Enterprises need governed data foundations to move AI to production

As enterprises move from experimenting with generative AI to deploying agents and AI-driven applications in production, the challenge is increasingly shifting from model capability to the underlying data architecture.

Organisations need AI systems to work across fragmented data environments while maintaining governance, security, interoperability and control. The challenge is particularly pronounced in banking, financial services, telecommunications and government, where data is distributed across on-premises infrastructure, private clouds, hyperscalers and specialised platforms.

“Over the last couple of years, there has been an evolution from data to decision. The view has shifted from collecting data to having decision-grade data,” says Remus Lim, Senior Vice President for Asia Pacific and Japan, Cloudera.

Decision-grade data needs to be both fresh and trustworthy, he says, particularly as AI systems increasingly consume enterprise data and use it to generate recommendations or actions.

From data consolidation to interoperability

The enterprise data landscape is also becoming less amenable to the idea of bringing everything into a single repository. Instead, organisations are looking at how data residing across different environments can be accessed and governed consistently.

“Enterprises and public governments will have different data platforms. The era of bringing everything into one place is gone. The question is how we build interoperability across the platforms to ensure that you can move from one place to another or leverage all those things,” says Sergio Gago, Chief Technology Officer, Cloudera.

For India, this shift is playing out across sectors handling large and increasingly complex data estates.

“India is one of the largest and fastest-growing geographies for us across the globe. A lot of momentum is happening across the industry, primarily driven by the volume of data,” adds Mayank Baid, RVP for India and South Asia, Cloudera. He points to banking, telecom, fintech, capital markets and government as areas where organisations are building data foundations for AI. According to Baid, three-fourths of India’s largest banks use Cloudera’s platform.

AI agents raise the access challenge

The emergence of agentic AI adds another layer of complexity. Unlike conventional applications, agents may need to interact with multiple data sources and systems to complete a task.

That creates a tension between access and control. Restricting agents too heavily can limit their usefulness, while broad access without appropriate controls can create security, compliance and operational risks.

“An agent needs full access, but full control on that axis,” affirms Gago.

This requires permissions, lineage, governance and orchestration to work together. The issue becomes more acute when an AI application moves beyond experimentation into production.

Gago says some organisations have approached AI experimentation by copying enterprise data into separate environments, including vector databases. While this can help with experimentation, production deployment raises questions around data provenance, access rights, auditability and cost. “Without those answers, no enterprise will deploy AI into production,” he points out.

Sovereignty adds another layer

The rise of sovereign AI and sovereign cloud is further complicating infrastructure decisions for regulated organisations. The question is no longer simply whether workloads should run on-premises or in the public cloud, but how organisations can retain control over data while accessing the broader AI ecosystem.

“We bring AI to data rather than pushing data all the way to AI,” says Baid.

For Gago, however, sovereignty should not result in technological isolation. Enterprises also need the flexibility to use different models, infrastructure providers and deployment environments as requirements evolve. “What we recommend precisely is optionality. You still want to use the latest models because they are state of the art. But you also want to be able to move from one hyperscaler to another, or from a hyperscaler to a sovereign cloud provider or a data centre.”

This flexibility is becoming more significant as AI systems become embedded in business-critical processes. Dependence on a single cloud, model provider or infrastructure environment can create operational and business-continuity considerations.

Infrastructure must keep pace with agents

The next challenge is scale. Existing enterprise infrastructure has largely been designed around human users and predictable application workloads. Autonomous agents could significantly increase the volume and frequency of data queries and system interactions.

“The vast majority of infrastructure systems that are out there today are not ready for agents,” says Gago.

For CIOs, the implication is that AI readiness cannot be separated from data architecture. Moving AI from pilots to production requires trusted data, interoperability, controlled access and infrastructure capable of supporting increasingly autonomous workloads.

Comments (0)
Add Comment