AI Models Are Commoditizing: Your Data Isn't
Last quarter, a logistics firm I advise deployed an agentic AI system to automate supplier contract negotiations. Within weeks, the agents had renegotiated terms with three major suppliers, acting on customer demand signals, product margin data, and supplier performance scores pulled from across the enterprise. But the customer records in the CRM conflicted with those in the ERP, the product margin figures hadn’t been reconciled since a platform migration eighteen months prior, and the supplier scores reflected ratings from a division that had been reorganized out of existence. The agents performed exactly as designed, which turned out to be the problem. They executed with confidence on data that didn’t deserve it.
The technology worked. The asset underneath it had never been managed as one, and agentic AI exposed that gap at operational speed.
The Emerging “Best AI Wins” Fallacy
Most executives still assume that competitive advantage in AI accrues to whoever has the best models. The economics increasingly argue otherwise. LLM inference prices have declined at a median rate of 50x per year across major benchmarks, according to Epoch AI, and API pricing across major providers dropped roughly 80 percent between 2024 and 2026. The trajectory shows no sign of leveling off..
For anyone who remembers the early days of cloud economics the pattern will be familiar. Infrastructure differentiation collapsed within a few years; competitive advantage migrated up the stack to applications and data. The same migration is happening again, but downward, to the data layer. When every competitor can deploy equivalent agents at comparable cost, the differentiating variable becomes the quality, governance, and contextual richness of the data those agents consume.
The Asset That’s Seldom Managed Like an Asset
I have spent much of my career arguing that data meets every standard economic definition of an asset: a resource with measurable value, capable of generating future economic benefit, and subject to management practices that either enhance or erode that value over time. The Infonomics framework I developed quantifies this through intrinsic, business, and performance value dimensions. Yet most enterprises manage data with less rigor than they apply to office furniture depreciation schedules. That was always a waste. Now it is a liability.
Autonomous agents making real-time decisions about customers, suppliers, and products require data that carries not just accuracy but organizational context: relationships between entities, hierarchies, lineage, interaction history across domains. An agent negotiating a supplier contract needs to understand how that supplier connects to specific product lines, customer segments, and regulatory obligations simultaneously. That requires continuous entity resolution across systems, master data that reconciles conflicting records into a single trusted view and keeps it current as the business changes. Static data quality, in other words, is necessary but insufficient. What agents require is governed, unified master data that functions as a contextual model of the business itself.
Mind the Governance Gap
The global CDO study I recently co-led with EY, combining in-depth interviews and surveys of data, analytics, and AI leaders across five continents, quantifies how wide this gap has become. Eighty-five percent of CDOs rank data as a major challenge in deploying AI, and the barriers they cite are not technological: change management, data governance, and data quality top the list. As one banking-sector CDO put it: “We’ve spent 20 years building data for eyeballs; we now have 20 months to rebuild it for AI.”
The bifurcation is already visible. Organizations that have invested in continuously unified, governed, real-time data across domains, with entity resolution and master data management built into a cloud-native architecture, can deploy agents that act with warranted confidence. Organizations still running on legacy platforms designed for batch reporting and static compliance are discovering that those architectures cannot serve agents operating in real time. A data warehouse refreshed overnight is not the same as a data foundation delivering trusted, contextually rich records to agents in milliseconds, and no amount of AI tooling compensates for that structural gap. Top-performing organizations in our study invest roughly half again as much as the rest in using AI within data management itself, and they put more into defensive capabilities like governance and quality, treating both as preconditions for scale rather than overhead. Sixty percent of these top performers received budget growth exceeding 11 percent in the past year, versus 35 percent of laggards. That funding gap is self-reinforcing. Measurement credibility attracts further investment, which widens the maturity advantage.
The most consequential finding, though, may be structural. Machine consumers of enterprise data, meaning AI agents, are imposing quality standards that human dashboard consumers were willing to tolerate. When a governance shortfall caused a report to be slightly off, someone caught it. When it causes an autonomous agent to act on flawed data at scale, the consequences surface faster and more publicly than any reporting error ever did. Ross Schalmo, CDO at Eaton, captures the operational reality: “If you can’t trust the data, you might as well not start building anything on top of it.”
The broader research tells the same story from different angles. An HBR Analytic Services survey found that 94 percent of respondents rate trust in data reliability as important to successful AI adoption, yet only 39 percent report high proficiency in delivering it. BARC’s Data, BI and Analytics Trend Monitor 2026, surveying nearly 1,600 practitioners, found data quality management reclaiming the top priority position, driven by AI adoption exposing what poor foundations actually cost. And MIT’s Project NANDA found that 95 percent of generative AI pilots failed to deliver measurable business impact.
Who Owns “The Data Versus AI” Problem
Chief Data and AI Officers face a mandate that has shifted from “build AI models” to “make enterprise data AI-ready,” where governance, cross-domain unification, and real-time contextual delivery are the deliverables. The EY study found that top performers rate data as more strategic than AI and dedicate significantly more AI capacity to managing data itself. That priority inversion is worth sitting with. The organizations furthest ahead on AI got there by investing first in data.
CEOs and boards need to ask a question most have not yet even conceived yet. What is the economic value of our enterprise data, and is it architected for real-time autonomous consumption? For CFOs and investors, the calculus is becoming concrete. Data asset maturity will increasingly separate valuation premiums from valuation risk, as autonomous decision-making exposes data quality failures at speed and scale.
As one CDO in the study observed: “In the AI era, the quality of your data is the only unfair advantage you have left.” The next wave of enterprise value creation will not come from who has the best AI, but from who has the best data for AI: governed, unified across domains, delivered in real time, and rich with organizational context. The companies that treated data as an economic asset before it was fashionable are about to collect on that investment.
