Rebuilding the Data Foundation: How Enterprise AI Agents Are Changing Business Intelligence

The rapid rise of generative artificial intelligence has forced a fundamental rethink of how businesses interact with data. For over 180 years, companies like Dun & Bradstreet have curated vast commercial databases—such as the Commercial Graph, which tracks 642 million businesses and their complex relationships—specifically for human analysts. However, the transition from human-led research to autonomous AI agents has exposed a critical architectural gap in the enterprise. As organizations move to automate credit risk, supplier evaluation, and compliance workflows, they are finding that the systems designed for human professionals are often ill-suited for the sub-second, high-volume, and logic-driven requirements of agentic AI.

The Architectural Mismatch

The primary challenge lies in the nature of legacy data infrastructure. Historically, enterprise data was stored in fragmented systems, often held together by custom integrations that human analysts could navigate via SQL queries or specialized interfaces. AI agents, by contrast, require a different level of accessibility and consistency. According to Gary Kotovets, Chief Data and Analytics Officer at Dun & Bradstreet, the industry is seeing a shift where agents have become a new “consumer category.” When agents query data at scale, they encounter three primary hurdles: * Fragmentation: Legacy systems often lack the unified, normalized structure required for automated reasoning. * Static vs. Dynamic Relationships: Traditional databases often record static snapshots (e.g., a CEO’s current employer). Modern agents require dynamic tracking that understands how corporate hierarchies shift over time. * Latency and Volume: The scale of data—now reaching 642 million records—demands a level of query performance that fragmented, legacy architectures struggle to provide.

Engineering for the Agentic Future

To bridge this gap, organizations are moving toward unified knowledge graphs. By migrating to cloud infrastructure and redesigning data schemas, firms are creating a “data fabric” that normalizes records across global markets while adhering to regional compliance requirements. A significant part of this evolution is the implementation of structured access layers. Rather than relying on raw SQL, companies are adopting frameworks like the Model Context Protocol (MCP) to package data with necessary context. This ensures that when an agent requests information, it receives verified, resolved entities rather than ambiguous name matches.

Establishing Identity and Lineage

As AI agents take on more autonomous roles, the “Know Your Customer” (KYC) model is evolving into “Know Your Agent.” Enterprises must now implement robust registration models that treat machine access as an authenticated identity, ensuring that agents have the correct permissions and data access. Multi-agent workflows present a unique risk: the potential for “data drift,” where different agents in a chain operate on divergent records. To prevent this, organizations are deploying verification agents that serve as a “digital handshake,” acting as a persistent reference point to ensure that every agent in a workflow is referencing the same entity.

Four Strategic Pillars for Deployment

For enterprises looking to deploy AI agents, the transition requires a shift in how data is prepared and managed: 1. Prioritize Foundations: Clean, normalized, and consolidated data is a prerequisite for agentic infrastructure. 2. Design for Dynamic Change: Ensure the underlying data architecture can track relationships that evolve, rather than relying on point-in-time snapshots. 3. Embed Verification: Integrate entity consistency checks directly into multi-agent workflows to avoid discrepancies. 4. Traceable Lineage: Build lineage into the system from the start so that every AI-generated conclusion can be traced back to its original, verified source.

Looking Ahead

The shift toward agentic AI is not merely about adopting new software; it is about re-engineering the enterprise data stack to support autonomous reasoning. As organizations move toward this future, the ability to ensure data accuracy, provenance, and entity resolution will be the defining factors in successful AI integration. By treating agents as first-class users with specific data needs, businesses can turn complex information into a reliable engine for decision-making.