Among Slovak firms engaged with AI, hiring tied to AI is reported about twice as often as layoffs, but many firms report both simultaneously — indicating workforce reconfiguration rather than simple net job loss.

Main Finding

Among 693 AI-engaged Slovak firms, reported or planned AI-related job creation was significantly more common than reported or planned elimination (18.0% vs. 9.2%), but creation and elimination co-occurred within firms far more often than expected by chance (odds ratio 7.23; bivariate-probit correlation ρ = 0.54). More advanced AI deployment (adoption stage) predicts reporting creation, while employee job‑threat concerns predict reporting elimination; foreign ownership predicts joint occurrence only in an exploratory model. Results are associational and concern incidence (realized or planned events), not counts or net employment change.

Key Points

• Sample: 693 AI-engaged firms in Slovakia (351 current users, 211 pilots, 131 planning adoption).
• Outcomes: two binary indicators combining realized and planned/expected events:
  • Creation: introduced or planning to introduce AI-related positions.
  • Elimination: AI has led or is expected to lead to layoffs/position cuts.
• Frequencies:
  • Creation reported/planned: 18.0%
  • Elimination reported/planned: 9.2%
  • Co-occurrence (both outcomes) markedly above chance (OR = 7.23).
• Multivariate associations:
  • Adoption stage (more advanced deployment) → higher odds of reporting creation (OR 1.49; FDR-adjusted p = 0.004).
  • Employee job-threat concern → higher odds of reporting elimination (OR 1.58; p = 0.014).
  • Foreign ownership → higher relative risk of joint occurrence (Reconfigurer category) in an exploratory multinomial model (RRR 2.29; unadjusted p = 0.027), but this was not a primary, FDR‑adjusted finding.
• Hypotheses tested included that creation > elimination (supported), positive within‑firm association (supported), adoption stage predicts incidence (supported for creation), orientation/maturity predict elimination (mixed: job-threat concern predicted elimination; self-assessed maturity did not show the same association), and foreign ownership predicts reconfiguration (exploratory support).
• Important measurement caveats: outcomes merge realized and planned/expected events; they indicate incidence (whether a firm reports/anticipates the event), not number of jobs or net employment change.
• Typology: four firm types by incidence pattern — No-change, Creators (creation-only), Eliminators (elimination-only), Reconfigurers (both).

Data & Methods

• Data source: AI-ImpactSK organization-level survey (Černý et al., 2026); open-license dataset. Initial file 816 firms; analytical sample restricted to 693 firms that use, pilot, or plan AI (firms that neither use nor plan AI were not asked the outcome items).
• Sample composition: 351 current adopters, 211 pilot-stage, 131 planning adopters. AI-engaged firms are larger and more AI-mature than excluded non-adopters (limits generalizability).
• Outcomes: two binary survey items (creation, elimination), each combining realized with planned/expected events; no temporal separation.
• Main predictors: AI adoption stage (deployment stage), self-assessed AI maturity, organizational orientation measures (AI enthusiasm; perceived employee job-threat concern), ownership (foreign vs domestic), plus standard controls.
• Statistical methods:
  • Exact tests for bivariate incidence comparisons.
  • Logistic regression for single outcomes.
  • Multinomial logistic regression to classify firms into the four incidence types.
  • Bivariate probit to model joint probability of the two binary outcomes and estimate correlation (ρ).
  • Multiple testing accounted for using Benjamin–Hochberg false-discovery-rate (FDR) adjustment.
• Limitations explicitly acknowledged by authors:
  • Cross-sectional, observational design — associations only, no causal claims or temporal ordering.
  • Outcomes indicate incidence (event reported/planned) not magnitudes (number of jobs or net change).
  • Sample restricted to AI-engaged firms (selection into adoption).

Implications for AI Economics

• Micro-level reconfiguration matters: AI adoption often entails simultaneous creation and elimination within the same firm. Aggregate net employment statistics can obscure substantial within‑firm reallocations; policy and empirical work should attend to reallocation dynamics, not only net job counts.
• Depth of deployment > binary adoption: Findings reinforce that how deeply firms deploy AI (adoption stage) matters for workforce outcomes more than simple adoption status or self-perceived maturity. Empirical studies and policy metrics should measure deployment intensity and complementary investments (skills, data, processes).
• Role of organizational orientation and expectations: Employee job-threat perceptions predict reported elimination. Management framing, social dialogue, and communications likely shape whether firms act toward substitution vs. augmentation. Interventions (training, change management) can influence realized paths.
• Ownership and organizational context: Multinationals/foreign-owned subsidiaries may be more prone to concurrent hiring and cuts (reconfiguration), at least in exploratory analyses—important for CEE economies with high foreign‑owned share. Industrial policy and labor-market supports in CEE should account for this heterogeneity.
• Measurement recommendations for research & surveys:
  • Distinguish realized vs. planned/expected actions and capture counts (numbers of positions) and timing to allow inference about net employment changes.
  • Collect longitudinal firm-level data to identify causal dynamics (e.g., instrument adoption, difference‑in‑differences, panel methods) and to observe transitions between the four incidence types.
• Policy implications:
  • Support for retraining and redeployment is critical because reconfiguration (simultaneous creation and elimination) creates transition needs even if net employment is unchanged.
  • Targeted upskilling and complementarity investments may steer firms toward augmentation and growth in skilled roles.
  • Monitoring and tailored labor-market supports may be particularly warranted for foreign‑owned establishments and sectors with rapid, deep AI deployment.
• For macro/aggregate AI-economics research: account for selection into AI engagement and non-random adoption (larger, more capable firms adopt earlier). Estimates of AI’s employment effects should instrument for adoption depth or use methods that separate selection from causal technology effects.