Winning AI Playbooks from Chase, BofA, NatWest and More
As banking institutions everywhere work to define their AI strategies, themes and best practices are emerging from the technology's pioneers. A new report from Euromoney details multiple case studies and concludes that (as in many things) success often depends on how you define it – and definitions can vary, a lot.
By David Evans, Chief Content Officer at The Financial Brand
The report: The AI in Banking Best Practices Playbook
Source: Euromoney
Why we picked this report: Best practices in AI implementation in banking are coming into focus. Lessons learned from early adopters and industry leaders can assist fast followers and laggards keep up with fewer stumbles.
Executive Summary
Financial institutions are discovering that AI can significantly enhance both employee efficiency and customer experience. Banks with strategic AI implementation are already seeing productivity gains of 15-30% in certain departments, while customer-facing AI applications show customer satisfaction improvements of up to 150%.
However, AI adoption demands a careful balance between innovation and risk management, requiring flexible technological infrastructure and organizational readiness. Leading banks are centralizing AI strategy while devolving execution to business units, investing in proprietary AI platforms that allow model-agnostic approaches, and prioritizing employee upskilling across all levels. Those who can successfully navigate AI’s complex implementation challenges while maintaining trust will gain significant competitive advantages.
Key Takeaways:
• Flexible infrastructure is critical: Banks must build AI implementations that remain agnostic to specific models and vendors, allowing quick adaptation to technological developments.
• Organizational balance matters: The most successful banks centralize AI strategy, governance, and cost management while devolving use-case execution to business units.
• Low-risk cases yield high returns: Internal applications like developer tools and employee assistants demonstrate immediate ROI with 15-30% efficiency gains.
• Customer-facing applications require caution: While more complex, customer-facing AI implementations show promise with significant satisfaction improvements.
• Human-AI collaboration is paramount: Even as banks experiment with agentic systems, maintaining human oversight builds trust and ensures regulatory compliance.
Key Data Points:
• NatWest recorded a 150% increase in customer satisfaction among those using its Gen AI-powered Cora+ chatbot
• JPMorgan Chase has deployed AI tools to approximately 200,000 employees (two-thirds of its workforce) via its LLM Suite platform
• UBS recorded one million AI prompts from employees in January 2025 alone, compared to 1.75 million for the entire year of 2024
What we liked about this report: Case studies and lots of them. Some of the anecdotes are familiar from press reports and elsewhere but in aggregate the example create a useful mosaic.
What we didn’t: Some readers may find the report’s advice too fungible. The report’s focus on "balance" in AI implementation can be translated as, "Best practices in AI vary from institution to institution depending on its goals, markets and other factors".
Today’s AI revolution differs from previous technological shifts in both the pace of advancement and the breadth of potential applications. Unlike incremental improvements in traditional banking technology, AI’s capabilities are expanding exponentially, with each new model generation unlocking previously unimaginable functionalities.
The rapid development of large language models (LLMs) since late 2022 has created both opportunity and urgency for banking executives. Staying competitive now demands strategic AI adoption, yet implementation presents complex challenges around model selection, data security, regulatory compliance, and organizational readiness.
The Organizational Foundation for AI Success
Successful AI implementation begins with organizational preparation. Leading banks have discovered that neither completely centralized nor fully decentralized approaches work optimally. Instead, a hybrid model proves most effective: centralizing AI strategy, governance, and cost management while devolving use-case development and execution to business units.
"I’m a big believer that the CEOs of the businesses are best placed to understand where AI can help them achieve their strategic objectives," notes Teresa Heitsenrether, Chief Data & Analytics officer at JPMorgan Chase. "People talk about AI strategies. That’s a misnomer. There are business strategies, and there are places where AI can enable the strategy."
This balanced approach ensures AI initiatives align with business goals while maintaining enterprise-wide standards and preventing risk proliferation. Central AI teams should function as strategic shepherds rather than bottlenecks to implementation, guiding business units toward use cases with scalable potential across the organization.
Morgan Stanley exemplifies this approach, having built its central AI function from a team working on wealth management applications that subsequently expanded across divisions. The bank’s AI assistant for wealth management advisors became the foundation for similar assistants in other business lines, demonstrating how targeted implementation can scale across an enterprise.
Building Technical Flexibility into AI Infrastructure
Banks face a dual challenge: capturing AI’s benefits today while preparing for tomorrow’s technological advances. This requires building flexible infrastructure that can adapt to rapidly evolving capabilities.
"Everything you build in AI should allow you to be agnostic as to which LLM you use, and as to the vendor," advises Christoph Rabenseifner, Chief Strategy Officer for Technology, Data and Innovation at Deutsche Bank. "It should allow you to quickly adapt to new technological developments."
Forward-thinking institutions have developed proprietary platforms that act as abstraction layers between underlying models and customer-facing applications. JPMorgan Chase’s LLM Suite, Goldman Sachs’ developer platform, and Santander’s Luiza are representative examples, allowing these banks to swap underlying models without disrupting user experiences. This approach offers critical advantages:
1. Cost flexibility: Banks can negotiate with multiple vendors rather than being locked into single-provider pricing
2. Technological agility: Institutions can quickly adopt the latest models without rebuilding applications
3. Risk management: Security protocols can be consistently applied regardless of the underlying model
NatWest’s experience illustrates the necessity of this flexible approach. Before ChatGPT’s release, the bank was developing a call-summarization system using its own natural language processing technology. After ChatGPT emerged, NatWest discontinued the project, recognizing that third-party LLMs could accomplish the same tasks more effectively. Subsequently, the bank started developing a complaints-automation procedure based on OpenAI’s models, upgrading through four model generations during the nine-month development period.
The People Element: Upskilling Across the Organization
AI’s successful deployment depends as much on human capabilities as technological infrastructure. Banks must develop talent strategies encompassing both specialized AI researchers and the broader workforce that will use AI tools daily.
"Many of the best ideas have not been found yet, because those ideas will come from frontline staff when they truly understand what this technology can do," explains Chris Purves, Co-head of Emerging Technology at UBS. "That’s why you need education."
UBS has implemented a comprehensive approach to AI training, targeting all organizational levels. The bank launched a year-long data science program attracting 1,700 enrollments, established reverse-mentoring where junior employees train senior executives on AI developments, and created prompting competitions to build institutional knowledge. This investment is paying dividends, with UBS recording one million AI prompts from employees in January 2025 alone, compared to 1.75 million for all of 2024.
DBS Bank in Singapore took a creative approach to talent development by establishing what it calls a "Data Chapter" – essentially a modern guild for data professionals. This community of practice brings together 700 data specialists from across the bank, providing training, networking, and career development opportunities. The structure helps retain talent by showcasing diverse use cases across the organization while building institutional data capabilities.
Low-Risk, High-Value Implementation Targets
While AI offers transformative potential across banking operations, prudent implementation prioritizes lower-risk applications with demonstrable returns. Internal tools for employees, particularly software development aids and knowledge management systems, represent ideal starting points.
Beyond development environments, employee-facing AI assistants deliver substantial productivity improvements by simplifying access to institutional knowledge. At Discover Financial Services, an AI tool reduced policy-and-procedure search time by 70% in call centers. These efficiency gains directly translate to improved customer experiences, with faster response times and more consistent service delivery.
BBVA demonstrates how even seemingly minor AI applications can deliver substantial value. When the bank’s mortgage team needed to update collateral values in its database, they leveraged AI tools to build a system verifying and updating values based on public housing market data in just days – a task that previously would have required a substantial IT project with specialized technical teams.
Navigating Customer-Facing AI Applications
Customer-facing AI implementations offer greater potential rewards but involve correspondingly higher risks. Leading banks are proceeding cautiously, building sophisticated safeguards before deploying AI in direct customer interactions.
NatWest has enhanced its existing Cora chatbot with an opt-in generative AI feature called Cora+. When customers elect to use this capability, the system asks permission to retrieve information and present it in conversational form rather than simply directing customers to relevant website sections. This approach has yielded a 150% increase in customer satisfaction metrics and 50% fewer handoffs to human agents.
Bank of America has taken a different approach with its Erica virtual assistant. Rather than deploying generative AI to produce responses (which could risk hallucination), the bank uses AI to better understand customer queries while maintaining controlled outputs. Through continual refinement including more than 55,000 updates since launch, Erica’s accuracy has improved from 65% to 95%.
Both approaches demonstrate the essential balance between innovation and risk management in customer-facing AI applications. As these systems mature, they will increasingly handle more complex customer needs with less human intervention, though always with appropriate safeguards.
The Future: Towards Agentic Banking
As AI models gain greater reasoning capabilities, banks are cautiously exploring "agentic AI" systems that can orchestrate complex tasks across multiple applications. While definitions vary, these systems typically involve AI agents that can understand user needs in natural language, formulate plans to address those needs, and execute those plans by interacting with relevant systems.
Commonwealth Bank of Australia exemplifies this approach with its card dispute resolution system. Previously, customers navigated multiple screens and dropdown menus to report disputed transactions. With the new AI-powered system, customers describe their issues conversationally, with an AI agent assessing problems in real-time, requesting additional information when needed, and orchestrating resolution. For simple disputes, this could reduce resolution times from days to minutes.
Capital One has deployed a similar system called Chat Concierge for auto finance, which allows customers on participating dealer websites to compare vehicles, explore financing options, get trade-in estimates, and schedule test drives through natural language interaction. The system demonstrates how agentic frameworks can transform customer experiences while operating within low-risk domains.
These early implementations suggest a future where banking becomes more conversational and intuitive, moving beyond the self-service model that has dominated digital banking. Instead of navigating complex app interfaces, customers will simply state what they need and AI agents will navigate bank systems on their behalf.
Building Trust: The Critical Success Factor
For all AI’s technical potential, successful adoption ultimately depends on trust – from customers, employees, regulators, and society at large. Banks must develop AI governance frameworks that ensure responsible use while maintaining the agility to innovate.
"Credit is the highest risk part on the risk slope in using AI," notes Prem Natarajan, Chief Science Officer at Capital One. "If the transformation is enduring, what is the rush to get it done? Let’s get it done right. Nowhere is that more important than in credit, because it fundamentally affects people’s everyday lives, in a deep way."
This emphasis on responsibility extends beyond high-risk applications like credit decisioning to all AI implementations. Effective governance includes developing internal standards for AI deployment, creating diverse leadership councils to assess use cases, implementing technical safeguards to prevent bias and hallucination, and maintaining appropriate human oversight.
"We always make sure that gen AI tools are suggestions for our people," explains Marco Argenti, Chief Information Officer at Goldman Sachs. "I don’t think we are at the stage now where you can take output from AI and consider that a finished product… It’s very important that we keep that element of human supervision."
Editor’s note: This article was prepared with AI language software and edited for clarity and accuracy by The Financial Brand editorial team.
