Share with your CDO
TransUnion’s Chief Data and Analytics Officer Dinesh Chawla is making a direct case that enterprise AI has hit a data ceiling, not a model ceiling. Drawing on McKinsey data showing 88% of organizations use AI in at least one function but only 7% consider their data fully AI-ready, Chawla argues that fragmented data, inconsistent quality, and governance gaps are the real blockers to scaling AI from pilots to production. TransUnion’s own approach pairs Google Cloud’s Gemini models with proprietary identity and fraud data, treating the combination as the competitive asset rather than the model alone.
What this means for your business
The 7% figure is the tell. If your organization is in the 93% and you’re still approving AI pilots, you’re not behind on models, you’re behind on the prerequisite. The deciding trait here isn’t industry or company size, it’s whether your data governance covers what AI agents can do with data, not just who can see it. Companies whose governance stopped at access controls are operating with a framework that was designed for a world where humans made the decisions and AI just surfaced information.
Chawla’s point about agentic AI sharpens the stakes considerably. A generative AI tool producing a flawed summary is a nuisance. An AI agent acting on incomplete or stale data across credit, fraud, or risk workflows is a liability event. The shift from “AI that answers” to “AI that acts” means data lineage, the audit trail showing where data came from and how it was transformed, stops being a compliance formality and becomes an operational requirement. Enterprises that haven’t invested in metadata management and master data governance (the discipline of maintaining a single authoritative record for customers, products, and transactions across systems) are not just slowing their AI programs, they’re accumulating unpriced risk.
The harder organizational truth buried in Chawla’s framing is that the technical infrastructure is rarely the actual bottleneck. Getting agreement on common definitions of a “customer” or a “transaction” across business units, legacy systems, and acquired platforms is a political problem dressed as an architecture problem. CDOs who treat data readiness as a tooling exercise will keep hitting the same wall. The leading indicator to watch is whether your AI governance charter has been updated to include agent identities, decision auditability, and human escalation thresholds. If it hasn’t, your agentic AI timeline is optimistic.
Concept deep-dive: Agentic AI
Agentic AI refers to systems that don’t just respond to queries but take sequences of actions autonomously, such as retrieving data, calling APIs, updating records, and triggering downstream workflows, to complete a goal. Think of the difference between a search engine and an employee who goes and does the thing you searched for. The business implication is significant: every governance assumption built for human decision-makers needs to be rewritten for a system that acts without pausing to ask permission.
Based on reporting from AI ambition is running ahead of data readiness: TransUnion’s Chief Data & Analytics Officer, originally published 2026-09-07 11:13:00.
