Nearly every enterprise is investing in AI, but only 5% say their data is ready

Nearly halfway into 2026, enterprises are beginning to see tangible returns on their AI investments. Yet many are discovering that scaling requires something far less glamorous than flashy frontier models and state-of-the-art benchmarking: Clean, interoperable, governed data.

According to a new AI Momentum Survey from Dun & Bradstreet, 97% of organizations report active AI initiatives, but just 5% say their data is ready to support them.

This reflects the messy reality of AI as enterprises struggle to move beyond experimentation to operationalization.

“You do not need enterprise-wide AI-ready data to launch pilots or isolated AI use cases,” said Cayetano Gea-Carrasco, Dun & Bradstreet’s chief strategy officer. “But you do need it to scale AI reliably across mission-critical workflows and systems.”

Early gains seen

Organizations are all-in on AI in 2026 and view it as a mission-critical imperative, according to the D&B report. Well over half (67%) are seeing “early signs or pockets” of ROI, and 24% report “broad or strong” returns.

Further, more than half (56%) of the 10,000 businesses polled by the data and analytics firm say they are planning to increase AI investment in the next 12 months. Around one-third (30%) are scaling AI into production and 26% are operationalizing the technology across multiple core processes.

As adoption rapidly increases, early returns are more common now than even just a year ago, D&B noted, but they still remain uneven. Dovetailing with this, concerns around data readiness are “even more profound” than in 2025.

This is for a variety of reasons, including problems with access to data (reported by 50% of those polled by D&B), privacy and compliance risks (44%), and data quality and integrity concerns (40%). Further, 38% report lack of integration across systems, while 37% say there is a shortage of qualified AI professionals.

Concerningly, however, just a small number of enterprises (10%) say with high confidence that they are able to identify and mitigate AI-related risks.

“The key question is no longer whether organizations are experimenting with AI,” said Gea-Carrasco. “It’s whether they have the data and infrastructure required to deploy AI reliably at enterprise scale.”

He noted that it’s relatively easy for enterprises to launch copilots, chat interfaces, or departmental AI tools using general-purpose models and get “impressive results in a controlled environment.” But far fewer are able to deploy AI into production workflows, where accuracy, accountability, explainability, interoperability, and consistency directly impact business decisions. This includes areas like onboarding, compliance, risk management, and customer operations. “That’s where data readiness becomes critical,” said Gea-Carrasco.

The data hurdle

The challenges around data are only compounded as enterprises move from copilots to more autonomous agentic workflows. “Most enterprise data environments were built for human workflows, not autonomous AI systems operating continuously across the business,” he pointed out.

While AI systems can produce outputs that sound coherent, they can be difficult to trust operationally, due to hallucinations, conflicting recommendations across systems, and compliance issues, Gea-Carrasco noted. This is problematic for all enterprises, but particularly for those in regulated industries like banking, insurance, healthcare, and financial services, where trustworthy and auditable outputs are “non-negotiable.”

Organizations seeing the most progress are those working to ensure that their data is high-quality, reliable, and governed. They are investing in consistent identity resolution and data interoperability and maintenance, so that AI can “reliably consume” and act on information, he explained.

Where enterprises are seeing ROI

Enterprises are beginning to see ROI in areas where underlying data environments are more mature, thus making it easier for AI to be directly embedded into real workflows, according to Gea-Carrasco. This includes areas like sales intelligence, onboarding, compliance workflows, customer research, risk analysis, workflow automation, prospecting, screening, supplier evaluation, and business verification.

ROI is typically reflected in reduced manual research, faster onboarding and review cycles, improved operational consistency, accelerated sales workflows, and better decision support for employees, he said. “In many cases, organizations are using AI to help teams process and synthesize large amounts of information significantly faster than before.”

He emphasized that AI is most successful when it augments existing operational processes rather than fully replacing human decision-making. “Organizations are finding success where AI helps employees work faster, make better decisions, and it reduces repetitive manual work while humans remain involved in oversight and final approvals,” he said.

Enterprise approach to agentic AI

Agentic AI is beginning to enter production environments, although it is “still relatively early and targeted,” Gea-Carrasco pointed out.

Most enterprises today are deploying agents that are narrowly scoped rather than fully autonomous, he said. The near-term pattern is supervised autonomy, where agents execute portions of workflows while humans remain involved in approvals, oversight, and exception handling. Thus, agents are entering what he referred to as “clearly defined workflows,” such as research, onboarding support, and workflow orchestration.

Over the next several years, AI will move from standalone copilots to more connected agentic systems embedded directly into enterprise workflows, he noted. They will increasingly coordinate work across customers, suppliers, partners, employees, and enterprise apps. Agents will likely become ever more prominent in workflows around sales operations, onboarding, compliance, procurement, customer research, risk management, supplier evaluation, and monitoring.

“Enterprise AI is becoming less about isolated productivity tools,” said Gea-Carrasco, “and more about building intelligent operational systems that can support decision-making and workflow execution at scale.”