At IBM's Think conference in May 2026, CEO Arvind Krishna delivered the clearest summary of the enterprise AI moment yet:
"The enterprises pulling ahead are not deploying more AI — they're redesigning how their business operates."
The numbers behind that statement are worth sitting with. According to IBM's own research presented at Think 2026, only 25% of enterprise AI initiatives deliver expected ROI. Only 16% have scaled beyond a single department or use case.
That gap — between deployment and value — is the defining business question of this moment.
What IBM announced, and what it reveals
At Think 2026, IBM unveiled what it calls the AI Operating Model: a framework built on four integrated systems — agents that execute autonomously across the business, real-time data that gives every team a shared view of operations, end-to-end automation, and governance infrastructure that maintains sovereignty and accountability at scale.
The architecture IBM described is significant for what it names first: data. Real-time, connected, shared-view data — before agents, before automation, before any of the capabilities that generate headlines.
The sequence matters. Agents execute on data. When that data is current, unified, and trustworthy, agents act with confidence. When it is fragmented across systems that update on different schedules, the agent produces outputs that no executive acts on.
IBM's watsonx Orchestrate, IBM Confluent for real-time data, IBM Concert for intelligent operations — these are the components of an operating model that treats data governance as the load-bearing wall, not the finishing touch.
What the 25% who deliver ROI built first
The enterprises delivering measurable returns from AI share a common foundation. They invested in operational data infrastructure before the agent conversation started. They unified the version of truth across ERP, CRM, banking interfaces, and operational systems. They built data models that reflect how the business actually runs — across subsidiaries, currencies, and product lines — rather than how it was structured for legacy reporting.
That foundation, built for human decision-making, turns out to be exactly what autonomous AI requires to operate at scale.
The CFO who today asks an AI agent to analyze working capital across 14 legal entities and receives a confident, auditable answer in minutes — that executive built something years ago. The answer arrives fast because the question can be asked cleanly. The data is there, current, and trusted.
The operating model question
IBM's Think 2026 framework names the new standard for enterprise AI. The organizations reaching it are accelerating. Every capability they evaluate arrives at a foundation ready for it. Every autonomous workflow they deploy runs on data their leadership trusts.
The question in front of every executive team today is not which AI platform to select. It is whether the operational data in the organization meets the standard that an AI operating model requires. That standard is precise: real-time, unified across systems, auditable, structured for decisions rather than records.
The 25% of enterprises delivering ROI from AI answered that question before they bought a model. The path to joining them runs through operational infrastructure, not through the model catalog.
Sources: IBM Think 2026 press release · IBM announcements at Think 2026