. We therefore characterize H3 as descriptively supportive but statistically uncertain at this N.

Indirect effect (a × b): 0.262; 95% CI [−0.203, 0.789] (cluster bootstrap, B = 2,000).

The three-way Div × CDE × Integration term is negative and imprecise (θ₃ = −0.776, HC3 SE 0.811; wild-cluster p 0.485). Simple slopes of Diversity show attenuation when CDE and Integration are both high (slope ≈ 0.019; wild-cluster p ≈ 0.908), versus positive slopes in the other three cells. The full interaction model is reported in Table 6, and conditional simple slopes are summarized in Table 7. We treat H4 as suggestive (boundary-consistent) but not decisive. (See Appendix D for influence checks; notably, the three-way term’s sign does not flip under leave-one-firm deletion.)

Figure 2 extends this pattern by conditioning on integration, plotting predicted DIO across combinations of low/high CDE, diversity, and integrative structures.

Given N = 36 coder-stacked observations (6 firms × 2 years × 3 coders), we report HC3 standard errors and wild cluster bootstrap p-values by firm and emphasize effect directions over magnitudes. Wild bootstrap used Rademacher weights, B = 2,000.

This study advances UET by showing that leadership influences digital innovation not only through demographic proxies but also through substantive expertise and the integration of top-management processes. The findings indicate that while diversity in TMTs provides raw cognitive variety, its positive effects on innovation are contingent on the expertise of the CEO and the presence of integrative structures.

First, our results extend UET by moving beyond traditional demographic indicators to capture domain-specific leadership qualities. CDE emerges as a substantive capability that conditions how TMT diversity translates into digital outcomes. Importantly, the findings highlight a nonlinearity: expertise enhances diversity’s benefits up to a point, but excessive centralization by a highly expert CEO can stifle the very plurality that diversity brings. This supports a more nuanced view of UET in digital contexts, emphasizing that leadership must balance authority with openness to distributed inputs. Integration roles—CIO, CDO, CTO, and cross-functional councils—play a mediating role in this balance, translating diversity into coordinated action.

A further contribution lies in methodology. By employing expert coding, the study demonstrates a transparent and replicable alternative to conventional survey measures of executive cognition. The coding protocol, triangulation rules, and reliability checks show that qualitative evidence can be systematically transformed into quantitative constructs suitable for comparative analysis. This approach complements traditional survey and archival designs, offering a viable path for studying leadership phenomena in data-scarce or high-discretion domains such as digital transformation.

Limits of diversity without integration: diversity can raise coordination costs and create fault-lines; without integrative roles and routines, its upside is muted. The practical implication is balanced authority + empowered integrators—ensure CEO digital fluency is complemented by CIO/CDO/CTO authority and cross-functional councils so plurality is translated into coordinated innovation.

For practitioners, the results caution against both under- and over-reliance on CEO expertise. Boards and investors should recognize that CEO digital fluency is valuable but must be balanced by empowering integrators who distribute decision-making and translate strategic intent into execution. Diversity in the top team alone is not sufficient; it must be coupled with integration mechanisms that convert cognitive variety into innovation. Organizations that invest simultaneously in leadership expertise, team plurality, and integrative structures are more likely to achieve sustained digital performance.

Taken together, these findings contribute to theory by refining UET with a capability-and-architecture perspective on digital leadership; to method by validating expert coding as a transparent, replicable measurement strategy; and to practice by specifying how boards can balance CEO expertise with empowered integrators to harness—not mute—the benefits of TMT diversity. In short, digital innovation is strongest when substantive expertise, plurality, and integration are jointly designed rather than pursued in isolation.

In sum, we find partial support for H1, while H2–H4 are pattern-consistent under small-sample uncertainty; formal tests are reported but interpreted cautiously. Examining leadership configurations through the UET lens, we show that leadership matters for digital transformation, but its effects are conditional. TMT diversity relates positively to product launches and, more imprecisely, to digital revenue share, yet is negatively associated with patenting once CDE and integration are controlled. CDE correlates strongly with innovation outputs and integration—consistent with an amplifying role—even though inverted-U moderation tests are inconclusive at this N. Diversity predicts integration, which in turn is associated with innovation outputs, but full mediation awaits larger samples. The hypothesized stronger moderation under empowered integrators is directionally consistent but not adjudicated. Overall, expertise and integration emerge as the channels through which TMT diversity is translated into digital innovation.

The study makes three core contributions. Theoretically, it extends UET by moving beyond demographic proxies to highlight CDE and integrative structures as key drivers of innovation in digital contexts. Methodologically, it demonstrates the feasibility of expert coding as a transparent, replicable alternative to surveys, offering a viable strategy for measuring executive cognition when archival data are limited. Practically, it cautions boards against both under- and over-reliance on CEO expertise: digital fluency at the top is valuable, but only when balanced by empowered integrators who prevent excessive centralization and translate diversity into coordinated action.

At the same time, the study has limitations. It is based on a small sample of 12 firm-years concentrated in global technology firms, and its reliance on manual expert coding constrains generalizability. Inferential power is limited, and the findings should be viewed as pattern-consistent rather than definitive.

Future research should address these limitations by extending the dataset across industries and countries, scaling up the coding approach, and integrating automated techniques such as natural language processing and AI-based text analysis to complement expert judgment. Such advances would allow the field to test these hypotheses on a larger scale and refine understanding of how leadership architectures drive digital transformation.

In sum, the evidence suggests that digital innovation is strongest when CEO expertise, team diversity, and integration mechanisms operate together. Leadership in the digital era is not a matter of any single attribute but of carefully balancing authority, plurality, and coordination.

The supplementary material for this article can be found online.