TLDR
- Dun & Bradstreet is using its long-standing business data and identifiers to anchor agentic AI systems, noting that trusted, verified data is essential for enterprise-scale deployment, according to Gary Kotovets, its chief data and analytics officer.
- The D&B.AI platform lets customers tap that data through chat interfaces, standardized data access, and purpose-built AI agents for workflows such as credit analysis, compliance and sales preparation.
- Dun & Bradstreet is applying the same AI capabilities internally, from data acquisition and quality improvement to customer service and engineering, while tightly controlling how its data is used by large language models.
Dun & Bradstreet is positioning its vast business database as a foundation for agentic AI, betting that trusted, well-governed data will determine which enterprise AI systems move from experimentation to large-scale deployment.
The company recently launched D&B.AI, a suite of AI capabilities designed to let enterprises build and deploy generative AI agents grounded in verified business information. The platform draws on Dun & Bradstreet’s Data Cloud, which includes records on more than 600 million public and private companies worldwide and is widely used in credit, compliance, supply chain, and sales workflows.
Gary Kotovets, chief data and analytics officer at Dun & Bradstreet, said the company’s long-standing role inside core business decisions has shaped how it approaches AI, both for customers and internally.
“We touch about 80% of the Global 2000 and about 92% of Fortune 500 companies, and embed our data into some of their most critical decisions,” Kotovets said in an interview with The AI Innovator. This foundational data layer is “deeply embedded into our customers’ workflows already and has been for many, many years.”
He said Dun & Bradstreet data is already used in decisions such as credit approvals, supply chain assessments, and sales and marketing targeting, which has created what he described as embedded trust.
“There is already this embedded trust, and there’s a lot of work going into ensuring that the data is clean and trusted in a way that they can use it without having to continuously go back and question their decisions.”
That foundation, Kotovets said, has made it easier for customers to apply the data to agentic AI use cases, where autonomous or semi-autonomous agents can take actions across workflows.
“Because the data is already trusted, there has not been a lot of testing and revalidating that had to happen,” he said. “The time to value for us and for the customer has been dramatically shorter than maybe for a lot of other organizations in this space.”
D&B.AI features
D&B.AI includes several core components, including a natural-language chat interface, standardized access to data through Model Context Protocol servers, and agent-to-agent connectivity that allows AI systems to communicate securely. Kotovets said Dun & Bradstreet has given customers multiple ways to use the platform.
“We’re giving them the options,” he said. “You can use the data in MCP for your agents. You can use our agents, and then we also built an enterprise interface that interacts with our data.”
The company has also begun rolling out purpose-built agents aimed at specific business workflows.
“We’ve launched really three agents,” Kotovets said. “So one is around KYX – know your counterparty or know your customer. Another one is credit analysis. And the third is a sales call prep agent.”
Those agents are being deployed externally by customers and also internally at Dun & Bradstreet, where AI is used for tasks such as customer service support, code assistance, and data management. Kotovets said the company uses AI extensively across its own data operations.
“We use LLMs and gen AI across multiple parts of the data management workflow,” he said. “One is for data acquisition,” including extracting information from documents, websites, and third-party sources.
He said AI is also used to improve data completeness, such as identifying missing corporate email addresses or domains for company executives. “That would be like a data quality improvement capability.”
Despite its push into AI, Dun & Bradstreet has taken a cautious approach to how its proprietary data is used with large language models. Kotovets said the company does not allow its data to be used for model training.
“We have traditionally, or up to this point, allowed for RAG (Retrieval Augmented Generation) and some aspects of grounding, but not for training,” he said. While there are limited scenarios where training might be explored in secure, customer-specific environments, those arrangements would include controls to shut off access if a customer relationship ends.
“Overall, more generally speaking, we have not allowed it,” Kotovets said.
Governance on tap
As for challenges encountered in building the platform, he said “one is around governance and compliance and ensuring that we have the right frameworks in an operating model, as well as the technological environment in which we would feel comfortable having these LLMs interact with our data.”
He said the company set up a governance structure comprised of review committees that monitor legal, compliance, ethics, cybersecurity, and product issues, along with oversight of language models approved for use. Dun & Bradstreet takes a multi-model approach, using systems such as ChatGPT, Gemini, and Llama depending on the use case.
Accuracy and transparency are also built into the system, according to Kotovets.
“We have full lineage,” he said. “You can literally click on the answer and see where it came from and what the original data elements that were used in the answer and the source of those data elements.”
Kotovets said AI is being applied across many industries, not just financial services. Use cases span telecommunications, manufacturing, pharmaceuticals, insurance, and supply chain management; it is particularly helpful for automating repetitive, labor-intensive tasks.
“Some clients see in terms of productivity gains or efficiency gains anywhere from 50% to almost 70% to 80%,” he said.
Looking ahead, Kotovets said agentic AI will increasingly reshape how work gets done, as multiple agents from different vendors are linked together into end-to-end workflows. “We’re seeing a lot of that happening all over the place – this idea of stringing these agents together into a comprehensive workflow is really rapidly evolving.”
In those environments, he said, Dun & Bradstreet’s business identifier plays a critical role.
“The D-U-N-S number becomes the connective tissue between the multiple agents,” Kotovets said, allowing systems to know they are referencing the same entity as information moves across platforms.
Asked what advice he would give business leaders deploying AI, Kotovets pointed to the importance of thinking holistically.
“Think about and look at your adoption journey,” he said. “Think about what is it that you have to take into account across your technology, your data integration, your tools, and your people.”
