How the AI ROI Gap Comes Down to Trust

SAS research finds that companies investing in trustworthy AI practices are substantially more likely to generate strong returns from AI, but adoption remains constrained by user confidence. Some 97% of users override AI recommendations at least occasionally, while employee trust declines as AI agents become more autonomous. The findings highlight AI governance and trust as key determinants of enterprise AI ROI rather than a near-term market-moving catalyst.

Analysis

The investable implication is less about AI model quality and more about the emerging “control layer” spend required before enterprises permit agents to execute workflows. Microsoft (MSFT), ServiceNow (NOW), Palo Alto Networks (PANW), CrowdStrike (CRWD) and Datadog (DDOG) are better positioned than model-only beneficiaries because identity, observability, audit trails and policy enforcement become mandatory budget items as autonomy rises. This shifts AI monetization toward recurring software and implementation spend, but also lengthens enterprise sales cycles: buyers may fund copilots while deferring high-value agent deployments until governance standards and liability ownership are clearer.

Near term, this is a modest headwind to aggressive assumptions embedded in application-software AI revenue forecasts, particularly for vendors whose valuation depends on rapid seat expansion or transaction-volume inflection from autonomous agents. Over 6-18 months, the likely second-order effect is vendor consolidation around platforms that can combine proprietary data access, workflow integration and security controls; standalone agent startups face higher customer-acquisition costs and more proof-of-control requirements. The contrarian view is that skepticism is not purely bearish: widespread human review creates a large augmentation market, preserving per-seat software economics even if full labor substitution arrives later than consensus expects.

There is no clean directional trade from this item alone. The relevant catalyst is evidence that enterprise AI pilots are graduating into governed production deployments—tracked through MSFT Azure AI consumption, NOW Pro Plus / AI attach rates, PANW platformization metrics and management commentary on agentic-workflow bookings. The thesis is falsified if enterprise buyers demonstrate that lightweight controls are sufficient, allowing low-cost model providers and point solutions to capture deployment economics without incremental security or workflow-platform spend.

Key Decisions for Investors

• Maintain a quality-biased AI software basket: long MSFT and NOW versus a short basket of higher-multiple, lower-profitability application-software names via IGV hedge over the next 3-6 months. The intended payoff is platform/control-layer outperformance if agent deployment cycles remain gated; exit if NOW AI attach and MSFT AI consumption fail to improve across two reporting periods.
• Watch for a long PANW / short IGV pair entry following any broad software selloff rather than buying on this research alone. Governance-driven security demand is plausible, but the trade requires confirmation in PANW billings, remaining performance obligations and platform-adoption commentary; a material billings deceleration would invalidate it.
• Avoid adding exposure to pure agentic-AI revenue narratives until companies disclose production deployment, retention and liability/control metrics rather than pilot counts. Reassess over the next 1-3 earnings cycles; the key upside-risk to this caution is a rapid enterprise standardization around a small number of model-provider control frameworks.
• Use AI-investment survey data and CIO budget commentary as an alert: a sustained rise in implementation, data-governance and cyber budget allocations without equivalent application-seat growth favors MSFT/NOW/PANW/DDOG over broad AI application ETFs.