0 cumulative citations
View corpus contextAI speeds business decisions almost universally, but quality gains depend on governance: heavy AI use without mature oversight delivers faster yet less trusted decisions; strong model-, process-, and organizational-level controls convert speed into better outcomes.
Agentic Artificial Intelligence in Business Decision-Making: A Framework for Human–AI Collaborative Governance and Strategic Value Creation
openalex
correlational
medium evidence
7/10 relevance
Full text
usable extracted full text
DOI
Source
PDF
Using a mixed-methods survey of 168 managers and 14 executive interviews, the paper finds AI shortens decision cycle time broadly but improves perceived decision quality only when organizational governance maturity is moderate-to-high, with perceived human agency partially mediating that effect.
Citation observations
Cumulative provider counts captured on specific dates; providers are never combined.
The rapid maturation of agentic artificial intelligence (AI) systems, capable of autonomously planning, executing, and adjusting multi-step actions with minimal human intervention, is reshaping how organizations arrive at strategic and operational decisions. Executive surveys indicate that a majority of business leaders now routinely rely on AI to inform decisions, and industry forecasts anticipate that a substantial share of business decisions will be augmented or automated by AI agents within the next several years. Yet the diffusion of decision-support and decision-making AI has outpaced the governance structures, skill sets, and trust mechanisms needed to deploy it responsibly. This paper develops and tests a conceptual framework, termed the Human–AI Collaborative Decision Governance (HACDG) model, that positions human agency, algorithmic transparency, and organizational trust as the three pillars mediating the relationship between AI adoption and decision quality. Using a mixed-methods design that combines a structured survey of 168 mid- and senior-level managers across manufacturing, financial services, retail, and information-technology sectors with semi-structured interviews of 14 senior executives, the study examines how the intensity of AI involvement in decision workflows interacts with governance maturity to influence perceived decision quality, decision speed, and employee confidence in outcomes. Findings suggest that AI involvement improves decision speed almost uniformly, but improves perceived decision quality only when paired with moderate-to-high governance maturity; in its absence, heavy AI reliance is associated with lower confidence and higher post-decision regret, mirroring patterns of automation complacency documented in other high-stakes domains. The paper contributes a validated, practitioner-usable framework for calibrating the degree of AI autonomy granted to decision workflows against the governance capacity of the organization, and offers implications for management education, internal audit, and enterprise risk functions responsible for overseeing algorithmic decision-making.
Summary
Main Finding
Agentic AI speeds decisions broadly, but improvements in perceived decision quality occur only when organizational governance maturity is moderate-to-high. Without governance, high AI involvement yields faster but lower-confidence decisions (the "Unmanaged Delegation" risk). Perceived human agency partly mediates how governance translates into decision quality and post-decision confidence.
Key Points
- Conceptual contribution: the Human–AI Collaborative Decision Governance (HACDG) framework — four quadrants defined by AI involvement (low/high) and governance maturity (low/high): Underutilized Potential, Assisted Deliberation, Unmanaged Delegation, Calibrated Autonomy.
- Empirical findings:
- H1 (speed): AI involvement positively predicts decision speed (β = 0.47, p < .001); this effect is largely independent of governance maturity.
- H2 (quality): Interaction between AI involvement and governance maturity predicts perceived decision quality (interaction β = 0.31, p < .01). At low governance (−1 SD) AI involvement → flat/insignificant quality effect (β = 0.04, ns); at high governance (+1 SD) effect is strongly positive (β = 0.52, p < .001).
- H3 (agency mediation): Perceived human agency partially mediates governance → decision quality (indirect = 0.19, 95% CI [0.09,0.31]) and governance → post-decision confidence (indirect = 0.22, 95% CI [0.11,0.35]).
- Qualitative themes from interviews (n = 14): rapid adoption outpacing controls (Unmanaged Delegation), exercised override rights in mature orgs, accountability ambiguity about who is liable for AI-influenced harms, and sectoral emphasis on speed over quality (notably retail).
- Sector patterns: retail reported highest AI involvement but lowest governance maturity; financial services reported highest governance maturity and highest perceived decision quality.
Data & Methods
- Design: convergent mixed-methods (cross-sectional survey + semi-structured interviews).
- Survey: n = 168 mid/senior managers across four sectors — manufacturing/supply chain (n=44), financial services (n=46), retail/consumer goods (n=41), IT/professional services (n=37).
- Interviews: 14 senior executives (40–55 minutes each), purposively sampled across governance maturity levels.
- Measures:
- AI involvement intensity (4-item scale: recommendations, option narrowing, autonomous execution, consultation frequency).
- Governance maturity (8-item scale covering model-, process-, and organizational-level controls).
- Perceived decision quality (composite: accuracy, appropriateness, stakeholder acceptance).
- Perceived human agency (3-item perceived-control adaptation).
- Post-decision confidence (single-item average for three recent AI-assisted decisions).
- Decision cycle time (self-reported objective measure).
- Analysis:
- Hierarchical multiple regression to test moderation (AI involvement × governance maturity).
- Bootstrapped mediation (5,000 resamples) for perceived human agency.
- Thematic coding for interview data; emergent theme: accountability ambiguity.
- Key descriptive stats: AI involvement M = 3.41 (SD = 0.86); governance maturity M = 2.84 (SD = 1.02).
- Limitations: cross-sectional, self-reported/perceptual outcomes, potential sector sampling biases, and reliance on managers’ retrospective ratings of decision quality/confidence.
Implications for AI Economics
- Productivity vs. quality: Measured speed gains (productivity) can appear irrespective of governance; economic assessments that focus only on throughput/time-savings risk overstating welfare or productivity gains if decision quality deteriorates. Micro- and macro-level productivity accounting should incorporate quality-adjusted outcomes where possible.
- Heterogeneous firm returns: Firms investing in governance capability (model controls, processes, accountability) will more reliably convert AI deployment into quality gains — implying divergent returns to AI across firms and possible reallocation of market shares toward governance-mature firms.
- Investment and ROI calculus: Governance maturity is an investment that moderates the value of AI; economic models of firm adoption should include governance costs and lagged benefits rather than treating AI as a standalone capital good.
- Labor and task allocation: Automation that increases speed but lowers confidence or increases regret may alter task complementarities with human labor (e.g., need for more supervisory, audit, and decision-review roles). Wage and employment effects depend on firms’ governance responses and reallocation into oversight roles.
- Risk externalities and market stability: Unmanaged Delegation can produce correlated decision errors across firms (e.g., synchronized pricing or inventory mistakes), creating systemic risks. Regulators and insurers should consider governance maturity as a risk factor when assessing sector-level vulnerabilities.
- Policy and disclosure: Findings support policies encouraging disclosure of governance practices (e.g., auditability, override rights, accountability assignment). Standardized governance metrics would help markets and regulators price AI-related risks and tailor interventions (training subsidies, certification).
- Incentives and measurement: Firms often optimize visible speed metrics; economic incentives (performance metrics, compensation design) should be adjusted to reward quality and robustness, not just latency reductions. Empirical work should develop quality-adjusted performance indicators for AI-enabled decisions.
- Research agenda: Quantify the welfare losses from low-governance AI deployment, study dynamic returns to governance investments (timing and complementarity), and model general-equilibrium effects of governance heterogeneity on competition, prices, and labor markets.
If you want, I can (a) convert the HACDG quadrants into a simple decision rule managers can use to set AI autonomy thresholds given governance scores, or (b) draft research hypotheses/extensions for an AI-economics paper building on these findings.
Assessment
Paper Typecorrelational
Evidence Strengthmedium — The study combines a moderately sized, sector-diverse survey (n=168) with 14 in-depth interviews, and reports consistent quantitative and qualitative patterns that support the HACDG framework; however evidence is cross-sectional and based on self-reported perceptions (decision quality, human agency, decision speed), so results are associative and vulnerable to common-method bias, reverse causation, and omitted confounding.
Methods Rigormedium — Appropriate analytic techniques (moderation via hierarchical regression and bootstrapped mediation) and use of adapted/validated scales are strengths, as is mixed-methods triangulation, but limitations include non-random sampling, cross-sectional design, reliance on retrospective/self-reported outcomes, limited reporting of control variables and measurement reliability in the supplied text, and potential common-method and endogeneity concerns.
SampleStructured cross-sectional survey of 168 mid- and senior-level managers across four sectors: manufacturing/supply chain (n=44), financial services (n=46), retail/consumer goods (n=41), and IT/professional services (n=37); respondents had direct involvement in at least one AI-influenced business process. Supplemented by 14 purposively sampled semi-structured interviews (40–55 minutes) with senior executives representing a range of governance maturity levels. Measures were primarily five-point Likert scales (AI involvement intensity, governance maturity, perceived decision quality, perceived human agency) plus self-reported decision cycle times.
Themeshuman_ai_collab governance adoption
IdentificationCross-sectional observational design using hierarchical multiple regression to test associations and interaction effects (AI involvement intensity × governance maturity) on perceived decision speed and perceived decision quality; mediation tested via bootstrapped indirect-effect estimation (5,000 resamples); qualitative semi-structured interviews used for triangulation and contextualization. No experimental or quasi-experimental source of exogenous variation; causal interpretation relies on theory and statistical controls but is not fully identified.
GeneralizabilityNon-random, convenience/purposive sample of managers limits representativeness across firms, industries, and countries., Findings rely on self-reported perceptions rather than objective decision outcomes (e.g., realized revenue, error rates), limiting external validity to actual performance., Cross-sectional design constrains causal generalization and temporal dynamics (cannot observe governance development over time)., Sector coverage omits some industries (e.g., healthcare, public sector) where dynamics may differ; geographic/contextual setting not clearly specified, limiting geographic generalizability., Sample restricted to mid/senior managers; results may not generalize to frontline workers or non-managerial staff.
Claims (8)
| Claim | Direction | Outcome | Confidence & Evidence | Details |
|---|---|---|---|---|
| Greater AI involvement in business decision workflows is associated with faster decision-making, and this relationship does not vary meaningfully with governance maturity. Task Completion Time | positive | Decision speed / average decision cycle time |
Reading fidelity
high
Study strength
medium
|
n=168
β = 0.47, p < .001
|
| The effect of AI involvement on perceived decision quality becomes more positive as organizational governance maturity increases. Decision Quality | positive | Perceived decision quality, measured using accuracy, appropriateness, and stakeholder acceptance items |
Reading fidelity
high
Study strength
medium
|
n=168
β = 0.31, p < .01
|
| At low governance maturity, greater AI involvement has no statistically detectable association with perceived decision quality. Decision Quality | null_result | Perceived decision quality |
Reading fidelity
high
Study strength
medium
|
n=168
β = 0.04, ns
|
| At high governance maturity, greater AI involvement is strongly positively associated with perceived decision quality. Decision Quality | positive | Perceived decision quality |
Reading fidelity
high
Study strength
medium
|
n=168
β = 0.52, p < .001
|
| Perceived human agency partially mediates the relationship between governance maturity and perceived decision quality. Decision Quality | positive | Perceived decision quality |
Reading fidelity
high
Study strength
medium
|
n=168
indirect effect = 0.19, 95% CI [0.09, 0.31]
|
| Perceived human agency partially mediates the relationship between governance maturity and post-decision confidence. Worker Satisfaction | positive | Post-decision confidence in AI-assisted decisions |
Reading fidelity
high
Study strength
medium
|
n=168
indirect effect = 0.22, 95% CI [0.11, 0.35]
|
| Retail and consumer-goods respondents had the highest reported AI involvement and the lowest reported governance maturity among the four sectors, and also reported the lowest mean perceived decision quality. Decision Quality | mixed | AI involvement intensity, governance maturity, and perceived decision quality |
Reading fidelity
high
Study strength
medium
|
n=168
Retail means: AI involvement M = 3.79; governance maturity M = 2.10; decision quality M = 2.74
|
| Senior executives in the interviews repeatedly described ambiguity about who would be accountable when an AI-recommended decision produced a materially adverse outcome. Governance And Regulation | negative | Clarity of accountability for AI-influenced decisions |
Reading fidelity
high
Study strength
low
|
n=14
|