It’s fall 2024, where are we at with AI taking over financial services. Here’s a case study I’ll share with you from leading digital brokerage apps in the US. Earlier in the year they made the decision to start rolling out AI-led customer service. Even crossing that Rubicon many FIs would yet shudder to consider, promoting the AI from copilot status to directly responding to customers (no human in the loop). Contrary to what you might think, it was the engineers at first that were hesitant to dialing the models in production. The LLM performed well in testing but no model is perfect and there could always be unexpected categories of mistakes. From the leadership perspective though, managing and attempting to consistently train and deal with the way thousands of human agents could (and do) make unpredictable mistakes. As least with the AI, if it makes a mistake you can fix it one place. And so over the course of the last few months, they’ve been gradually dialing up the AI. And it turns out that delivering a model that can handle a large share of customer service needs, faster, with fewer errors good customer satisfaction is well within reach of current tech.
I should probably say that again. Assuming you’ve got enough scale, and you know what you are doing on the engineering and data science competencies… AI today is ready to take the wheel for many, if not most, routine customer service interactions in banking.
But here’s where the firm is stuck. It’s actually still too difficult to secure enough elastic or spot GPU capacity, which allows for flexible and scalable computational resources on-demand. And so the program can’t yet scale to full GA. We’re still in a world where COOs and CTOs are leaning on personal relationships with insiders at Amazon or Microsoft to book capacity. The economics and availability are particularly challenging around managing peak vs baseline volume. Inferencing costs though are steadily coming down, but it make still take a year or so before we can take for granted that you can rely on cloud providers to have elastic supply of AI capacity.
Here is the moral though for FIs that aren’t this far along yet on your AI journey. You may have some breathing room, but AI is coming for all of our businesses. For your digital ready competitors, their cost advantage in servicing customers will only accelerate. And their CX advantage too. Don’t assume the quality of AI service is lesser. Myself, I would take an impersonal but instant, knowledgeable and effective resolution to my needs by a robot… over hours of hold music to reach a human agent who may often have no idea how to help me. (ask me about my recent 8 week horror story of just trying to purchase ordinary services from an US health insurance carrier…)
Or if you ever need help on your own AI Transformation, I’m always happy to provide fellow product leaders some free advisory time or recommend some solid partners on the acceleration/implementation side.