Conversational UI in Banking: Say Good Bye to Your Boring App
How many times do we all log in to our banking app and struggle to find information? We end up searching and encountering a plethora of FAQ links. Then, we attempt to connect with an agent and find ourselves in a 30-minute queue. Sounds familiar, right?
This is precisely where Conversational UI banking is revolutionizing the retail banking industry. The vision is simple: log in to the app, authenticate yourself, and pose questions naturally. The app responds promptly with accurate answers. For instance, you can inquire about your current balance or whether you’ve paid last month’s water bill. Beyond just offering information, you can instruct the app to perform actions, such as paying the water bill.
In this article, we will explore the key use cases in retail banking where Conversational AI is set to play a pivotal role.
What is Conversational UI
Conversational UI, or User Interface, is a way of interacting with computers or machines using natural language, similar to how we talk to people. Instead of clicking buttons or typing commands, you can have a conversation with a computer through text or speech. It’s like chatting with a friend, but you’re communicating with a program or system that understands and responds to what you’re saying in a human-like way.
It’s important not to confuse this with traditional bots. Current traditional chatbots operate using pre-defined rules; for instance, they follow a decision-tree workflow like responding “Y” when the user says “X.” They resemble automated phone menus where users navigate through selections to find answers.
Conversely, Conversational AI bots possess context awareness and are trained to comprehend user intent. They engage in free-flowing conversations, fueled by a Large Language Model that serves as a bridge between users and backend systems, ensuring a seamless user experience.
Examples of Conversational UI in Banking.
There are 5 categories of interaction the Conversational AI can help. The use cases below I will mix Bank and credit card related use cases.
1. Account related questions
“Conversational AI can help streamline account-related queries. Users can swiftly inquire about balances, transactions, and account details, receiving immediate, accurate responses. The AI can also guide actions like password resets, fund transfers, and account updates, providing a user-centric experience that simplifies financial management and enhances interactions. Here are some examples of questions users can ask:
- Balance Inquiry: User: “What is the current balance in my checking account?”
- Transaction History: User: “Can you show me the transactions from the past week on my credit card?”
- Card Activation: User: “I received a new debit card. How do I activate it?”
- Password Reset: User: “I forgot my online banking password. How can I reset it?”
- Transfer Assistance: User: “I’d like to transfer $500 from my savings account to my checking account. How can I do that?”
In the above example, the Large Language Model takes the user request for account balance and translates that to an API call and sends that to the backend system to respond back. When the backend responds back, the LLM translates the information in to a meaningful sentence to respond back to the user.
2. Payment related questions
Conversational AI can simplify payment queries, allowing users to inquire about due dates, bill history, and even schedule payments seamlessly. The AI can fetch accurate information and assist in tasks like setting up auto payments, making transactions, or updating payment methods. Examples include:
- When is my next credit card payment due?
- Have I paid the mortgage for this month?
- Can I schedule a payment for my water bill?”
- Can I view my payment history for the past month?
If you notice, filters are applied to the query along with corresponding actions. For instance, if a user requests information about bills due next week, the LLM translates this into an API call to retrieve bills due and adds a time filter for the upcoming week. Furthermore, in the subsequent interaction, the LLM utilizes the references for the identified bills to schedule payments on their respective due dates. Please note that the LLM can also respond back with charts in addition to text.
Note that the advantage of such systems is multi language support. Here is an example of similar question asked in an Indian language.
3. Updating information
Bid farewell to confusing forms. Simply express your desired action to the app, and it will intuitively prompt you for the necessary details before efficiently carrying out the task. Say hello to a streamlined, user-friendly experience that simplifies your interactions.
In the near future, conversational AI bots will likely take over the role of handling most existing forms, engaging with users more effectively.
4. Promotions
A conversational AI chatbot has the capability to analyze past customer information, enabling it to recognize chances for upselling and cross-selling within the existing customer base. This enhances personalized interactions, fostering effective marketing strategies and improved customer engagement for businesses.
5. Unsubscribe
We’ve all paid for services we never use and often find ourselves being lazy when it comes to quickly checking, canceling, and unsubscribing. Conversational UI empowers users to effortlessly check, cancel, or unsubscribe from services they rarely use. By offering an intuitive platform for quick interactions, it eliminates the hassle of managing subscriptions and enhances user control over their expenses.
Backend integration for seamless operation
Large Language Models (LLMs) play a pivotal role in bridging user inputs with backend inquiries, retrieving responses, and presenting them in a dialog format. However, achieving this involves more than just transmitting LLM-generated text and receiving responses. It entails accessing specific data stored in systems, often via APIs with robust security measures.
For instance, when a user seeks their account balance, the LLM must log in on the user’s behalf through APIs, formulate a query, and retrieve the desired information, subsequently formatting and delivering it to the user in a coherent manner.
The strategy to connect to various systems can be programmed into plugins. These plugins guide the LLM in selecting the appropriate plugin based on the nature of the request it’s handling. This intricate architecture involves an interaction layer linking an array of plugins, which, in turn, establish connections with backend systems.
Consider the scenario of initiating a transaction. The LLM must furnish relevant context to the backend system, such as the account ID and transaction particulars. This context is relayed to the plugin responsible for this type of transaction, which then gathers the necessary data. The LLM takes this data and seamlessly crafts it into a dialog format for the user, ensuring a fluid conversational experience.
In essence, effective conversational AI involves intricate backend integration. LLMs act as intermediaries between users and backend systems, driven by plugins that enable specific functionalities. This combination ensures smooth interactions, whether it’s responding to FAQs or retrieving data from databases
What to watchout for?
There’s a potential downside to AI chatbots in finance. Large Language models tend to generate incorrect information, a phenomenon referred to as “hallucination.” It’s crucial for chatbots to be trained to provide accurate information to prevent misinformation and privacy breaches. Failing to do so risks eroding trust and customer satisfaction. Additionally, if chatbots make it difficult to connect with human representatives, customers might lose trust in the institution and its services.
One common metric used to measure the success of Conversational AI is containment. Containment indicates whether the entire conversation remained within the AI bot and didn’t escalate to a human agent. However, high containment doesn’t always guarantee issue resolution. If AI cannot seamlessly transfer users to human agents, frustration might lead users to abandon the conversation, despite technically achieving containment.
In some cases, success is achieved by directing customers appropriately. For intricate matters like mortgages, steering users toward human experts proves more effective, even if it doesn’t strictly adhere to containment.
I will write a separate article on North Start metrics for Conversational AI.
The Road Ahead: Embracing Conversational UI Banking
As the banking industry continues to evolve, Conversational UI emerges as a transformative force poised to revolutionize user experiences. The ability to swiftly provide responses, engage in personalized interactions, and seamlessly execute tasks empowers users to navigate their financial landscape with unparalleled ease.
Much like how many banks embraced dashboards in the past, the next 12 months are likely to witness a surge in banks adopting conversational AI interfaces. Some institutions will opt for an internal development approach to create these interfaces, while others will choose to procure the technology from vendors.
This marks the point where traditional Business Intelligence (BI) vendors can leverage their expertise to introduce conversational UI interfaces and develop such applications. Similarly, low-code vendors, already equipped with an array of plugins, can seize this opportunity. Additionally, specialized AI chatbot vendors catering exclusively to banking will provide comprehensive, out-of-the-box experiences.
Indeed, the future holds promising prospects for the integration of conversational AI, poised to transform how we engage with our banks. Exciting times lie ahead in this evolution of banking interactions.
