Finn Chat– Addressing a Deficiency in Financial Customer Service By Exploring the Capabilities of an AI-Powered FinTech Chatbot
Background
In recent years, the relevance of AI powered robo-investors has seen a fairly significant uptick in response to general demand for done-for-you investing portfolios. And in November of 2022, the release of ChatGPT sparked a wave of interest in AI powered chat tools.
ChatGPT responded to user prompts with seamless, uncannily human-like responses. Utilized properly, ChatGPT had the potential to serve as a one-stop-shop research tool that could enable users to tap into the vast database of the internet in the blink of an eye– with several caveats, of course.
ChatGPT was by no means perfect, but it did lend itself to providing novel solutions to complex, unsolved problems.
The Challenge
Customer service in the financial industry has always presented difficult problems that couldn’t be solved with conventional solutions.
Customers are often handed off to different contacts and representatives who each need to be filled in on all the details of a specific case again, and again, and again. Much-needed financial advisors and wealth managers can be unreliable and difficult to reach even at the best of times, as they are often juggling multiple responsibilities, accounts, and clients across various verticals.
This lack of effective communication and timely follow-up from financial institutions makes for an extremely frustrating and impersonal experience that in turn results in lower customer satisfaction, customer retention, and overall reputation.
But, could a sophisticated AI chatbot like ChatGPT offer relevant, meaningful advice while playing the role of a financial professional?
As our thought experiments with this idea grew in complexity, we realized that a 1-to-1 chat scenario probably wouldn’t be all that compelling. After all, 1-to-1 conversations tend to be goal or transaction-oriented, which wouldn’t leverage the more nuanced potential of something like ChatGPT.
How could we make the current state of AI actionable for a financial institution? AI chatbots offered a scalable alternative to traditional banking procedures. It could simultaneously assist multiple users, provide a secure messaging experience, better visualize financial scenarios, and lay the groundwork before engaging with human financial professionals.
It wouldn’t work very well in a 1-on-1 scenario, but what about in a group setting? With that, we decided to make a slight pivot.
The Solution
Our experience in the financial and machine learning sectors, told us that an efficacious robo-advisor was a market-viable solution.
Our goal was to create a sort of “Google at the Dinner Table” effect– a system that provides accurate information in the context of a conversation, making the conversation more productive. But let’s be clear about one thing: we’re not saying we want Google or ChatGPT at a dinner party. That sounds annoying– exhausting, even.
This specific use-case for ChatGPT would integrate into a bank’s existing systems and provide the user accurate information and advice in the context of a nuanced conversation by taking into account the user's financial information. Of course, this platform would also have to ship with water-tight compliance standards to protect privacy and security.
So we got to work developing a prototype and non-clickable demos.
The Process
An Accessible Design
What if we could set up situations where business partners or a married couple could discuss a line of equity or a home purchase while an AI chimed in from the sidelines? The advice wouldn’t be definitive, but it could certainly be deeply useful.
And since the conversations would persist, new participants who join part way through the conversation (like a human advisor) could also provide helpful, contextual advice that takes the user’s financial situation into account.
Essentially, a FinTech chat platform would most likely find its footing in a collaborative group chat setting. This setting would allow the AI platform, which we’ve named Finn, to create the most productive and effective experience for both users and financial professionals.
With all that in mind, our teams made the necessary adjustments to accommodate for the new direction we were heading with this platform. We created new flows for a husband-and-wife and CEO-and-CFO conversation as well as designs for a messaging interface, SSO, bank account linking, AI decision trees, persistent conversations, and profiles.
The Result
An Accessible Design
We wanted the design to be easy to follow and intuitive. The notion of an AI platform already carries with it an inherent bias towards complexity and esotericism. With how far AI has come and the advent of AI as a Service platforms, we know that this notion couldn’t be further from the truth.
However, to encourage third-party buy-in, we would have to reduce friction and immediately address the most top-of-mind objections on first contact. To that end, UI and UX were sculpted around simplicity and accessibility.
Features - Persistent Environments
The persistent chat environment belongs to a suite of functions integral to the Finn Chat experience. Stakeholders and financial advisors may join an existing chat room and have access to the prior conversation at any time. This would eliminate the need for a customer to repeat the details of their circumstances, inform new entrants, and ultimately provide a much more meaningful and much less frustrating experience.
Features - Dedicated Chat Rooms
Segmented Chat Rooms would enable organizations to provide more specific, highly relevant information to users. Very clear segmentation allows users to access the information they need as fast as possible while also making it much easier for organizations to manage and allocate personnel on a need-basis.
Additionally, specific segments would also enable much faster data training and AI modeling for the Finn Chat AI. Allowing specific instances of Finn to train on only the data it needs would allow the platform to stay up-to-date on the latest developments across various verticals with minimum delay.
Features - Core Messaging Interface
Finn Chat would of course ship with core instant-messaging functionalities such as private channels, direct messaging, and group messaging. Users can also exercise management privileges to add or remove users from a group chat and send invites to uninitiated members.
Features - Single Sign On
SSO would ship with a simple onboarding function and a sign-in/sign-up screen. To adhere to cybersecurity best practices, this feature would also support 2 Factor Authentication.
Features - Profile and Bank Info
A profile feature would allow users to view their user information and make changes as necessary. Data fields include Name, phone number, email, security, and bank account management.
As far as bank account management goes, we would likely utilize the Plaid API to facilitate linking to financial institutions and any monitoring functionality. Users would be able to link bank accounts (checking, saving, 401K, etc) to Finn Chat so that Finn can view relevant data. The depth of monitoring would not go deeper than simple information read, as Finn is not designed to make payments or facilitate transactions (though these functions can be added while adhering to compliance).
Demonstration - Husband and Wife Flows
We also wanted to showcase how Finn could integrate itself into the conversation when prompted. To that end, we developed two conversation flows. This flow demonstrates a conversation between husband and wife.
Finn is non-intrusive and only enters the conversation when prompted. He behaves much like Iron Man’s Jarvis AI, without the cheeky, unsolicited advice of course. Having been integrated into a financial institution’s systems, Finn can access the couples’ financial data and give relevant feedback based on the latest market data Finn has been trained on.
Finn can even recall contextual information from prior conversations to make suggestions based on the couples’ expressed interests and preferences in the past.
Demonstration - Business Partner Flows
This flow demonstrates Finn’s capabilities to provide on-demand data. If a user were to be talking to their business partners about the type and size of a credit line, it would be deeply valuable to have an AI provide the facts and figures they would need, when they need them. It would be even more useful if the AI were to give interesting advice, as long as it was clear that what it offered was nothing more than advice.
Future Development
The launch build of the Finn platform would have to encompass a variety of features to truly be considered market-ready. The following is a comprehensive view of a potential MVP build of Finn Chat.
Decision tree-based onboarding: Users can ask financial questions and be guided through a decision tree of prepared questions and answers generated by ChatGPT. The decision tree is designed to help users understand their financial situation and make informed decisions. This would also guide users to the appropriate chat rooms or professionals as needed.
Natural Language Processing (NLP): ChatGPT’s NLP would most likely drive the core conversational features of the Finn Chat platform, as this is what allows users to have a more conversational experience. ChatGPT would be used to understand the user's intent and provide relevant financial advice.
Pre-trained models: Finn Chat would also have pre-trained models on financial data to provide more accurate and relevant advice to users. We would have to procure data models across relevant topics, but as mentioned earlier, segmentation would facilitate rapid training and marketability.
Fine-tuning: The platform allows users to fine-tune the model on a smaller dataset of financial data to improve its performance on financial advice. This feature can be geared towards larger, highly specific accounts. It’s always nice to have an on-call human representative, but on-demand access to an AI tool designed to specifically assist a single client could prove to be highly useful.
Integration with other tools: Finn Chat would be designed to be seamlessly integrated into other financial tools across a variety of existing systems and tech stacks. Financial institutions and banks may adopt Finn Chat to supplement customer-facing features such as calculators, budgeting apps, and other financial management platforms to provide more comprehensive financial advice.
Multi-user support: Users can invite other people, such as their spouses or business partners, to the chat and view previous conversations. Users can also have separate channels for different topics such as home loans, saving, investing, and credit.
Bank integration: Users can link their bank account to the platform and view their balance and transaction history to aid in decision making.
Live support: If users still have questions, they can be connected to a financial specialist who can view previous conversations to better assist them.
Security: The platform would have a secure messaging experience for families to better understand their financial scenarios and engage with financial professionals from their bank.