Enterprise AI adoption is outpacing the development of necessary data infrastructure, creating a bottleneck for companies seeking to move from pilot projects to production. According to a report by QuantumBlack AI by McKinsey, 88% of organizations globally are using AI in at least one business function, but only 7% say their data is fully AI-ready.
Dinesh Chawla, executive vice president and chief data and analytics officer at TransUnion, notes that the biggest barrier to scaling AI is no longer the model itself, but rather the readiness, governance, and trustworthiness of enterprise data.
AI’s Data Problem
The issue is shifting increasingly away from access to advanced AI models toward ensuring those models can consistently access the correct data, interpret its context, and function within suitable safeguards. Cloudera estimates that 56% of global organizations identify data silos as a key barrier to unlocking AI value.
Agentic AI intensifies the problem, as systems are expected to make choices and execute actions across business operations. Chawla observes that when data is scattered, confidence in such actions diminishes rapidly.
Chawla states that the inability of AI initiatives to move beyond testing to full deployment stems more from the operational environment than from model capabilities. While enterprises can access powerful foundational models, scaling them demands data integration, quality control, oversight, and accountability.
Obstacles to AI Scaling
Chawla identifies three enduring challenges: fragmented data across organizational systems, subpar data quality and preparedness, and governance. Many AI pilots succeed because teams manually refine and organize datasets, but this method becomes unmanageable when AI is applied across various departments and workflows.
A world-class model cannot compensate for weak data quality, disjointed systems, or a lack of trust in outputs, Chawla said. For large enterprises, establishing a trusted data foundation extends beyond data quality to include lineage, metadata, identity, master data management, access restrictions, oversight, and real-time data availability.
The complex part, Chawla said, is often not the technology but forging shared definitions, ownership, and policies across business units, legacy systems, and acquired platforms. This makes building a unified view of customers, products, or transactions an organizational and architectural hurdle.
The emergence of agentic AI raises the governance challenge, requiring enterprises to regulate what AI systems can do with data rather than just controlling access. Chawla notes that oversight must shift from managing data access to governing data, decisions, and actions.
This requires assigning AI agents distinct identities and permissions, ensuring human review for high-risk choices, and maintaining full tracking of data accessed, decisions made, and actions executed. The governance gap is evident, with few organizations deploying agentic AI claiming to have mature oversight frameworks.
For TransUnion, where AI can influence decisions related to credit, fraud, identity, and risk, explainability is a core requirement rather than a trade-off against model performance. A highly accurate model that cannot be explained, validated, or governed has limited value in regulated and high-stakes environments, Chawla said.
The firm employs reason codes and model explanations alongside monitoring for bias and model drift, embedding oversight throughout the AI lifecycle. The goal is to enable business users, regulators, and consumers to comprehend and validate AI outcomes, with TransUnion adhering to this principle when deciding whether to develop internally or leverage external technology.
Chawla expects the company to retain ownership of capabilities that differentiate it, such as its proprietary data, identity intelligence, fraud insights, analytics, and governance, while using external foundation models and cloud infrastructure where they can accelerate development, as seen in its collaboration with Google Cloud.
Over the next 12-24 months, the focus will increasingly shift from building AI foundations to scaling them, with Chawla expecting greater adoption of AI-enabled analytics, automation, and agentic capabilities, alongside continued investment in cloud, data engineering, and platform capabilities, and the India GCC taking on greater AI ownership.
TransUnion’s India GCC is taking on a more significant role in the company’s AI strategy, moving beyond traditional technology and analytics delivery. The centre is now contributing to global platforms, data, and AI capabilities, including the development of OneTru, a centralized solution-enablement platform.
The India-based teams are working on various aspects of AI, cloud transformation, data engineering, fraud, security, and intelligent automation. This shift in role is enabling the GCC to drive innovation, solve complex business problems, and influence product roadmaps.
AI Readiness and Data Infrastructure
The company’s approach to AI readiness is focused on making data reliable, governed, and usable at scale. This involves combining external foundation models with proprietary data, domain expertise, governance, and semantic context. The collaboration with Google Cloud is an example of this approach, using Gemini models through Vertex AI to accelerate development.
The competitive advantage comes from leveraging external models and infrastructure, rather than building foundation models internally.
As the focus shifts from building AI foundations to scaling them, the India GCC is likely to play a deeper role in global products and platforms.
