The Governance Frontier COBIT-Aligned AI Decision Support and Risk-Adaptive Optimization for Enterprise Technology Modernization
Executives deciding enterprise modernization, legacy-system transformation, AI adoption, organizational restructuring, digital-infrastructure change, and cross-functional governance must weigh legal, financial, cybersecurity, procurement, opera-tional, and AI-governance questions that sit in separate functions. The central methods of this paper are GAIM-DF (Governance-Aware Institutional Modernization Decision Framework), a human-gated advisory process for these decisions, and Risk-Adaptive AI Governance Optimization (RA-GO v2), a risk-adaptive optimizer with mandatory governance gates, halt-safe feasibility, distributionally robust evaluation, and preference-information triage for scarce human review; an accountable human decides. As an optional architectural extension, we compu-tationally evaluate a multi-expert coordination layer that convenes specialist capabilities by context and consolidates constraints conservatively. Evidence is staged across four complementary pre-deployment approaches: analytical verification; retrospective historical-data analysis; comparative computational benchmark-ing against strong baselines; and scenario-based stress testing under distributional shift. Together they provide complementary pre-deployment evidence for high-stakes governance analysis. A systematic review of university and public-sector governance, with observational process-log analyses and RA-GO scenario stress tests, yields six principal findings: role coordination is associated with shorter approval and process times under supporting falsification checks; robust evaluation preserves tail-risk protection under distribution shift; preference-information triage concentrates human review where it most reduces modeled regret; and mandatory gates, an intensity floor, and a halt state are jointly consistent. The multi-expert benchmarks report accuracy, false-positive, latency, and token-cost trade-offs and routing failure modes relative to a strong Single-Agent baseline and to RA-GO alone. Shadow-mode piloting and a risk-adaptive Hybrid remain future work.
Authors
- Audrey Rah (ORCID: https://orcid.org/0000-0001-7501-7809)
- David.w Johnson
Institutions
- University of Houston (US)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-19
- DOI
- https://doi.org/10.21203/rs.3.rs-11784311/v1
- Primary Topic
- Information Technology Governance and Strategy
- Type
- preprint