Why Banks Should Launch AI Pilots Before Perfecting Their Data Infrastructure

The Data Modernization Dilemma

The banking industry faces a fundamental tension between the promise of AI and the reality of legacy infrastructure. Bain's research identifies three critical barriers that legacy systems create for AI implementation: data silos that fragment inputs and blind AI agents, outdated pipelines that provide stale information for decision-making, and brittle integrations that slow deployment timelines.

These challenges are particularly acute in banking, where institutions often operate on decades-old core systems. A 2023 study by Deloitte found that 43% of banks still rely on mainframe systems for critical operations, with some running code written in the 1970s. The complexity of replacing these systems while maintaining regulatory compliance and operational stability has created what many executives see as an insurmountable barrier to AI adoption.

However, Bain's recommendation to proceed with pilots despite imperfect data infrastructure represents a significant departure from traditional IT modernization approaches. This strategy acknowledges that waiting for complete modernization could mean missing the competitive advantages that early AI adoption provides.

The Case for Parallel Implementation

The dual-track approach offers several compelling advantages that address the urgency many banks feel around AI competition. First, it allows institutions to begin generating value from AI investments immediately rather than waiting years for infrastructure overhauls. McKinsey research indicates that banks implementing AI pilots report average cost reductions of 15-20% in targeted processes within the first year.

• Second, early pilots create organizational learning that informs broader modernization efforts. When JPMorgan Chase began experimenting with AI for fraud detection in 2016, the bank discovered data quality issues that weren't apparent in their infrastructure assessments. These insights helped prioritize their subsequent $12 billion technology modernization program.
• Third, pilot programs build internal momentum and expertise that prove essential for larger-scale implementations. Goldman Sachs' experience with Marcus, their digital banking platform, demonstrates how starting with focused use cases can create the technical capabilities and organizational confidence needed for broader transformation.

The approach also addresses resource constraints that many banks face. Rather than dedicating all available resources to infrastructure upgrades, institutions can allocate teams to both modernization and pilot programs, potentially accelerating overall timelines.

The Hidden Risks of Imperfect Foundations

While Bain's approach offers clear benefits, it also introduces risks that demand careful consideration. The most significant concern involves the potential for AI systems built on poor data quality to make flawed decisions with serious consequences. In banking, where AI increasingly influences lending decisions, fraud detection, and customer service, errors can result in regulatory violations, financial losses, and damaged customer relationships.

The 2019 case of Apple Card's gender bias controversy illustrates how AI systems can amplify existing data problems. The algorithm, developed by Goldman Sachs, appeared to discriminate against women in credit limit decisions, leading to regulatory investigations and public relations disasters. While the exact cause remains disputed, the incident highlights how launching AI systems before addressing underlying data issues can create significant risks.

Regulatory compliance presents another challenge for the dual-track approach. Banking regulators increasingly require institutions to demonstrate robust governance frameworks for AI systems, including comprehensive data lineage, quality controls, and bias monitoring. The Federal Reserve's 2023 guidance on AI governance emphasizes that banks must ensure data quality and model validation before deploying AI systems that affect consumers.

There's also the risk of creating technical debt that becomes more expensive to resolve over time. Quick pilot implementations often involve workarounds and shortcuts that must eventually be addressed. If not properly managed, these solutions can become embedded in operations, making future modernization more complex and costly.

Learning from Industry Leaders

Examining how leading banks have navigated these challenges provides valuable insights for institutions considering the dual-track approach. DBS Bank in Singapore offers a compelling example of successful parallel implementation. The bank began AI experiments in customer service and fraud detection while simultaneously modernizing core systems. Their disciplined approach included strict data quality thresholds for pilot programs and clear criteria for scaling successful experiments.

DBS's success came from establishing what they called 'data corridors' – specific data domains where quality was sufficient for AI deployment while broader modernization continued. This approach allowed them to launch chatbots and fraud detection systems that generated immediate value while building toward more comprehensive AI capabilities.

Conversely, Wells Fargo's challenges with AI governance following their account fraud scandal demonstrate the risks of insufficient preparation. The bank's struggles with data quality and governance led to regulatory restrictions on their growth and technology implementations, highlighting how AI initiatives can amplify existing operational weaknesses.

European banks operating under GDPR provide additional lessons about implementing AI with imperfect data infrastructure. ING's approach involved creating 'AI sandboxes' with curated, high-quality data subsets that enabled experimentation while maintaining compliance with privacy regulations. This model shows how banks can pursue innovation within constrained environments.

A Framework for Balanced Implementation

For banks considering Bain's dual-track approach, success requires a structured framework that balances innovation with risk management. The first element involves establishing clear data quality thresholds for pilot programs. Rather than accepting any available data, banks should define minimum standards for accuracy, completeness, and timeliness that AI systems require for reliable operation.

• Second, institutions need robust governance frameworks that can evolve with their AI capabilities. This includes establishing AI ethics committees, implementing model validation processes, and creating feedback loops that capture lessons from pilot programs to inform broader modernization efforts.
• Third, banks should prioritize use cases that offer high value while maintaining lower risk profiles. Customer service chatbots, internal process automation, and decision support systems typically present fewer regulatory and operational risks than systems that directly impact customer financial outcomes.

Finally, successful implementation requires clear integration pathways between pilot programs and broader modernization initiatives. Banks should avoid creating isolated AI solutions that cannot eventually integrate with modernized infrastructure, as this approach ultimately undermines both innovation and modernization goals.

The Competitive Imperative

The urgency driving Bain's recommendations reflects the competitive reality facing traditional banks. Fintech companies and digital-native financial services providers are gaining market share by leveraging AI capabilities that established banks struggle to match. Research from BCG indicates that banks implementing comprehensive AI strategies achieve revenue growth 6% higher than their peers.

This competitive pressure makes the wait-and-see approach increasingly untenable. Banks that delay AI implementation until achieving perfect data infrastructure risk losing market position to more agile competitors. The dual-track approach offers a middle path that acknowledges both the need for speed and the importance of foundation building.

However, competitive pressure shouldn't override prudent risk management. The banking industry's history includes numerous examples of institutions that pursued innovation too aggressively without adequate preparation, leading to operational failures and regulatory consequences.

Implementation Recommendations

For banking leaders considering this approach, several practical steps can increase the likelihood of success:

• Begin by conducting comprehensive data quality assessments to identify domains where AI pilots can launch safely. Focus on internal processes and customer service applications that offer clear value while minimizing regulatory exposure.
• Establish dedicated teams for pilot programs that include both technology and risk management expertise. These teams should operate with clear mandates to experiment and learn while maintaining appropriate controls and governance oversight.
• Create formal feedback mechanisms that capture insights from pilot programs and incorporate them into broader modernization planning. This ensures that early experiments inform infrastructure investments rather than creating competing priorities.
• Finally, maintain transparent communication with regulators about AI initiatives and governance approaches. Proactive engagement can help identify potential compliance issues before they become problems and demonstrate commitment to responsible innovation.

The Path Forward

Bain's dual-track approach represents a pragmatic response to the competing demands of innovation and stability that banks face. While the strategy offers clear benefits in terms of speed and organizational learning, it also requires careful risk management and structured implementation to succeed.

The key insight is that perfection shouldn't be the enemy of progress, but neither should speed compromise safety. Banks that can navigate this balance successfully will likely gain sustainable competitive advantages in an increasingly AI-driven financial services landscape.

The most successful institutions will be those that view the dual-track approach not as a shortcut around proper preparation, but as a more sophisticated way of building AI capabilities that combines learning from experimentation with systematic infrastructure development. This perspective transforms the choice from whether to wait or proceed into how to proceed most effectively.

Ultimately, the banking industry's AI future belongs to institutions that can move quickly while building responsibly. Bain's framework provides a roadmap for achieving this balance, but success will depend on execution quality and sustained commitment to both innovation and risk management.

For more insights on modernizing data for AI, you can explore the details on this Bain & Company resource.