Gary Kotovets, Chief Data, Analytics, and AI Officer at Dun & Bradstreet.
As enterprises accelerate their transition from AI as a querying tool to AI as an autonomous executor, the fundamental role of data within organizational architecture must evolve in kind. The passive data layer, a reservoir sitting beneath the application stack and waiting to be queried, is no longer sufficient to meet the demands of agentic systems.
Agents tasked with onboarding suppliers, approving customers, flagging risk exposures or recommending next-best actions require more than access to raw information. They need enriched, trusted business context delivered precisely at the moment a decision is being considered, with enough confidence and clarity to support action inside the workflow itself.
Provenance separates context from noise.
Not all data is equal. What separates useful context from noise in an agentic workflow is provenance, meaning a clear understanding of where the data originated, how it was verified and whether it is authoritative enough to serve as ground truth for the decision at hand.
This is where the concept of the data originator as truth becomes operationally significant. When an AI agent needs to evaluate a counterparty, assess a supplier’s risk profile or validate a business relationship, the only data worth trusting is data that has been verified at its source, structured around a persistent business identity and enriched with the commercial relationships that define how that entity actually operates in the world.
Persistent identity becomes the anchor.
In commercial contexts, that source of truth should begin with a persistent business identifier. Such an identifier is not simply a reference code; it is an anchor point that connects a verified legal identity to the broader web of attributes, relationships and signals that define a business.
That connected view is what many organizations are beginning to understand as a commercial graph. At its best, a commercial graph links business identity, ownership structure, financial health signals, trade behavior, operational history and supply chain relationships into a structured view of how an entity participates in the global economy.
In this model, a commercial graph can also provide deterministic answers to core identity and linkage questions. An agent should not have to infer probabilistically what trusted business context can establish in advance.
For an AI agent, this distinction is critical. When an agent evaluates a business record anchored by a persistent, verified identifier, it is not acting on loosely matched names, scraped web data or self-reported attributes alone; it is acting on grounded business context that has been organized around a durable identity.
Data must become decision infrastructure.
This reframes data not as a static input, but as dynamic decision infrastructure. It becomes an operational substrate that must be woven into the execution layer itself rather than retrieved from the periphery only after a user or system knows which question to ask.
The shift from retrieval to execution also changes the meaning of “next-best action.” In traditional analytics environments, next-best action is often treated as a recommendation surfaced to a human decision-maker. In agentic workflows, it becomes a governing framework for how autonomous systems evaluate options, apply policy and determine the appropriate next step.
Next-best action requires judgment.
That next step may be simple, such as routing a supplier for additional review. It may also be consequential, such as recommending customer terms, flagging an ownership change, escalating a compliance risk or identifying a disruption in a critical supply chain relationship.
These are not merely search problems; they are judgment problems. And judgment requires more than access to information; it requires context, confidence, thresholds and an understanding of relationships that may not be visible in any single data point.
This is why accuracy, provenance and the identity of the data originator become non-negotiable as AI agents move from answering questions to executing high-stakes business decisions. Truth in an agentic workflow must be grounded and deterministic, not approximated, because the cost of approximation rises sharply when software is empowered to act.
Leaders must rethink data strategy.
Enterprise leaders should therefore begin rethinking their data strategies not as a foundation beneath AI, but as an integrated component of the AI execution layer. Data should participate actively in every decision cycle, carrying with it verified identity, originator-backed intelligence, relationship context and risk signals that agents can apply in real time.
That requires new scrutiny of the data environment. Leaders must ask not only whether information is available, but whether it is authoritative, current, connected and fit for autonomous decisioning.
They must also ask whether their systems can distinguish between a company name and a verified business identity. Without that distinction, AI agents may confuse entities with similar names, miss relevant ownership relationships or act on incomplete risk signals hidden several layers away in a corporate family or supply chain.
We are entering the next frontier of enterprise AI.
The organizations that solve this problem are the ones I believe will be better positioned to move AI beyond copilots and chat interfaces into true operational execution. Their agents will be more capable of distinguishing signals from noise, applying business context at the moment of decision and recommending next-best actions with a clearer understanding of the commercial reality surrounding each entity.
As AI moves deeper into procurement, credit, compliance, sales and customer operations, autonomous systems will need more than access to information. They will need access to verified business truth, structured through persistent identity, enriched by a commercial graph and embedded directly into the workflows where decisions are made.
This is the next frontier of enterprise AI. The new industry leaders will not be those with the largest reservoirs of data, but those with the most trusted decision infrastructure.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?