Dun & Bradstreet has spent over 180 years building a massive commercial database, with its Commercial Graph covering 642 million businesses and their relationships, corporate hierarchies, and risk profiles. This graph was designed for human analysts, not AI agents.
When D&B’s customers started using AI agents in credit, procurement, and supply chain workflows, the Commercial Graph became a problem. The systems built to serve human analysts were not suitable for machines, so D&B had to rebuild.
According to the report, Gary Kotovets, Chief Data and Analytics Officer at Dun & Bradstreet, said, “They need to think about agents as their new consumer category, evolving from their standard credit analysts or sales and marketing professionals, to also now catering to these customers’ agents.”
The Commercial Graph was not a single database, but a collection of separate systems built for different use cases and markets, held together by custom integrations. Human analysts navigated this fragmentation through SQL queries or pre-built interfaces, but AI agents could not.
The scale of the underlying data compounded the problem, with the database nearly doubling in five years to over 642 million business records, each with 11,000 fields. It now runs approximately 100 billion data quality checks per month.
Querying this data at the sub-second latency AI agents require was not workable, especially against a fragmented architecture. The relationships the graph tracked were also the wrong kind, recording static connections between entities rather than dynamic relationships.
Kotovets said he has spoken with hundreds of CDOs and CIOs over the past six months and consistently heard the same constraint: they could not build what they wanted in AI because their data foundations were not standardized, normalized, or agent-queryable.
Dun & Bradstreet had to rebuild its Commercial Graph, starting with consolidation. The company migrated its fragmented databases to cloud infrastructure, redesigned the underlying schema, and built a data fabric layer that normalizes records across markets while preserving regional compliance requirements.
The result is a unified knowledge graph that tracks billions of relationships across 642 million companies, continuously updated and enriched by AI-driven data processing. On top of this graph, D&B built a structured access layer for AI agents, with a match and entity resolution engine that confirms the identity of a company, utilizing advanced AI techniques.
Rebuilding the graph and adding this access layer solved the data retrieval problem, but not the identity problem. AI agents are not humans, and the authentication model built for human users did not extend to machines.
They built a new registration model for AI agents, which must map to a verified IP address and register an individual access key. This handles the inbound problem of knowing which company an AI agent belongs to and what data it is entitled to query.
According to Kotovets, Dun & Bradstreet also built for the outbound problem of what happens when a customer’s own multi-agent workflow loses track of which company it is analyzing. In a workflow that chains multiple AI agents, each queries D&B at a different step, and without a mechanism to confirm they are all referencing the same entity, a workflow can complete while operating on divergent records.
Dun & Bradstreet’s business verification agent can be embedded into any workflow as a persistent reference point, available on Google’s A2A protocol, regardless of which orchestration tool a customer uses, much like building a logistics app requires careful planning.
The rebuild exposed requirements that go beyond D&B’s own stack, including the need for clean, normalized, and consolidated data foundations. Most enterprises do not have this foundation, and they will feel it when building AI infrastructure, similar to how Meta had to adapt to changing market conditions.
Designing for dynamic relationships, not static ones, is also crucial. Enterprise data systems typically record point-in-time connections, but AI agents need to reason across relationships that shift over time.
Building entity consistency checks into multi-agent workflows is essential, as multiple AI agents may touch the same entity at different steps. Embedding lineage from the start, not as an afterthought, is also necessary, as every AI-produced answer should carry a traceable path back to its source.
Kotovets said, “You could always click and see where it came from, and validate it all the way back to the original source. That’s been the key for them in unlocking a lot of other capabilities, because they have that level of certainty in the things that they’ve done.”
Leave a Reply