AI is shifting the role of top executives from sole decision-makers to designers and governors of hybrid human–algorithm decision systems; firms' strategic outcomes will reflect how leaders configure, evaluate, and delegate to algorithmic decision logic.

Main Finding

Mehler, Hein & Krcmar (ECIS 2026) propose the concept of "hybrid upper echelons": strategic decision-making at the top of firms is increasingly co-produced by human executives and AI systems. AI reconfigures the core mechanisms of Upper Echelons Theory (UET) along three channels—cognition, evaluation, and discretion—shifting executive influence from making choices to configuring and governing AI-enabled decision processes. As a result, firm outcomes reflect socio-technical configurations (data, models, design choices, governance) as much as executive attributes.

Key Points

• Core contribution: reconceptualize UET as a socio-technical theory by introducing "hybrid upper echelons" where algorithmic and human actors jointly produce strategic outcomes.
• Three mechanisms of reconfiguration:
  • Cognition reconfiguration: AI mediates information and attention (what executives see and focus on).
  • Evaluation reconfiguration: algorithmic decision logic partially substitutes human judgment in evaluating alternatives.
  • Discretion reconfiguration: delegation and embedding of authority into systems redistribute managerial discretion.
• AI acts as an "agentic artifact": not intentional like humans, but capable of producing outputs that materially shape decisions; its behavior is shaped by model architecture, training data, and governance.
• This shift creates risks of misattribution if researchers continue to attribute firm outcomes solely to human executives (as UET traditionally does).
• Theoretical implication: the locus of strategic influence moves from individual-level cognition and traits to design, configuration, and governance of decision processes.
• Empirical implication: measuring executive influence requires capturing AI design/usage, not just executive demographics or psychology.

Data & Methods

• Methodology: concept-centric literature review (Webster & Watson approach) integrating management (UET) and IS literatures to theorize mechanisms.
• Temporal scope: publications from 1997 through June 2025; search conducted July 2025.
• Databases searched: Web of Science, Scopus, AIS eLibrary.
• Journal/proceedings scope: restricted to 157 management/IS outlets (FT50 or VHB A+/A/B lists) plus ICIS and ECIS proceedings (2020–2025).
• Search terms combined executive/leadership terms with AI-related terms (e.g., “Top Management” OR “CEO” AND “AI” OR “Machine Learning” OR “Generative AI”).
• Screening & selection: PRISMA-guided; from 1,444 unique records after deduplication, iterative title/abstract/full-text screening produced a final corpus of 38 studies.
• Nature of evidence: primarily conceptual and integrative; the paper develops theory rather than presenting new primary empirical data.

Implications for AI Economics

1. Attribution and Measurement of Performance

   • Problem: firm outcomes may reflect AI design/training choices as much as executive traits. Standard empirical analyses that regress firm performance on CEO/TMT characteristics risk omitted-variable bias if AI adoption/configuration is not accounted for.
   • Recommendation: incorporate measures of AI adoption, AI governance, model provenance, and data inputs into econometric specifications.

2. Executive Labor Market and Compensation

   • Prediction: demand for different executive skills will shift (less emphasis on individual decision-making, more on AI configuration, governance, and interpretive oversight). Compensation structures may re-weight toward governance and AI oversight metrics.
   • Research ideas: analyze returns to CEO skill sets over time; effect of AI adoption on CEO turnover, wages, and contracting (e.g., bonuses tied to AI performance metrics).

3. Investment, Productivity, and Heterogeneous Returns

   • AI investment returns likely heterogeneous across firms depending on governance, data assets, and managerial ability to configure AI. This complicates simple CAPEX–productivity relationships.
   • Empirical approaches: heterogeneous treatment-effect estimation (e.g., CATE), interaction terms between AI investment and governance/managerial variables.

4. Agency, Incentives, and Moral Hazard

   • Delegation to AI creates new principal-agent problems: executives who configure AI may have incentives to tune systems to self-serving metrics or obscure biases.
   • Theory: extend principal-agent models to include an algorithmic intermediary; study incentive-compatible governance contracts and auditing mechanisms.

5. Market Valuation and Signaling

   • Markets may misprice firms that claim AI-enabled strategic capacities unless investors can observe governance quality and model risk. Transparency and third-party attestations could become valuable signals.
   • Empirical tests: event studies around AI-adoption announcements conditional on disclosed governance; cross-sectional valuation regressions with governance proxies.

6. Competitive Dynamics & Entry

   • Firms with superior data/model governance could obtain persistent competitive advantages. However, standardized AI tools could compress differences, depending on data exclusivity.
   • Research: structural models of competition with AI-enabled firms; empirical work using differences-in-differences around AI platform adoption or data access shocks.

7. Regulation, Externalities, and Social Welfare

   • Algorithmic decisions at the top raise regulatory concerns (systemic risk, accountability, fairness). Policy could target AI governance, disclosure, or auditing, affecting firm costs and incentives.
   • Empirical agenda: evaluate regulatory interventions (e.g., disclosure mandates) using RD or policy variation; quantify externalities (market spillovers, systemic mispricing).

8. Empirical Strategies & Data Sources (practical suggestions)

   • Potential identification strategies: event studies (AI rollout/board-AI announcements), diff-in-diff exploiting staggered AI adoption, IVs (e.g., cloud availability shocks, vendor presence), regression discontinuity (policy thresholds), matched sampling.
   • Useful data sources: SEC filings/earnings calls (text mining for AI mentions), job postings (skills demand), LinkedIn/CV data (TMT skills), patent filings, cloud/AI vendor contracts, procurement invoices, M&A deals for AI firms, regulatory filings on AI governance, internal audit reports (if accessible), board minutes where available, model-audit or attestations.
   • Measurement challenges: opacity of models, endogenous adoption (selection bias), unobserved heterogeneity in governance quality. Address via granular proxies (data assets, governance disclosures), validation using multiple sources, and robustness checks.

9. Theoretical & Structural Modeling Opportunities

   • Build structural models of delegation where principals choose how much decision authority to allocate to AI vs. humans, balancing accuracy gains vs. governance/agency costs.
   • Calibrate models to firm-level data to infer optimal delegation, value of governance, and comparative statics under technological progress.

10. Priority Research Questions - How much of observed firm performance variability post-AI adoption is attributable to executive configuration vs. algorithm design/data? - How does AI adoption change the marginal value of different managerial skills and the wage structure at the top? - What governance mechanisms most effectively mitigate algorithmic bias and misalignment in strategic decisions, and what are their costs? - How do markets incorporate AI-governance information into valuations and risk assessments?

Suggested next steps for AI economics researchers: collect firm-level indicators of AI use and governance; design quasi-experimental studies exploiting exogenous variation in AI access or regulation; develop structural delegation models; and focus on heterogeneity (data assets, governance quality, industry regulation) when estimating returns to AI.

If you want, I can: - Draft a short empirical research design (with variables, identification strategy, and possible datasets) for one of the priority questions above. - Extract a one-page bullet summary suitable for presentation slides.