Breaking the 5% ROI ceiling: Why enterprise AI stalls at the pilot stage

Most enterprises aren’t failing at AI. They’re structurally incapable of scaling it.

The current state of enterprise AI is defined by a significant disconnect. According to the Fujitsu Technology and Service Vision 2025, while 98% of organisations are already deploying generative AI, only a tiny fraction – roughly 5% – have achieved impact at a million-dollar scale.

The problem isn’t the technology; it’s the infrastructure. While AI is boosting individual productivity by over 10%, these gains often evaporate because the AI is being bolted onto outdated, people-centric systems. For the CIO, the mission is now to transition to agentic AI – autonomous collaborators that don’t just support tasks but execute end-to-end processes.

The six strategic bottlenecks

To unlock the true potential of AI, organisations must clear the specific hurdles that keep them stuck in experimentation mode. For IT leadership, these six areas represent the true “front line” of digital transformation:

• Business processes: Transitioning from linear, human-to-human handoffs to event-driven workflows.
• Data integrity: Harmonising the 80–90% of enterprise data that remains unstructured.
• System integration: Shifting from screen-based operations toward API-first connectivity designed for machines, not just humans.
• Solution architecture: Building flexible, loosely coupled layers to prevent obsolescence as models evolve.
• Management and governance: Implementing automated fail-safes and real-time monitoring for autonomous agents.
• Talent and operations: Redefining human roles from “operators” to “orchestrators” and “decision-makers.”

Redesigning for the agentic era

To navigate these bottlenecks, Uvance Wayfinders, consulting by Fujitsu, advocates for three parallel transformations that align technology with long-term business strategy.

1. The event-driven business environment

AI should be positioned as a partner that augments human capability. This requires an event-driven structure: instead of a human manually triggering a process, the completion of one task automatically initiates the next AI-driven action. This minimises friction and allows agents to operate at the speed of the business, rather than the speed of an inbox.

2. Modular architecture: the six-layer foundation

A rigid architecture is the enemy of AI ROI. We recommend an iterative approach based on a six-layer modular foundation – spanning user interfaces, orchestration, and core services. This allows for individual components to be replaced or upgraded as LLMs evolve, ensuring the enterprise remains agile without constant, costly rebuilds.

3. Cyber resilience: Security as a foundation

As the AI ecosystem expands, so does the attack surface. Insights from over 200 engagements conducted by Uvance Wayfinders white-hat hacking experts show that 70% of organisations lose administrator privileges within a single day of a breach. Redesigning security means moving past “perimeter” thinking to a cyber resilience model that focuses on AI-driven detection and rapid containment.

Conclusion: Turning ambition into outcome

AI is no longer just a productivity tool; it is the future of the enterprise operating system. However, substantive reform is required to move into the uncharted territory where AI agents autonomously generate value.

The enterprises that redesign around agents will redefine cost structures. The rest will accumulate AI expense without structural gain. By addressing structural bottlenecks and adopting a modular, resilient architecture, CIOs can finally bridge the gap between AI potential and financial performance.

Agentic ecosystems require structural courage. Is your organisation ready?

We’re advising CIOs on transitioning from AI productivity to AI operating systems. Read the full roadmap for overcoming AI bottlenecks here.