Dun & Bradstreet CDAO: How a 185-Year-Old Firm is Leaning into the Agentic Era

TLDR

• Dun & Bradstreet is adapting to the agentic AI era, believing that deployment of autonomous AI agents will increase the need for trusted business identity rather than replace commercial data providers.
• Enterprise AI adoption is accelerating, but fragmented data remains the biggest barrier to scaling AI agents. Only 6% of organizations say their data is fully ready for AI at scale.
• The company is preparing for an agent-to-agent economy by making its data, workflows and AI agents available through MCP while exploring whether AI agents themselves may eventually need unique business identifiers.

As agentic AI risks upending many business models, Dun & Bradstreet is betting AI agents will make its business more valuable.

For a company whose business has revolved around helping organizations identify, verify and evaluate customers, suppliers and business partners for 185 years, the rise of agentic AI might appear to offer a rival capability.

Modern AI agents can already search the web, analyze SEC filings, reconcile information from multiple sources and answer many of the questions companies once relied on commercial data providers to solve.

Dun & Bradstreet believes its services are tough to replicate robustly.

“We collect data from 20,000-plus sources. We synthesize it, we clean it, standardize it, normalize it – all of that work that you otherwise would have to do on your own, plus huge proprietary data sets that we have that cannot be replicated,” Gary Kotovets, Dun & Bradstreet’s chief data and analytics officer, said in an interview with The AI Innovator.

The company believes the rise of autonomous AI agents will make trusted business identity more important than ever. As enterprises increasingly deploy agents to negotiate with suppliers, assess customers and automate business processes, those systems first need confidence that the company appearing in one application is the same company appearing in another.

That doesn’t mean Dun & Bradstreet isn’t adapting to the agentic era. It is expanding its role as a commercial data provider to add what it calls an identity and context layer for enterprise AI, exposing its data, workflows and proprietary AI agents through Model Context Protocol, or MCP, so enterprise agents can retrieve verified business information directly.

Instead of requiring an employee to open a Dun & Bradstreet application to verify a company or assess a supplier, for example, an agent can retrieve that information as part of its own workflow. Dun & Bradstreet is also making available some of its existing processes – including business verification, entity matching and know-your-customer checks – so customers can incorporate them into their own agentic systems.

Enterprise AI momentum accelerates

The strategy comes as enterprises move beyond AI experimentation into production deployments. According to Dun & Bradstreet’s latest AI Momentum Survey of 10,000 businesses across 32 countries, more than three-quarters of organizations now report measurable returns from AI.

Nearly half said they are seeing “pockets of ROI,” while another 28% reported broad or strong returns across multiple initiatives. At the same time, only 6% said their enterprise data is fully ready to support AI at scale, suggesting many organizations are deploying AI faster than they are preparing the data those systems depend on.

“The tools are there to build your AI capabilities, but there’s a lot of work that needs to happen to take the data out of these fragmented systems that exist across the enterprise and putting it in place that is centralized, that is accessible by your AI tools,” he said. “The big problem is rationalizing and cleaning that data up to a point where you feel comfortable that your AI tools can execute tasks autonomously with a high degree of confidence.”

That gap is becoming the defining challenge of enterprise AI.

“We’re in an environment where there’s more adoption and more clarity on what the ROI is,” he said. “People seem to be running faster in the deployment of AI capabilities, and sometimes they’re running almost ahead of the overall enterprise readiness.”

The momentum reflects a broader change in how organizations are adopting AI. Early experimentation has given way to wider deployment as easier-to-use coding assistants and enterprise AI tools lower the technical barriers for employees across organizations.

Engineering teams are using coding assistants such as Cursor or Codex while other departments are using tools like Copilot or various chatbots. These two trends are converging to result in wider adoption across a company, he said.

The rise of ‘agent swamp’

Success with scaled AI deployments, however, has been uneven.

Many companies are deploying thousands of internally developed AI agents without fully understanding whether those agents are delivering measurable business value.

“I hear things like ‘agent swamp,’” Kotovets said. “People talk about overbuilding agents and then nobody’s really using them. They’re kind of like zombie agents walking around.”

Organizations are seeing stronger returns when AI projects target specific business functions, such as finance or risk management, rather than attempting enterprise-wide transformation all at once, he said. Those targeted deployments explain why so many organizations report pockets of ROI instead of broad gains across the enterprise.

The larger obstacle, however, is not building AI agents but preparing the data they rely on.

For decades, enterprises have stored customer, supplier and financial information across disconnected systems such as Salesforce, SAP, contract management platforms and internal databases. Those systems often refer to the same company in different ways, making it difficult for AI agents to confidently determine that multiple records belong to the same business.

“The agent does not understand always that the name of company ABC that’s in ERP is the same company that’s sitting in the CRM,” Kotovets said. “We help you connect those two together and say this is the same company.”

Connecting disparate data

That identity problem lies at the center of Dun & Bradstreet’s strategy.

The company maintains what it calls a commercial graph containing information on approximately 650 million businesses worldwide, collecting up to 11,000 data elements on each company. Those records include corporate officers, ownership structures, supplier relationships, financial indicators and risk scores. Customers match their internal business records to a D-U-N-S Number, which Dun & Bradstreet then enriches with verified commercial data.

The D-U-N-S Number, introduced in 1963, has long served as a unique identifier for businesses. Kotovets believes it now has a second life as AI agents begin operating across enterprise applications.

A company may exist as a supplier inside SAP, a customer inside Salesforce and a legal entity inside a contract management platform. Employees understand those records describe the same organization. AI agents frequently do not.

By matching those records to a common identifier, enterprises can connect Dun & Bradstreet’s verified commercial data with their own internal information, such as purchase histories, customer interactions and financial relationships.

Access 3 layers through MCP

To make that information easier for AI systems to consume, Dun & Bradstreet recently expanded its use of Model Context Protocol (MCP), a popular industry standard that lets AI assistants connect to external data sources and software so they can retrieve data and perform tasks.

Rather than simply making business records available, the company is making three layers accessible through MCP. Customers can retrieve commercial data, invoke Dun & Bradstreet’s business workflows or use proprietary AI agents the company has already developed, including business verification, entity matching and know-your-customer agents.

“We’ve provided all our data via MCP,” Kotovets said. “Not only can you get access to our data, you can get access to our workflows. And then the third layer is we give you all of the proprietary agents that we’ve built that sit on top of our data.”

Kotovets argues that make those capabilities accessible through MCP allows enterprises to incorporate Dun & Bradstreet’s expertise directly into their own agentic workflows instead of requiring users to switch applications.

While the emphasis on data quality may sound familiar – enterprises have struggled with fragmented and inconsistent data since the early days of cloud computing – AI has finally created enough economic incentives for businesses to finally solve its data readiness problems, he said.

“It’s the same problem,” Kotovets said. What’s different is that “there’s tremendous opportunity for the enterprise to realize an enormous amount of productivity gains as a result of AI. That pressure is different than the pressure they had with the cloud.”

D-U-N-S Number for agents?

Looking ahead, Kotovets believes the importance of trusted business identity will only increase.

As organizations deploy larger numbers of autonomous agents, those systems will generate growing volumes of synthetic data. That makes it even more important for enterprises to maintain what he described as a “golden source” of verified information that agents can reference before making decisions.

The company is already considering what that future might require.

Asked whether Dun & Bradstreet could eventually assign D-U-N-S Numbers not only to companies but also to AI agents representing those companies, Kotovets said the concept has become an active area of discussion.

“Agents from companies will interact with one another as counterparties,” he said. “Knowing and understanding what those agents are, which businesses they belong to, and what they actually do becomes very important for us. It is definitely a topic that we’re looking at, and it is going to be an important one for us to tackle.”