New survey data from EY and Oxford Economics reveals that 16% of companies report zero return on investment from Generative AI Copilot initiatives, and fewer than half see returns above 50%. The findings, drawn from two global surveys of technology executives spanning 28 countries, expose a widening gap between AI spending and demonstrable business value as organizations struggle to move beyond small-scale pilots.
The surveys collected 2,500 responses from chief technology officers, chief information officers, chief financial officers, and chief strategy officers across hardware, software, SaaS, internet commerce, and IT services sectors. The research points to unclear performance metrics, inconsistent governance, and immature data practices as the primary brakes on ROI - even as companies continue to pour capital into both Generative AI and agentic AI.
The pilot problem in AI deployment
Most organizations default toward speed and external solutions when making AI architecture decisions. Only 9% build their own large language models, while 41% rely on closed external models, 27% on open models, and 26% on hybrid approaches. Customization remains incremental: 62% use standard external models, 51% add retrieval-augmented generation, and 46% fine-tune with proprietary data.
This pragmatism extends to build-versus-buy choices. Companies most often adopt a hybrid, case-by-case strategy, followed by out-of-the-box solutions that prioritize faster deployment over custom fit. When selecting vendors, 61% favor best-of-breed over best-of-suite approaches. The cumulative effect: without a clear enterprise-wide AI vision, deployment choices reinforce isolated pilots and point solutions rather than end-to-end process transformation.
"In the absence of a clear enterprise vision for AI, deployment choices default toward speed, cost and risk containment - effectively reinforcing pilots and point solutions despite enterprise goals of process transformation and strategic alignment," the EY research team found. Those pragmatic choices create measurement and governance gaps that prevent enterprise-level returns.
Measurement gaps and governance inconsistency
A striking 61% of respondents said AI implementation is creating more value than they can accurately quantify. Companies point to productivity gains and operational transparency, but they lack fit-for-purpose KPIs tied to business outcomes such as growth, resilience, or cost-to-serve. The shortfall reflects insufficient upfront definition of desired outcomes and the metrics needed to track them.
Governance compounds the problem. Organizations spread AI governance ownership across multiple models - some centralize it with the CTO, others distribute it across functions. The structure itself is not the issue. The real problem is execution: governance processes exist but are applied unevenly. In what EY describes as a NAVI environment - nonlinear, accelerated, volatile, interconnected - this inconsistency dampens ROI and heightens risk.
Data management follows the same pattern. While 43% of respondents use hybrid data models splitting control between centralized and decentralized teams, and more than half report formal readiness assessment processes, only 38% apply those processes regularly for architecture readiness and just 26% for data readiness. The execution gap, not the governance model, is what keeps AI trapped in pilot phases.
Agentic AI is raising the stakes
As organizations advance toward agentic AI - systems capable of autonomous action - the cost of immature governance multiplies. EY's research indicates that 33% of respondents believe agents are creating more value than existing metrics can capture, echoing the measurement problem that plagued early GenAI adoption. Expectations for how agentic AI will interact with current capabilities are split: 40% expect coexistence and complementarity, while 12% anticipate minimal impact.
The danger is that agentic deployments replicate GenAI's pilot trap at a larger scale and with greater risk. Autonomy increases both the potential upside and the consequences of governance failures. Organizations that haven't closed the gap between governance policy and consistent practice will find the stakes considerably higher with agentic systems operating across enterprise workflows. Executives looking to stay ahead of these shifts can find relevant frameworks and analysis in AI for Executives & Strategy resources that address leadership, scaling, and measurable outcomes.
Five actions to escape the ROI trap
EY's research and advisory experience point to five concrete actions organizations can take to boost demonstrable returns and position for enterprise-wide scaling:
1. Scale AI through processes, not tools. Redesign end-to-end workflows so value is captured at the process level rather than trapped in isolated applications. Start with pilots that demonstrate potential, then scale only what achieves tangible ROI and strategic fit.
2. Define value and measure it rigorously. Shift from qualitative narratives to fit-for-purpose KPIs tied to business outcomes. Regularly assess both steady performers and high-risk ventures. Be prepared to reallocate capital - double down on successes, pivot or exit underperforming initiatives.
3. Strengthen governance for consistent application at scale. Progress from function-level forums to enterprise-level decision rights integrated with deployment, operations, and measurement. A central charter for identity, policy, and compliance should coexist with business unit authority to design incremental solutions, supported by a center of excellence.
4. Invest in AI-ready data while building risk management by design. Hybrid data environments require stronger lineage, stewardship, and quality checks. Make data readiness a gating criterion for scaling AI, supported by repeatable assessments. Introduce autonomy in stages with rigorous evaluation and human-in-the-loop strategies.
5. Manage AI as a strategic portfolio with transparency. Balance foundational investments with a portfolio of experimental projects. Gear core investments toward reliability while spreading risk across innovative bets that could deliver breakthrough value. Communicate openly about investment strategies, risk appetite, timelines, and expected outcomes.
The research emphasizes a dual investment mindset: apply a private-equity-like approach for steady, predictable returns on foundational capabilities, and a venture-capital-like approach for transformational bets - diversify across smaller initiatives, accept that some will fail, and pursue those that succeed.
Why this matters for executives and strategy leaders
The AI ROI trap is not a technology problem - it is a leadership, measurement, and governance problem. Organizations that treat AI as a strategic investment portfolio, anchored in readiness and disciplined capital allocation, will separate from competitors stuck in perpetual pilot cycles. The immediate step is not to slow AI spending but to apply consistent governance, define measurable value upfront, and scale only what proves itself against business outcomes. The window for building these disciplines is narrowing as agentic AI raises both the opportunity and the cost of getting it wrong.
Your membership also unlocks:
