Press "Enter" to skip to content

Dun & Bradstreet CDAO: AI’s ROI Depends on Data Readiness

TLDR

  • Only 28% of businesses report broad or strong AI returns, even as 65% plan to increase spending, according to Dun & Bradstreet’s Q4 survey.
  • Companies with fully AI-ready data are 6.5 times as likely to report broad or strong returns as those with partially ready data, according to the company.
  • Dun & Bradstreet CDAO Gary Kotovets said businesses are becoming more selective about AI investments, focusing on a few high-value applications rather than deploying AI broadly.

Companies are pouring money into AI, but most have yet to see “broad or strong” returns on their investment, according to a new global survey by Dun & Bradstreet. One of the biggest obstacles has little to do with the technology itself: Their data isn’t ready.

The fourth-quarter AI Momentum survey found that 65% plan to increase spending over the next 12 months but only 28% of businesses reported “broad or strong” returns on their AI investments. (The company did not specify the financial thresholds used to classify those returns.)

The gap is even more pronounced among companies with differing levels of data readiness. Those with fully AI-ready data were 6.5 times as likely to report broad or strong returns as businesses whose data was only partially ready, according to the survey.

🔍
Meet Sherlock AI Need more clues? Ask the Sherlock chatbot to summarize this story, explain technical concepts or answer other questions.

In an interview with The AI Innovator, Dun & Bradstreet Chief Data and Analytics Officer Gary Kotovets said the findings reflect a persistent problem: Companies are buying AI models, platforms and applications before ensuring the information feeding those systems is accurate, consistent and trustworthy. This has a direct impact on ROI.

“There is much wider adoption of AI in the enterprise,” he said. However, “people are still struggling to get a significant amount of ROI for different reasons, whether it’s measuring the ROI or holding themselves accountable within the organization to get kind of the ROI they originally signed up for.”

The problem is hardly new. Businesses have spent decades trying to consolidate information scattered across databases, enterprise resource planning systems and customer relationship management software.

Previously, clean data was needed primarily for business intelligence dashboards, compliance systems and other applications. That’s still true. But now companies are also deploying AI agents that can analyze information, execute workflows and make decisions without human intervention.

That raises the stakes for getting the data right: only 8% of survey respondents had fully AI-ready data, up from 6% in the third quarter. While 33% say their data is mostly ready, it was down from 36%. Another 44% said their data was partially ready.

“Right now there is a much more, I would say, obvious need to get the data to a place to activate any of these AI capabilities,” Kotovets said.

Kotovets said businesses are discovering that making data usable for AI requires extracting it from different systems, resolving inconsistencies, identifying duplicate records and establishing a reliable foundation for automated decisions.

It’s a challenge that is complicated by a disconnect he has seen between executives at the top pushing for AI adoption and employees tasked with deploying AI in applications that generate ROI.

Data readiness separates the winners

Dun & Bradstreet’s quarterly survey of 10,000 businesses across 32 countries found that 98% of all respondents are deploying AI, while 37% said they were scaling pilots into production, up from 34%. Yet only 28% report seeing broad or strong returns.

The findings suggest that deploying more AI does not guarantee higher ROI.

Kotovets said he sees companies reconsidering their approach. Instead of distributing AI projects across every department, some are concentrating resources on two, three or four large enterprise functions where productivity gains or cost reductions could have a more substantial impact.

“People are being much more strategic as opposed to just running around and pushing it everywhere and hoping for the best,” he said. “They’re being much more calculated.”

On AI investments: “People are being much more strategic as opposed to just running around and pushing it everywhere and hoping for the best.”

Businesses are also scrutinizing productivity tools they rolled out to employees more closely, to ensure these bring business value.

For example, a company may know that employees frequently use Microsoft Copilot but still struggle to determine whether the investment has improved productivity enough to justify its cost.

Computing expenses are another concern. Kotovets said companies sometimes discover that the cost of operating AI systems exceeds the savings they expected.

For example, “people are saving $2 and and spending $5 on compute,” he said. “That kind of realization is also happening, and so there’s a lot of people now thinking about how else can I optimize this cost?”

Relationships are all you need

Dun & Bradstreet is seeking to capitalize on the growing demand for AI-ready data through its Commercial Graph, an interconnected database that identifies businesses and maps their relationships with other companies, individuals, physical assets and products. It covers data on more than 650 million entities, according to the company.

The system is built around Dun & Bradstreet’s D-U-N-S Number, a unique business identifier used by more than 600 million companies globally. By assigning consistent identities to companies across different information systems, the identifier helps resolve a common enterprise data problem: determining whether records that appear different actually refer to the same business.

Consider a salesperson who misspells a supplier’s name in a CRM system. The same supplier may also appear in an accounting or procurement database under a different name. Without a reliable way to match the records, the AI system might treat the two as separate businesses.

Kotovets said companies typically address such problems by extracting information from their systems and moving it into data platforms such as Snowflake or Databricks, where they attempt to clean, reconcile and standardize the records.

But “that process is long and tedious,” Kotovets said.

Dun & Bradstreet’s approach is to provide a layer of verified business information that can help resolve those identities and relationships before an AI application begins processing them.

The Commercial Graph goes beyond identifying individual companies. It also maps connections among businesses, their owners, executives and other entities.

For example, a financial institution evaluating a prospective borrower may need to establish not only the company’s identity but also its ownership structure, related businesses and connections to other individuals or organizations.

Similarly, an AI agent assessing suppliers could use the information to identify corporate relationships that might otherwise be missed.

Data that’s 30% to 40% ready

Kotovets said mapping these relationships is particularly important for applications involving know-your-customer and know-your-business checks, supplier onboarding, fraud detection and credit decisions.

The company describes its approach as a deterministic layer, meaning that business identities and relationships are established through predefined processes and validated information rather than being inferred independently by a generative AI model.

That distinction matters for autonomous agents, which may be expected to execute tasks without seeking human approval at every step.

An agent relying on inaccurate company records could misidentify a supplier, overlook a corporate relationship or reach an incorrect risk assessment. Giving it access to verified business information can reduce those risks, although it does not eliminate the need for oversight.

Kotovets said Dun & Bradstreet’s existing data infrastructure has also allowed the company to accelerate its own AI development.

Because much of its information was already standardized using D-U-N-S identifiers, the company could deploy new AI agents and workflows into production in weeks, he said, rather than first undertaking extensive data preparation.

Kotovets said the company has redesigned its product offerings around its data foundation, seeking to help customers adopt AI more quickly while retaining the reliability they associate with its commercial information.

However, the Commercial Graph is not a universal solution to enterprise data quality. Its strength lies in identifying businesses, connecting relationships and enriching commercial records. Companies must still address problems involving their own operational data, including inaccurate inventory information, inconsistent internal pricing or incomplete transaction records.

Despite the emphasis on data readiness, Kotovets does not believe every company must completely overhaul its information infrastructure before benefiting from AI. The required level of preparation depends on the application and the reliability of the results it needs to produce.

A company building an AI assistant for a limited internal function, for example, may not need to standardize every record across its organization. It can focus on preparing the information necessary for that particular task.

Kotovets suggested that some applications might operate satisfactorily with only 30% or 40% of the relevant data prepared, provided the results consistently meet the organization’s requirements.

Author

Get the latest insights about enterprise AI.

Subscribe to our newsletter. Thank you.

Be First to Comment

Leave a Reply

Your email address will not be published. Required fields are marked *

×