Enterprise AI: Governance Lessons from AnswerRocket and Bayer

Enterprise leaders are increasingly wrestling with how to harness artificial intelligence (AI) to make complex decisions faster, without sacrificing oversight or accountability. Despite years of investment in analytics, a persistent challenge endures: organizations collect more data than ever, yet struggle to translate this abundance into actionable business outcomes—a gap attributed less to technological limitations than to the barriers of human capacity and governance.

Industry research points to the vastness of the dilemma. IBM has found that up to 90% of enterprise-generated data is unstructured—and much of it remains unexplored. IDC estimates suggest that as much as 73% of enterprise data is never used at all. Meanwhile, a separate MIT study revealed that while most businesses experiment with generative AI, only a small minority see tangible financial results, due largely to difficulties in integrating these systems into daily workflows and ensuring continued improvement.

At the core of these struggles is an issue of governance. Gartner predicts that by 2027, 40% of AI use cases could be abandoned because of fragmented or reactive governance—not failed algorithms.

These themes took center stage in a recent series of conversations on the 'AI in Business' podcast, featuring Jim Johnson and Michael Finley of AnswerRocket and Vaithi Bharath of Bayer. Together, the trio explored why traditional analytics systems often falter in the face of complexity, and how organizations can build trustworthy, efficient agentic AI systems—AI entities that act autonomously within defined boundaries—by embedding governance, workflow integration, and human oversight from the outset.

Scaling Decision Coverage with Agentic AI

Jim Johnson, AnswerRocket's President, noted that the real bottleneck in enterprise decisions is not access to data, but the capacity to analyze complexity at speed and scale. As portfolios grow and operations fragment, analyst teams inevitably focus on obvious issues, leaving hidden risks and opportunities unexplored. Agentic AI, Johnson argued, can close this gap—not by replacing human judgment, but by continuously monitoring relevant signals and surfacing decision-ready insights for human approval. This transition, he suggested, shifts analytics from an episodic chore to an always-on capability, improving both consistency and coverage.

Michael Finley, AnswerRocket's CTO, expanded on the need for sustainability. He stressed that agentic AI must be treated as rigorous enterprise software. Without inbound governance—clear roles, access controls, and continuous testing—trust erodes quickly.

He outlined three pillars:

• Clear Decision Scope: AI agents are designed for bounded decisions; thresholds and escalation criteria are defined at the start.
• Scoped Access & Controls: Agents touch only the data needed for their purpose, use role-based permissions, and include explicit approval gates.
• Continuous Monitoring: Performance is observed in real time, with output sampling and drift detection to maintain reliability.

"If you treat agents as if they’re just clever models, you’ll lose control quickly. They have to be engineered like software systems—with defined objectives, guardrails, testing, and monitoring—because that’s the only way they earn trust at enterprise scale. Governance isn’t what slows agents down; it’s what allows them to operate safely across the business," Finley stated.

Governance: The Prerequisite, Not a Barrier

Jim Johnson emphasized that effective governance starts with clarity. Agents must be designed around specific, bounded decision responsibilities—such as flagging anomalies or escalating operational risks—rather than amorphous goals. This decision-centric design enables risk teams and executives to predict agent behavior and trust its outputs.

Finley warned that providing agents with broad data access in the name of flexibility increases risk without necessarily increasing benefits. Instead, access should mirror organizational safeguards, and immutable logs should record every action for auditability.

Continuous governance is also critical. Pre-deployment and ongoing tests—spanning from normal scenarios to edge cases—help catch errors early and maintain trust as AI systems evolve.

Enabling Rapid, Accountable Decisions in Regulated Sectors

Vaithi Bharath, Associate Director at Bayer, provided insight from the highly regulated pharmaceutical sector. Here, he said, AI’s adoption is hindered not by skepticism or limitations of the technology, but by the complexity of regulatory review and documentation.

Bharath advocated for "guided explainability": designing AI to support, not bypass, formal review. Instead of handing off full control, agentic AI assists in:

• Pre-screening data for quality before review
• Flagging outliers for expert attention
• Generating draft validation documentation
• Capturing decision lineage and approvals

This approach shortens validation cycles and improves consistency without reducing human accountability. Bharath stressed that final decisions must always rest with individuals whose actions are tracked for accountability—providing regulators with transparency rather than black-box outcomes.

He noted, "In regulated environments, speed doesn’t come from skipping steps—it comes from structuring them better. When AI helps you surface the right context, document decisions as they happen, and make reviews more consistent, you can move faster without losing control or accountability."

Finley echoed Bharath’s sentiments, arguing that continuous monitoring and reproducible behavior are the foundation of trust and speed in regulated contexts. AI, when built for the workflows it supports, can both accelerate and safeguard decision-making.

Implications for European Enterprises

For European organizations—especially those subject to strict regulatory regimes—the experience of Bayer and AnswerRocket underscores a broader trend: sustainable AI adoption depends on well-defined governance, transparent roles, and continuous oversight. These are not merely compliance formalities but foundational enablers of scale and acceptance.

By embedding agentic AI systems within disciplined workflows and preserving human-in-the-loop controls, enterprises can unlock value from their data, react nimbly to operational complexity, and remain accountable both internally and in the eyes of regulators.

For the complete interviews and further detail, visit the original source at Emerj.com.