The data platform bet: Why financial AI initiatives stall and how the winners scale

The adoption of AI is accelerating rapidly across financial services companies. However, a significant disconnect exists between ambition and operational reality. Many organizations invest heavily in advanced models only to find their projects stuck in endless testing phases. The root cause is rarely the model itself. The failure stems from the underlying data foundation.

Organizations often manage data in siloed systems, outdated architectures, and manual spreadsheets. AI requires speed, context, and flawless governance to function effectively. Without a unified data platform, organizations cannot deliver the real-time insights necessary to operationalize AI at scale.

These were among the topics I recently discussed with Dr. Efi Pylarinou, a top global fintech and tech influencer, and Mike Sisk, contributing editor at American Banker. We explored why data readiness determines AI success and how leaders can build a resilient foundation.

The widening gap in AI readiness

Financial services companies are not new to AI, but the demands of generative and agentic AI expose deep flaws in traditional infrastructure. The companies leading the market today began fixing their data architecture years ago. Organizations relying on batch processing and fragmented data stores are falling behind.

"More than 40% of financial services are still managing their data in spreadsheets," Pylarinou explains. "More than 50% have data that are locked in systems that generate that data."

When data remains trapped in silos, AI models lack the context required to make accurate decisions. This forces teams to spend excessive time cleaning and routing data manually. The business impact is severe. Slow data access prevents real-time fraud detection, delays customer service responses, and introduces massive compliance risks.

Why traditional data lakes fall short

Many organizations assume their existing data lakes or workflow automation tools are sufficient for AI. These systems serve a purpose for analytics and reporting, but they were not built for the instantaneous demands of modern AI agents. Data lakes hold historical information, while AI requires immediate context.

Pylarinou notes that these systems fail to solve the core problem of delivering the right data to the right model in a compliant way. To support advanced AI, a unified data platform must deliver the following capabilities:

• Fast access to data in milliseconds rather than seconds

• Contextual retrieval that brings relevant background to every query

• Cross-silo capability to span across different legacy schemas

• Built-in governance to maintain an audit trail and ensure correct access controls

When a platform unifies insights from onboarding, transactions, and behavioral signals, it enables the organization to respond to market changes instantly. This shift moves the business from reactive reporting to proactive, machine-speed decision-making.

"Data is the backbone of any AI success," Sisk adds. "Without a solid infrastructure, even the best models can’t deliver results."

Securing the perimeter at machine speed

The push for AI adoption also introduces severe security vulnerabilities. Autonomous agents can access vast amounts of information in fractions of a second. If data architectures lack proper access controls, a single breach can expose millions of records before human analysts even review the daily logs.

Pylarinou highlights a recent incident at a major consulting firm where an autonomous agent accessed thousands of confidential files in just two hours during a stress test.

"Preparing your data architecture is not only about serving your AI agents, it's about defending yourself against AI even if you haven't moved into transforming your internal processes," Pylarinou says.

For financial organizations, this means security and observability must converge. A unified platform allows security teams to monitor data access continuously. This comprehensive visibility is required to detect anomalous behavior early and protect the institution from catastrophic data loss.

Navigating complexity in legacy environments

Legacy systems carry decades of distinct data structures. Ripping and replacing these core systems is rarely feasible for large organizations. Instead, organizations must introduce an augmented layer that unifies data across disparate sources.

A unified schema must understand the importance of context. Metadata not only helps users understand what the data is for, but it also provides the context needed to drive an agent's or large language model’s (LLM) decision.

To achieve this unification without disrupting operations, leaders should focus on:

1. Augmenting existing systems rather than attempting full replacements

1. Creating a common schema that is readable by both humans and large language models

1. Prioritizing search engines capable of incredibly fast data recall

By applying this approach, financial services companies can extract transaction records from mainframes and enrich them with context. This enriched data powers immediate insights for fraud prevention and customer behavior analysis.

Governance as a sustainable advantage

As AI models become more autonomous, traditional risk management frameworks become obsolete. Organizations can’t rely on rule-based monitoring for non-deterministic models. Trust must be engineered directly into the data platform.

"The biggest gap in the market is clearly governance," Pylarinou states.

This point underscores the need for organizations to implement logging at every step. This makes every AI action auditable and explainable. When a company can prove exactly how an AI model reached a decision, it gains the confidence of regulators and customers. Governance transitions from a compliance burden into a competitive advantage. Building the foundation for future scale.

The companies winning with AI are not simply adopting better models. They are making better long-term platform and data architecture decisions. A unified, flexible, and real-time data platform is the only way to escape pilot purgatory.

By prioritizing data unification, open standards, and stringent governance, financial services companies can operationalize AI securely. The focus must remain on solving the data problem first.

_The release and timing of any features or functionality described in this post remain at Elastic's sole discretion. Any features or functionality not currently available may not be delivered on time or at all._