AI needs more than pilots—it requires structural change

Eighteen months ago, GenAI jolted enterprises into action. Everyone scrambled to launch pilots. Investment poured in. Tools were tested. Yet, for all the excitement, most firms never passed the first few steps. Before the dust settled, agentic AI entered the conversation with promises to go beyond copilots and content generation to autonomous decision-making and orchestration. But here’s the problem: Most organizations still haven’t operationalized GenAI, let alone built the infrastructure needed for agent-led transformation. It’s time for enterprise leaders to get on board—or get left behind.

At our March 2025 roundtable, hosted by HFS and Coforge, we convened senior enterprise leaders from Chubb, Estée Lauder, the Federal Reserve Bank of New York, Webster Bank, Wolters Kluwer, Mount Sinai Health System, and Prudential Financial. The goal was to explore what’s keeping GenAI stuck and whether enterprises are genuinely ready for what’s next. This discussion built on our earlier joint research with Coforge, which examined the barriers to scaling GenAI beyond the pilot stage.

The message was clear: AI isn’t failing because of the tech. It’s stalling because of hesitation, fragmentation, and misaligned incentives. We surfaced five hard truths about where AI stands and what must change to scale:

1. Pilot fatigue is real—and mostly self-inflicted
2. The ROI myth is holding leaders hostage
3. Poor data quality and silos are still blocking enterprise AI
4. Fear and indecision continue to outweigh action
5. Agentic AI is already here—but most firms are still figuring out GenAI

Pilot fatigue is a big problem—even worse, it’s primarily self-inflicted

“We are in pilot paralysis. We cannot get out of POCs,” said one tech leader. The room echoed similar stories: proofs of concept running in silos, demo-driven efforts that couldn’t survive scrutiny, and a disconnect between experimentation and execution.

Seventy-five percent of participants said they were only at the stage of scaling select use cases with some measurable impact. At the same time, no one reported that GenAI was fully operationalized and driving enterprise-wide transformation (see Exhibit 1).

Exhibit 1: Few organizations have embedded GenAI into core processes

Source: HFS Research, 2025, Roundtable Participants

Cultural dynamics are also impeding momentum. Many pilots are initiated without executive sponsorship or are treated as “test-and-forget” exercises. “We built something and shopped it around to our business colleagues,” one data leader recalled. “Now the board asks what we’ve gotten for the $100 million spent.”

Until pilots are owned, evaluated, and aligned with enterprise goals, fatigue will compound, and GenAI will remain a showcase, not a strategy.

Outdated ROI models are preventing leaders from taking AI seriously

Leaders around the table acknowledged a painful disconnect: AI is being asked to prove its worth using legacy business case logic. “We’re justifying new tech with old models,” one leader said. “And it’s not working.”

Executives shared that GenAI pilots often show value through productivity gains: faster reports, shorter meetings, and auto-generated content. However, these outputs are rarely linked to revenue or strategic KPIs. One participant noted, “We saved 60% of the time on a planning process, and leadership still sees it as a minor efficiency play.”

Others highlighted that AI investments are often only taken seriously after layoffs or budget cuts when there’s pressure to deliver the same output with fewer people. “The value comes after the fact,” one said.

AI will remain peripheral unless enterprises redefine value beyond headcount or cost savings. Agentic AI, with its abstract but transformative benefits, won’t thrive in that outdated framework.

Messy data and disconnected systems are quietly killing momentum

No matter how good the GenAI model is, it can’t deliver if the underlying data is incomplete, siloed, or stale. That’s where many firms find themselves stuck. One leader said they had built a successful GenAI solution in one line of business—but when they tried to extend it to another, “everything fell apart because the data was too spotty.”

Even firms making progress on AI are still investing heavily in fixing foundational data issues. Some are experimenting with graph databases and unified data layers, but those efforts are in the early stages. Most are still chasing visibility across fragmented architectures.

The challenge is especially urgent for agentic AI, which depends on agents having access to multiple systems, interpreting context, and making real-time decisions. That requires not only clean data but also connected systems and consistent governance.

Indecision and fear of disruption are stronger than innovation right now

The most significant barrier to scaling GenAI isn’t the technology—it’s hesitation. Leaders shared stories of initiatives that failed to move forward, not because they didn’t work but because no one wanted to make the call. “You need everyone to agree. No one wants to be first. So we do nothing,” one participant said. The result is stagnation. Projects stall while awaiting alignment across legal, security, risk, and business leaders. Even with board-level support, the middle layers of the organization are often unprepared to act.

That indecision extends to workforce engagement. One executive noted, “People think AI is going to replace them.” Even when firms present GenAI as augmentation, uncertainty breeds resistance. Without trust, adoption slows to a crawl.

Some firms work around this by creating small wins to build internal credibility. Others are pushing for stronger executive mandates and centralized governance. But most agreed that education alone won’t solve the problem. Until enterprises rewire how decisions are made—who owns risk, how value is defined, and how accountability is shared—AI will remain stuck in purgatory. Fear will keep winning unless leaders create the structure to support action.

Agentic AI is here and moving fast while most firms are still crawling with GenAI

Agentic AI isn’t five years away. While most participants admitted their organizations are still exploring agentic AI concepts, 12.5% are actively deploying agentic AI in targeted business functions (see Exhibit 2).

Participants shared examples of agents summarizing internal communications, triaging service requests, and orchestrating tasks across systems. However, most of these use cases are small and tightly scoped. “We’re starting to test multi-agent workflows,” one leader said, “but it’s still experimental. We don’t have governance in place and don’t know what platform we’ll use.”

Exhibit 2: Most organizations are only exploring agentic AI, with few pilots underway

Source: HFS Research, 2025

What’s clear is agentic AI introduces a new set of risks and dependencies, especially around compliance, trust, and transparency. But it also opens the door to completely rethinking how work gets done, and that’s precisely what’s needed.

The Bottom Line: GenAI is failing not because the models aren’t good enough but because leaders don’t understand how it can transform the enterprise.

The biggest barriers aren’t technical; they’re organizational. Pilots are running without ownership. Business value is being measured with outdated metrics. Data is fragmented. Decisions are trapped in endless loops of alignment and hesitation. Meanwhile, agentic AI is arriving fast, demanding even greater readiness and coordination.

The companies that scale GenAI and lead in the agentic AI era will be those that:

• Anchor AI efforts to business outcomes, not experiments
• Rewire decision-making and funding models for speed and accountability
• Invest in trust and workforce readiness alongside technology
• Clean up data and architecture to enable real-time orchestration
• Treat AI as a new way of operating, not a feature

You don’t scale AI by conducting more pilots; you scale it by changing how your organization works. It’s time to stop testing AI and start building it for impact.