AI adoption readiness among Ukrainian education managers: Barriers, typologies, and policy implications

Synthesis: Kremen and colleagues (2026) surveyed 395 education managers across 23 of Ukraine's 24 administrative regions in September 2025 to measure AI adoption readiness in a post-Soviet transitional setting. Personal readiness (M=3.83, SD=0.96) significantly exceeded system readiness (M=3.15, SD=0.93) — a 0.68-point gap (d=0.73) — with regulatory absence (58.5%) and digital competency gaps (53.9%) the most prevalent barriers. Latent Class Analysis identified six provisional manager typologies, the Competency-constrained class the largest at 25.6%, and fuzzy-set qualitative comparative analysis found no barrier necessary and no configuration sufficient for low readiness. The authors argue for differentiated, typology-based training and stronger regulatory frameworks.

Key Findings

1. A 0.68-point personal–system readiness gap. Personal readiness (M=3.83, SD=0.96) exceeded system readiness (M=3.15, SD=0.93), confirmed by a Wilcoxon signed-rank test (W=3538, p<.001, rank-biserial r=0.77, d=0.73). Top-2 box scores were 67.3% versus 32.9%.
2. Systemic barriers outweigh attitudinal ones. Regulatory absence was most prevalent (58.5%), then digital competency gaps (53.9%), AI distrust (44.8%), and resistance to change (24.6%).
3. Six provisional manager typologies. Competency-constrained (25.6%, the largest), Personalization champions (21.0%), Regulation-blocked (18.2%), Change-resistant automators (15.7%), Barrier-free skeptics (10.6%), Doubly-constrained (8.9%).
4. Low readiness is diffusely, not configurationally, determined. No barrier reached the 0.90 necessity threshold — low system readiness reached 0.84, resistance to change 0.24 — and none of the 16 four-barrier configurations met the 0.80 sufficiency threshold (highest consistency 0.33).
5. ChatGPT was the most widely adopted tool, at 69.4%, ahead of Gemini and Canva — familiarity with conversational interfaces rather than specialized educational AI tools.
6. Managers see AI as an analytics and automation tool. Perceived potential was highest for data analytics (61.0%), document automation (58.0%), and quality monitoring (53.4%), and lowest for personalization (29.4%).
7. Digital competency is the largest readiness shortfall, trailing its 0.80 target by 42.4%, against 21.3% for system readiness and 4.2% for personal readiness.

Managers as the Gatekeepers of AI Adoption

The paper starts from a premise the AI-in-education governance literature often underplays: education managers decide whether and how AI enters daily practice, so their readiness is a precondition for evidence-based policy rather than a peripheral staff concern. Ukraine matters as a post-Soviet transitional setting where regulatory infrastructure for AI is still embryonic; wartime conditions further raise the salience of digital tools while damaging the infrastructure institutions need to adopt them. Managers at all system levels — principals, vice-principals, regional administration department heads, institutional administrators — were recruited through regional education departments and professional associations.

The Personal–Institutional Readiness Gap

The central measurement finding is that the two layers diverge sharply: 3.83 versus 3.15, a 0.68-point difference with a large effect size (d=0.73). The authors read the asymmetry as structural rather than motivational — a post-Soviet administrative legacy concentrates authority and resources above the institutional level, so managers acquire readiness faster than their institutions acquire infrastructure, budgets, and AI governance. Readiness was measured as two separate single-item constructs (Q25 and Q7, Spearman ρ=0.39) so a composite could not mask the very gap of interest. Person-centered Latent Class Analysis then showed that "the manager" is a fiction: the six typologies range from the Competency-constrained (willing but unskilled, 25.6%) to Barrier-free skeptics (highest readiness at 3.52, yet highest AI distrust at 74%).

What Managers See AI Doing

Because no single barrier was necessary or sufficient for low readiness, there is no lone fix to administer, and the authors argue for simultaneous action on regulation, institutional capacity, and training tuned to each typology. Tool use concentrated on conversational interfaces, and the application areas managers prioritized — analytics, document automation, quality monitoring — locate AI within data-driven decision-making, where managerial readiness gates adoption of the analytics that later inform teaching and assessment. The authors flag the typology-differentiated tailoring as the provisional part: regulatory and capacity-building priorities rest on firmer prevalence and gap findings.

What this means for practice

• Administrators. Measure personal and institutional AI readiness as separate constructs — the 0.68-point gap is the actionable signal, and leaders who are personally ready may still be institutionally blocked.
• Administrators and policymakers. Treat regulatory absence (58.5%) and digital-competency gaps (53.9%) as the binding constraints and build AI Governance frameworks alongside capacity-building, since no single barrier explains low readiness.
• Policymakers. Differentiate professional development by manager typology rather than offering one generic AI training — Competency-constrained managers need skills building, Regulation-blocked managers need governance scaffolding, Barrier-free skeptics need evidence-based trust-building.
• Researchers. Treat the six typologies and the fsQCA null as exploratory: bootstrap stability was poor (mean ARI = 0.385) and the structural model could not be estimated, so larger-sample replication comes first.

Limitations

• The cross-sectional design precludes causal inference, and readiness is self-reported, so managers may overstate personal readiness given the policy salience of AI and their leadership role.
• Each readiness dimension was measured with a single item, and the four barriers as single binary yes/no items, so the facets of these constructs cannot be recovered and confirmatory factor analysis and structural equation modeling could not be estimated due to convergence difficulties.
• The six-class LCA solution is exploratory on the study's own evidence: bootstrap stability was poor (mean ARI = 0.385) and BIC fell monotonically across the 2–6 class range without a clear minimum.
• The sample represents 23 regions but may miss smaller rural institutions, and wartime conditions may have inflated or depressed readiness reports in ways the measures do not capture.

Connected Concepts

• AI Governance
• AI Regulation in Education
• Educational AI Policy
• Generative AI
• AI Literacy
• Technology Adoption Models
• AI in Education

Connected Articles

• Anticipatory governance and leadership for AI implementation in higher education: A scoping review — Anticipatory governance and leadership for AI implementation (scoping review)
• Policy Fragmentation or Institutional Alignment? Institutional Governance of AI in Universities and Business Schools — Institutional governance of AI in universities and business schools
• Governing generative AI in higher education: Emerging policy approaches and support ecosystems at innovative U.S. universities — Governing generative AI in higher education: policy approaches at US universities
• Generative Artificial Intelligence Policy: A Qualitative UNESCO Framework Analysis — Generative AI policy: a qualitative UNESCO framework analysis

Citation

Kremen, V. G., Spirin, O. M., Liashenko, O. I., Lytvynova, S. H., Malovanyi, Y. I., Pinchuk, O. P., Sokolyuk, O. M., & Semerikov, S. O. (2026). AI adoption readiness among Ukrainian education managers: Barriers, typologies, and policy implications. Computers and Education: Artificial Intelligence, 11, 100648.