Uncovering Hierarchical Asymmetries in Artificial Intelligence Transformation: Navigating the Bright and Dark Sides Across Organizational Levels

Summary

We describe a replicable survey methodology designed to probe hierarchical asymmetries in artificial intelligence (AI) transformation (AX) perceptions. Both executives and practitioners complete a validated questionnaire hosted on Qualtrics; the analytical pipeline runs in SmartPLS 4.0, covering outer-model validation, path analysis, MICOM invariance testing, and permutation-based multi-group analysis (MGA).

Abstract

Governance structures that align executive strategy with frontline implementation depend on understanding how organizational hierarchy shapes perceptions of artificial intelligence (AI) transformation (AX) success factors. We present a validated, replicable survey-based methodology coupling partial least squares structural equation modeling (PLS-SEM) with multi-group analysis (MGA) to detect and quantify hierarchical asymmetries in AX perceptions—going beyond aggregate analysis to expose structural divides that single-group methods leave hidden. Drawing on 293 professionals involved in AI projects, we assess four dynamic capability dimensions—operational agility, data readiness, customer proximity, and strategic process discipline—together with key enabling factors: technological support, AI-sensitive risk tolerance, environmental context, and proactive leadership. Compositional invariance is first verified through MICOM testing; only after confirming between-group comparability does the analysis move to permutation-based MGA significance tests. Both groups exhibit positive associations between dynamic capabilities and AX performance (executives: β = 0.637, R² = 0.406; practitioners: β = 0.531, R² = 0.282). Permutation-based MGA detects no statistically significant differences in structural path coefficients between the two groups (all p ≥ .178); the hypothesized moderating role of organizational position on path relationships (H5) therefore lacks support. Yet MICOM Step 3a shows executives rating all major constructs significantly higher than practitioners do (8 of 11 constructs, all p ≤ .045)—evidence that hierarchical asymmetry surfaces as a gap in perceptual levels, not in the structural relationships themselves. As a transferable methodological template, the protocol equips researchers studying perception gaps in organizational transformation with tools bearing directly on AI governance, change management strategy, and cross-level alignment in technology-intensive settings.

Introduction

Rapid advances in artificial intelligence have opened a dual landscape for business organizations—extraordinary value-creation opportunities on one side, significant risks and challenges on the other^1,2. Generative AI and conventional AI technologies hold out the promise of higher productivity, faster innovation, and stronger competitive positioning³, yet they simultaneously invite concerns around bias, job displacement, inequality, and ethical dilemmas^4,5. This bright-side/dark-side duality calls for careful scrutiny of h....

Protocol

All procedures complied with the ethical standards for human-participant research at aSSIST University, the University of Suwon, Sungkyunwakn University, and Dong-A University. Because every response was anonymized at the point of collection and no personally identifiable data was gathered at any stage, the Institutional Review Board at aSSIST University granted the study an exemption from full review. Co-author institutions relied on this lead-institution approval rather than pursuing independent IRB reviews; aSSIST University had sanctioned both the study design and the data-collection protocol. Within the Qualtrics platform, each participant encountered a digital i....

Results

As a preliminary assessment of common method bias (CMB), Harman’s single-factor test was conducted, with the first unrotated factor accounting for 30.21% of variance (below the 50% threshold). While this result provides initial evidence that CMB does not dominate the data, it is acknowledged that Harman’s test has limited sensitivity and specificity as a diagnostic tool. As additional procedural safeguards, the survey employed randomized item ordering, included reverse-coded items, and guaranteed respondent anonymity to .......

Discussion

The findings reveal that organizational hierarchy is associated with systematic differences in the level at which executives and practitioners perceive AI transformation constructs, rather than in the structure of relationships among them. Three independent MGA approaches—permutation-based MGA (1,000 permutations), Bootstrap MGA (5,000 subsamples)⁷⁴, and Parametric and Welch-Satterthwaite tests⁶⁶—consistently found no statistically significant differences in str.......

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Prolific (Online Participant Recruitment Platform)	Prolific Academic Ltd	https://www.prolific.com/	Online platform used to recruit survey participants. Participants were screened based on eligibility criteria (e.g., current employment in an organization with AI adoption experience) and compensated according to Prolific's fair-pay guidelines.
Qualtrics (Online Survey Platform)	Qualtrics, LLC	https://www.qualtrics.com/	Web-based survey platform used for questionnaire design, distribution, and response data collection. All measurement items were administered via Qualtrics using a 7-point Likert scale (1 = strongly disagree, 7 = strongly agree).
SmartPLS 4.0 (PLS-SEM Software)	SmartPLS GmbH	https://smartpls.com/	Used for all Partial Least Squares Structural Equation Modeling (PLS-SEM) analyses, including outer model assessment (composite reliability, AVE, HTMT), inner model path analysis (bootstrapping, 5,000 samples), MICOM 3-step compositional invariance testing (permutation, 1,000 draws), and PLS-MGA between-group difference tests.