1006. Insights: AI in banking: What it really means for the customer
🎯 Summary
Podcast Summary: 1006. Insights: AI in banking: What it really means for the customer
This 48-minute episode of Fintech Insider Insights cuts through the generative AI hype to examine the tangible, customer-facing applications of Artificial Intelligence currently being deployed, or held back, within the financial services industry. The core narrative explores the significant disconnect between the perceived potential of AI (especially GenAI) and the slow, cautious rollout of genuinely transformative customer experiences in banking.
1. Focus Area
The discussion centers on the practical application and customer perception of AI in banking, contrasting existing, often invisible, backend automation with the slow emergence of front-end, agentic customer experiences. Key themes include customer trust, the role of UX research in adoption, the technical realities of deployment (fine-tuning vs. foundational models), and the strategic importance of internal organizational culture in driving successful AI integration.
2. Key Technical Insights
- Model Mixing and Efficiency: Successful deployment involves a multi-modal approach, strategically mixing foundational models with smaller, fine-tuned models for specific tasks (e.g., date extraction). This approach saves costs, reduces environmental impact (fewer tokens/watts), and optimizes performance based on task complexity.
- Backend Automation Revolutionizing Advice: AI is significantly impacting the quality and speed of financial advice by automating administrative tasks for advisors, such as real-time transcription, automated note-taking, CRM updates, and rapid document generation, drastically reducing fulfillment time from weeks to hours/days.
- The MVP Trap: While building Minimum Viable Products (MVPs) with new AI tools is quick (5-10% of the challenge), the rigor required to move these solutions into secure, production-ready customer environments remains the primary technical and operational hurdle.
3. Business/Investment Angle
- Advice Accessibility as a Game Changer: Automation driven by AI is enabling firms to offer sophisticated advice services at significantly lower price points (in one example, reduced to one-eighth of the original cost), opening up previously underserved customer segments to professional advice.
- Trust as the Primary Barrier to Agentic AI: The biggest hurdle for customer-facing AI is not technological capability but earning public trust. Financial institutions must adopt a step-by-step approach, ensuring humans remain in the loop initially, to onboard customers to semi-autonomous behaviors.
- Empowerment Over Execution: As AI makes execution cheap and easy, organizational success hinges on structures that empower internal teams (close to tech, data, and business problems) to move validated ideas from prototype to production smoothly, avoiding “shiny object syndrome.”
4. Notable Companies/People
- Starling Bank (Oscar Barlow): Highlighted for its internal culture of empowering tech teams and its practical application of AI via the Spend Intelligence Tool, which incorporates generative elements while focusing on building customer trust step-by-step.
- Multiply AI (Vivek Medlani): Focuses on helping firms launch scaled advice propositions, leveraging new AI developments to enhance advisor productivity and accessibility.
- Bunk (Netherlands): Cited as a positive example of a chatbot that feels human-like and maintains customer control through reactive/revolving prompts based on recent activity and actively soliciting customer feedback to shape the AI.
- 11FS (Rosie Lee): Provided the crucial UX perspective, emphasizing that customers prefer AI tools for self-service but resist AI making decisions on their behalf, demanding control over the process.
5. Future Implications
The conversation suggests the industry is moving toward a hybrid model where AI handles high-volume, low-risk tasks in the background (KYC, process automation), while front-end customer experiences will evolve slowly, prioritizing control and transparency. The future success of advice propositions lies in leveraging AI to drastically reduce costs, democratizing access to personalized financial guidance. The persistence of off-channel AI use (e.g., customers using ChatGPT for financial queries) indicates a strong latent demand that regulated institutions must eventually meet responsibly.
6. Target Audience
This episode is highly valuable for Fintech Professionals, Banking Executives, Product Managers, AI/ML Strategy Leads, and UX/Customer Experience Strategists operating within regulated financial services who need to move beyond the hype and understand the practical, trust-based roadmap for AI adoption.
🏢 Companies Mentioned
💬 Key Insights
"I think when it comes to banking and using your banking app, I think that customers aren't quite there. They see banks need, banks, financial institutions need to build trust with their customers. And I don't think that they quite have that yet."
"We're now seeing AI therapists, AI coaches. These are such fun. There's no number crunching in there, right? These are fundamentally human services that are out there that are beginning to find some type of form factor in the digital world with with elements because there is an element of this technology which can understand language."
"there is that familiarity, there is that tone of voice aspect to it that I think will really demarcate the kind of average products out there from the ones that really, really lead the way when it comes to those sorts of these sorts of services."
"Voice makes sense because it's unstructured. I don't necessarily know actually what I'm talking about. I just have a few ideas and I want to bounce some ideas off of someone. So that works. And to what extent then will that then work when I'm speaking to an AI no matter how eloquent that AI actually is?"
"The thing that frustrates me is that I can't type fast enough for my thoughts. . . actually talking just makes the thoughts come out easier. And if we can build technical systems that can respond to meaning and intent, then voice, I think, is a natural modality for this."
"When we tested it with just the score itself, it didn't test very well. Customers didn't really understand what the score was showing them and why they were being showed this score. But when we broke the score down into components and we gave the most score out of 10 for each measure and they then tested better because customers want to understand not necessarily exactly how the AI is working, but what it means for them."