We recently participated in a webinar titled "Transforming CX in Financial Services with Generative AI" in partnership with the Financial Brand.
In addition to boost.ai’s VP of North America, Bill Schwaab, and Chief Customer Officer, Henry Vaage Iverson, we were joined by Ami Iceman Haueter and Ben Maxim, the chief research and digital experience officer and chief digital strategy and innovation officer for Michigan State University Federal Credit Union (MSUFCU).
Throughout this webinar, we discussed the potential that generative AI holds for institutions in the financial services industry and how those institutions can responsibly leverage large language models (LLMs). Here are some key takeaways from this eye-opening conversation:
An urgent need for fast, effective customer service
Since its first arrival on the scene, ChatGPT — and generative AI, in general — has been a popular topic of discussion in the financial services industry. With its ability to rapidly generate text using natural human language, banks, credit unions and other institutions have recognized potential use cases for this technology in customer service.
And it couldn’t come at a better time: Contact center wait times and engagement costs are steadily increasing, despite the introduction of automation. Employee attrition rates also remain high, placing strain on available resources. To cap it all off, customer expectations around self-service capabilities are at an all-time high.
Virtual agents are the obvious solution to these issues, able to provide high-quality, round-the-clock self-service at an affordable rate. But there remains confusion about what virtual agents can and can’t do, and what the risks are to using them.
Harnessing the power of generative AI
Immediately, two potential use cases for virtual agents come to mind: engaging in direct dialogue with customers and empowering human agents to be more efficient. But before financial institutions can reap the rewards of these applications, they first need to understand the implications.
Direct dialogue with customers
Virtual agents are the perfect complement to human agents, able to field lower level service requests and customer inquiries. But in order to successfully supplement their existing contact center staff with generative AI, financial institutions must first implement analytics to monitor conversations and improve upon their AI model and define service-level agreements (SLAs).
Although generative AI’s capabilities are impressive, they aren’t limitless. There may come a time when a virtual agent isn’t able to understand or answer a customer’s question. In times like these, it’s essential to have a clearly defined escalation path for handing off inquiries to live agents.
One way to streamline this process — and to avoid having a customer repeat their question — is to deliver an AI-generated summary of the inquiry and issue at hand to the human agent. This provides them with full context, so they can seamlessly pick up the conversation where the virtual agent left off.
Honesty is always the best policy when working with generative AI. More specifically, financial institutions should be candid about when customers are interacting with a virtual agent as opposed to a human agent and should tie all generative responses back to existing information within their knowledge base.
Empowering human agents
Generative AI is just as valuable on the back end as it is on the front end. Financial institutions can leverage this technology to rapidly build out knowledge bases, create onboarding materials for new agents and give live agents the tools they need to reduce handling times.
Let’s focus on that last piece. Just as customers can utilize virtual agents to answer common questions and resolve low-level issues, human agents can use them to quickly track down the information they need and provide customers with a rapid response. In situations where virtual agents need to escalate issues or questions, they can deliver detailed summaries to live agents to provide them with a more robust understanding of the matter at hand.
Say “hello” to Fran and Gene
For an example of generative AI in action, look no further than Michigan State University Federal Credit Union (MSUFCU). In 2019, the credit union launched “Fran,” a customer service virtual agent, with the help of another chatbot vendor. Although that particular solution wasn’t the right fit, MSUFCU learned some valuable lessons from the experience, which they brought to their partnership with boost.ai. With help from boost.ai, MSUFCU launched “Gene,” an internal virtual agent for employee service. Since its debut, Gene has been an asset to the MSUFCU team and has become an integral part of company culture.
MSUFCU’s success with Gene inspired them to relaunch Fran in 2021, again using the boost.ai platform. Fran’s reintroduction has enabled the credit union’s team of live agents to hand off simpler engagements and dedicate their time and focus to more value-adding areas of service. According to Ami Iceman Haueter, chief research and digital experience officer for MSUFCU, Fran is able to handle up to 23,000 conversations in a single month — 74% of the company’s total chat traffic.
MSUFCU’s journey with generative AI is just beginning; the company is in the early stages of exploring LLM models within the boost platform. But the biggest takeaway from their experience so far? That it’s essential to have a knowledgeable partner throughout the implementation and deployment processes. Haueter credits boost.ai for providing an unparalleled level of expertise and acting as a “personal tour guide” for MSUFCU as the credit union works to meet its AI pillars — security, accuracy, consistency, transparency and usability.
Addressing common generative AI concerns
During the webinar, we polled attendees to see what some of their chief concerns were related to generative AI and LLMs. The top responses were accuracy of information, data privacy, cost, effort and implementation. Our panel addressed each concern:
Accuracy
LLMs are sophisticated enough that they can consistently deliver accurate information, provided you build in the appropriate control mechanisms. The boost platform, for example, enables financial institutions to develop scripts for virtual agents to use based on existing content, such as pages on their website or information from their knowledge base.
However, we know that customers can and do go off-script. In times like these, the best short-term solution for ensuring accuracy is to escalate to a live agent. And by monitoring virtual agent interactions and collecting data from conversations that go off-script, financial institutions can actually retrain LLMs to better anticipate and respond to unexpected inquiries.
Data privacy
Regulatory compliance is a major concern for the financial services industry, which is why it’s important to work with a knowledgeable partner that can answer any and all compliance-related questions. MSUFCU found that partner in boost.ai. It also doesn’t hurt that boost.ai was founded in Norway, in one of the most stringent data privacy regulatory environments, which means that privacy protections are built directly into the platform.
Henry Vaage Iversen, chief commercial officer for boost.ai, provided additional context, explaining that the company never relies on customer data to train its LLMs, instead working with OpenAI to source data, and maintains a zero data retention policy.
Cost
Cost is a common obstacle to financial institutions considering generative AI. But if MSUFCU’s experience is anything to go by, the efficiency gains AI can offer are well worth the investment. According to Haueter, Fran — MSUFCU’s customer service virtual agent — can do the work of 60 employees, enabling the credit union to easily scale according to demand. Haueter also emphasized that MSUFCU hasn’t downsized its chat team, but has instead reallocated its experienced employees to handle more complex issues.
Effort
While building a custom application using the OpenAI API might appeal to some institutions, it requires significant effort and expertise. Financial institutions looking to avoid this degree of complexity should consider more turnkey solutions such as boost.ai.
Implementation
The key to a successful generative AI implementation? Taking a phased approach. By implementing generative AI in contained phases, rather than trying to boil the ocean, clients are able to drive immediate value while gradually progressing to more advanced use cases.
Closing thoughts
It comes as little surprise that, when polled, webinar attendees said improving customer services is their top priority with generative AI. MSUFCU is proof positive that it’s possible. Not only that, but the credit union has discovered that optimizing the member experience through the use of generative AI comes with added benefits, including lower operating costs, better resource utilization and greater scalability. With the help of boost.ai, MSUFCU will continue to find new and innovative ways to leverage AI to maintain a competitive edge in the market and create memorable experiences for their customers.
