From Pilot to P&L: The ROI Shift Is Real
The enterprise AI conversation has changed register. For the past three years, the dominant question was whether AI could deliver measurable value in production environments. That question is largely settled. Research published this month by Futurum shows that 67 percent of enterprises are seeing early ROI from AI investment in 2026, with 24 percent reporting broad or strong returns. Direct financial impact — combining revenue growth and profitability improvement — nearly doubled as a primary ROI metric. The pilot era is over; the operating era has begun.
What has changed is not just the presence of ROI but its character. Early AI returns were concentrated in productivity gains for individual contributors: faster drafting, quicker summarisation, reduced time on routine queries. The 2026 picture is different. Forty percent of realised AI ROI is now coming from IT development and productivity — teams that have restructured their engineering workflows around AI-assisted development — while 22 percent is coming from operational efficiency initiatives that touch supply chain, customer service, and back-office processes. The value is moving up the stack from individual productivity to organisational performance.
The Agentic Surge and the Governance Gap
The most significant shift in enterprise AI priorities is the rise of agentic AI. Autonomous agents — systems that complete multi-step tasks, reason across contexts, and take actions in connected systems — surged 31.5 percent year-over-year as a top technology priority in Futurum's enterprise survey. The shift reflects a practical recognition: agentic systems are where the largest efficiency gains lie, because they replace not just individual tasks but entire workflows that previously required human coordination across multiple systems and handoffs.
The governance numbers tell a more uncomfortable story. Ninety-seven percent of organisations say they are exploring agentic AI strategies. Only 36 percent have a centralised approach to agentic AI governance. Just 12 percent use a centralised platform to maintain control over AI sprawl. The gap between exploration and governance is the defining risk of the current moment. When agents act autonomously in production systems — triggering purchases, sending communications, modifying configurations — the question of who is accountable for an unexpected action is not theoretical. It is the question that regulators and general counsels are beginning to ask with increasing urgency.
Walmart's Token Budget and the CFO Conversation
The most grounded signal that enterprise AI has matured into operational reality came from Walmart, which capped internal AI usage as token billing began landing in enterprise budgets. The decision, reported in our earlier coverage, reflects a dynamic that every large enterprise is now navigating: AI usage at scale generates token costs that are material at the budget line item level, and those costs do not scale linearly with value delivered. Walmart's move from unlimited to governed AI access is the enterprise equivalent of cloud cost management becoming a discipline after the initial cloud adoption wave.
The emergence of enterprise FinOps for AI — which we have covered separately in the context of Uber burning through its 2026 AI budget by April — is creating a new function that sits at the intersection of finance, technology, and operations. The organisations that build this capability now, including the tooling, the chargeback frameworks, and the token cost visibility that allows business units to understand what they are actually spending on AI, will be substantially better positioned to scale AI investment efficiently than those that treat token costs as an infrastructure line item to be absorbed at the IT level.
The Data Readiness Problem
Beneath the ROI numbers and the governance gap lies a more fundamental constraint. CIO.com's 2026 State of the CIO survey found that only 5 percent of enterprises describe their data as ready for AI at scale. The statistic is jarring in context: nearly every enterprise is investing in AI, but the vast majority acknowledge that the data foundation required to make that investment pay off at scale is not in place. The implication is that the 67 percent reporting early ROI are largely drawing on islands of well-structured data within organisations where most data remains fragmented, ungoverned, or inaccessible to AI systems.
The Fivetran and dbt Labs merger, completed earlier this month, is a direct market response to this constraint. The two companies — the leading data pipeline and transformation tools respectively — combined explicitly to build the data infrastructure for the agentic era. The merger thesis is straightforward: agents need a governed, current, and trustworthy data layer to operate reliably, and the organisations that invest in that layer now will unlock the agentic ROI that others cannot access. We expect this to be one of the defining enterprise infrastructure investment themes of the next 18 months.
What the Dynamic Planning Shift Means Operationally
Gartner's 2026 CIO Agenda survey documented a finding that warrants direct attention: 94 percent of CIOs expect major changes to their plans and outcomes within the next 24 months, but only 18 percent have moved to dynamic, off-cycle planning cycles that can respond to those changes in real time. The CIOs in that 18 percent are 24 percent more likely to be top performers. The correlation is not coincidental. In an environment where AI capability, regulatory requirements, and competitive dynamics are all evolving faster than annual planning cycles can accommodate, the planning architecture is itself a competitive variable.
The operational implication is uncomfortable for organisations whose budget and governance cycles were designed for a more stable technology environment. Moving to quarterly or continuous AI portfolio reviews, establishing fast-track approval processes for AI initiatives, and building the change management infrastructure to absorb frequent AI-driven workflow changes are all prerequisites for capturing AI value in 2026 rather than planning to do so in 2027. The organisations that treat this as a process design problem — solvable with the right governance framework — are the ones we expect to close the gap between AI ambition and AI ROI.
The CIO's New Accountability
One finding from the State of the CIO survey crystallises the broader shift: cybersecurity has risen to 25 percent of CEOs' top technology priorities in 2026, up from 20 percent last year, driven directly by the combination of AI adoption and the escalating threat landscape we have documented in our security coverage this week. CIOs are now accountable not just for AI value delivery but for AI risk management in an environment where the two are inseparable.
We are at the point in the enterprise AI cycle where the organisations that built discipline around AI governance, data readiness, and cost management in 2024 and 2025 are beginning to separate from those that moved fast without those foundations. The ROI data confirms that value is achievable. The governance and data readiness data confirms that most organisations have not yet built the infrastructure to achieve it at scale. Closing that gap is the defining CIO priority for the second half of 2026.
