5 minutes
Plus, four challenges and a word of caution to consider when implementing member-facing AI
Artificial intelligence is far from a novelty these days, even in a risk-averse industry like financial services. Still, the best applications of AI—spam detection, fraud prevention, risk assessments—generally operate behind the scenes. They aren’t freely accessible or interactive.
That’s not the case for the world’s most virally popular AI system: ChatGPT.
Chatbots aren’t a new phenomenon either, but ChatGPT has made a big leap forward in terms of natural language processing, natural language understanding and generative text. The depth and breadth of answers, the fluidity of conversations, as well as the types of interactions are all remarkable.
This opens the door to a wealth of opportunities for credit unions, but there are critical challenges that can limit its deployability in the near term.
The Challenges of Using ChatGPT in Banking
Chatbots should streamline the member experience: The member has a question, and the chatbot answers it (or directs them to a solution, at least), saving everyone time. However, it’s difficult for credit unions to deploy advanced chatbots like ChatGPT in member-facing settings, because the purpose goes beyond talking to the member and understanding what they want. It’ also needs to perform an action to help solve the problem, which introduces several potential hurdles:
- Back-end infrastructure
- Regulatory compliance
- Liability ambiguity
- Human connection
First, AI-based chatbots are only as useful as the back-end infrastructure supporting them. Consequently, credit unions need to leverage API integrations to enable their systems to exchange data and facilitate smooth interactions. That often requires significant investments in IT architecture and third-party partnerships (i.e., it’s easier said than done).
Second, to no one’s surprise, compliance and regulatory constraints are a big deal in banking. ChatGPT isn’t perfect—it can be, and occasionally is, wrong. Generative models may rely on outdated information or misconstrue questions that are true under some circumstances but not others. Yet, regardless of accuracy, ChatGPT answers with extreme, believable confidence. And that’s particularly troublesome with subjective topics where it’s hard to pinpoint truth or fiction, which can certainly be the case in financial services.
Third, an overlooked benefit of human advice is knowing who’s ultimately responsible for that advice. There’s distinct, unequivocal accountability. Naturally, the pressure of being liable for incorrect advice leads to rigid processes with clear hierarchies and rules; people don’t want to be on the hook for mistakes, so they act with caution. That’s not necessarily the case with AI.
Finally, people can not only provide correct solutions and represent the brand but also empathize with the member. As we all know, money is a sensitive subject, and managing finances frequently requires emotional support, something humans alone can provide.
How ChatGPT Can Impact Banking Today
AI systems are currently best suited for high-volume, repetitive tasks that lack variability. With that in mind, the most viable use case for ChatGPT in banking is content creation, specifically for areas that aren’t subject to significant scrutiny.
For instance, an AI could generate templates for web pages or descriptions of products and services. Credit unions can maximize the effectiveness of this exercise by incorporating a hybrid model: ChatGPT handles the mundane heavy lifting while a human specialist proofreads, edits and polishes the text. This approach could save hours of time for the individual and increase efficiency with each iteration.
Where ChatGPT Can Impact Banking Tomorrow
We can be confident that ChatGPT will make predictable, linear improvements every six to 12 months or so—models will be updated, refined and, ideally, more accurate.
However, at its current trajectory, we’re accelerating toward a world of AI-to-AI interactions and services. In other words, people could have their own AI acting as an agent on their behalf. Your own personal/consumer AI could converse with a financial institution’s AI to proceed through the preliminary stages of an arrangement or negotiation, such as a mortgage loan, before the human counterparts take the reins to finalize or endorse conclusions reached by their AI agents.
That said, although AI systems will undoubtedly reshape workflows and roles, we shouldn’t expect mass replacements of human-filled positions. That certainly holds true in a highly regulated and service-oriented industry like financial services.
Words of Caution for Banks Investing in ChatGPT
AI is an investment. To maximize the return on investment, there are a few principles to keep in mind.
First, avoid dangerous KPIs (key performance indicators). The primary purpose of an AI like ChatGPT shouldn’t be to minimize the workload of employees—it’s to enhance the member experience. Therefore, chatbot systems that are rated prominently on a percentage of deflections (i.e., chats that aren’t escalated to a human) are dangerous. By prioritizing containment metrics, institutions could inadvertently box members in and reduce necessary human-to-human interactions. And since negative experiences are weighted much more by members than positive ones, this can cause fissures in member relationships.
Second, in conjunction with the first principle, understand that no AI system is comprehensive enough to resolve every inquiry. We’ve all raised our voices (or vigorously typed) to chatbots, exclaiming that we want to speak with a representative because our question isn’t being answered. Therefore, it’s imperative for credit unions to establish painless transitions to a human handler. Hurdles and delays in this outcome can lead to frustration and reputational damage.
Finally, iterate quickly. AI systems like ChatGPT are evolving rapidly, to a point where long-term projects face the constant threat of interruption from updates and advancements. Credit unions that operate in smaller increments that allow for weekly or even daily changes will experience better overall results, especially in terms of getting the most bang for their bucks.
Slaven Bilac is CEO of Agent IQ, a provider of digital customer engagement solutions specializing in making financial services more personal again. Before launching Agent IQ, Bilac worked at Google for more than a decade in both its Japanese and American offices. Specializing in software engineering, Bilac improved many of the search engine’s technical aspects, such as spell correction and query suggestion, as well as assisting to establish machine intelligence units in Google Cloud.
