The paradox requires approaches such as embedding governance, modernizing data architectures, and using modular AI to scale responsibly in regulated environments.

In the race to AI maturity, the financial services sector is navigating a paradox: While many organizations globally claim to have integrated AI into their core processes and have a clear data strategy, almost as many are claiming to be constrained by limited data access across complex environments.

Could this be a clue that ambition can outpace execution in the race to integrate AI?

According to Julia Tan, Regional Vice President (Southeast Asia and Taiwan), Cloudera, this gap matters because AI agents are speeding up loan approvals, automating client checks, and reducing onboarding times across banks in Singapore and globally. Yet, fragmented data environments in some cases continue to limit their ability to scale beyond isolated use cases. “Deploying AI alone does not guarantee returns. Without strong data foundations and embedded governance, even advanced models struggle to move beyond pilot stages. In regulated environments, success increasingly depends on how effectively AI is operationalized, making modern data architectures and consistent governance essential,”” Tan noted.

AI-adoption: sector success factors

Here are some key requirements for successful and responsible implementation of AI in the financial services sector that Tan is offering from her vantage point:

  • Build strong data foundations before scaling AI initiatives
  • Embed governance into AI workflows from the start
  • Modernize data architectures to support AI at scale
  • Ensure consistent governance across all data environments
  • Equip teams with secure controls while maintaining flexibility
  • When it comes to AI security, being “almost ready” is no longer sufficient but remains a systemic risk.

Despite decades of heavy investment in data infrastructure, many banks remain constrained by structural complexity: fragmented systems and manual processes that dilute returns and slow innovation. This is driving a shift toward more modular, low-overhead approaches to AI development, built on reusable frameworks and agents rather than bespoke systems. Recipe-based architectures that use specialized agents to streamline data discovery and preparation are emerging as a promising approach, with some notable benefits:

  • Simplifies how data is accessed and prepared
  • Accelerates time-to-value for AI deployments
  • Reduces deployment timelines
  • Minimizes manual effort in data preparation
  • Enables teams to focus on scaling innovation rather than managing infrastructure

This framework not only strengthens security measures, but also reduces false positives, improves response times, and lowers operational costs, all while reinforcing customer trust, the sector’s most critical asset.

Embed trust into every AI decision by allowing humans to retain responsibility for high-impact decisions:

  • Let AI execute tasks while humans validate outcomes
  • Maintain continuous oversight rather than episodic checks
  • Assign AI to high-volume, repetitive work (data preparation, reporting, pattern detection)
  • Keep humans responsible for high-impact decisions
  • Embed data lineage capabilities into workflows
  • Implement monitoring and policy enforcement directly in workflows
  • Create clear audit trails documenting who did what, when, and why

To enable this balance, organizations are embedding capabilities such as data lineage, monitoring, and policy enforcement directly into workflows. These built-in controls reduce reliance on manual checks while improving transparency through clear audit trails of who did what, when, and why.

Bridge the governance gap: To deploy AI safely, consistently, and with regulatory confidence:

  • Evolve governance from a control function into an enabling layer
  • Apply policies directly within workflows for real-time compliance
  • Keep data usage aligned with regulatory expectations from the outset
  • Ensure model outputs meet regulatory requirements before deployment
  • Integrate oversight into execution rather than after the fact
  • Give compliance teams visibility without slowing innovation

To address the gaps, governance needs to evolve from a control function into an enabling layer. Organizations are exploring agents that apply policies directly within workflows, ensuring compliance in real time rather than after deployment. This keeps data usage and model outputs aligned with regulatory expectations from the outset. By integrating oversight into execution, banks gain both speed and assurance, giving compliance teams visibility without slowing innovation.

These recommendations reflect industry patterns Tan has observed working with financial institutions across Singapore transitioning from AI experimentation to production deployment. While her perspective is informed by her role overseeing enterprise data and AI platform strategy in the region, the challenges outlined — data fragmentation, governance gaps, and the tension between innovation speed and regulatory compliance — are industry-wide. Banks that treat these as strategic priorities rather than technical afterthoughts will be better positioned to realize sustainable value from AI investments in an increasingly regulated environment.