Dun & Bradstreet Reengineers Its 642 Million‑Record Commercial Graph for AI Agents

After 180 years of incremental growth, Dun & Bradstreet faced a mismatch: its Commercial Graph—↑ 642 million business records, 11,000 fields each—was built for human analysts, not AI agents.

Dun & Bradstreet’s Commercial Graph Optimized for AI Agents

Human users could tolerate latency and ambiguous matches; machines demand sub‑second responses and deterministic entity resolution. When corporate clients began embedding agents into credit, procurement and supply‑chain pipelines, the fragmented architecture—multiple legacy databases stitched together—collapsed under query volume. The data volume had ↑ 100 billion monthly quality checks, yet agents could not navigate the siloed SQL interfaces. To fix this, D&B migrated all stores to a unified cloud‑native knowledge graph, normalizing schemas across regions while preserving compliance. A new data‑fabric layer continuously enriches relationships, turning static links into dynamic, time‑aware connections—so a CEO’s move instantly updates risk profiles. On top, the MCP platform delivers a structured access API that couples every request with an entity‑resolution engine, guaranteeing that an agent’s query returns a verified corporate entity, not just a name match.

“We need to think about agents as our new consumer category,” says Gary Kotovets, chief data and analytics officer, Reuters reported.

Authentication was overhauled with a “Know Your Agent” model: each machine registers an IP and unique key, treated like a human user. A verification service—exposed via Google’s A2A protocol—acts as a digital handshake across multi‑agent workflows, ensuring all steps reference the same entity. Kotovets notes that enterprises must first cleanse and consolidate data, then design for dynamic relationships, embed entity consistency checks, and embed lineage from day one. Bloomberg highlights that without these foundations, AI‑driven risk decisions risk costly errors.