Financial Chatbots Are Coming, But It’s More About the “Bot” Than “Chat”

Financial Chatbots

Tyler Griffin, Entrepreneur in Residence at CFSI, recently wrote a wonderful article on financial chatbots, covering the history of their use with a focus on the modern-day user experience and highlighting their current limitations, especially those not powered by “true artificial intelligence.”

Griffin’s article brought up a lot of great points. Because some financial chatbots operate through SMS messaging, they don’t require users to download another app. This is a huge bonus for consumers with limited storage space on their devices. And from a development perspective, chat functionality can be deployed fast, resulting in rapid validation of new features.

Retrieval-based Models vs. Generative Models

He’s also right about one of the biggest issues with financial chatbots: When bots lack strong natural language processing capabilities, users are forced to learn how to phrase their questions to get a meaningful response. For financial chatbots to be truly useful, they have to be able to process the complexities of normal conversational language— many simply aren’t capable of that yet, relying on retrieval-based models as opposed to generative models.

“The biggest difficulty with these products — at least the ones we’ve seen to date — is that there’s relatively little real artificial intelligence behind the scenes. Instead, there are basic text parsers that function an awful lot like a command-prompt interface. You might be able to ask an app “what can I spend this weekend,” but if you try a slight variation — say “how much cash do I have for Memorial Day,” you’ll probably get an error.”

While I agree with much of what Griffin wrote, I think he focused too much on the “chat” component of artificial intelligence technologies and neglected the “bot” part, specifically the use of machine learning to recognize patterns and make predictions. To be clear, I’m not saying chat functionality isn’t important; on the contrary, a financial chatbot that can’t communicate is pretty worthless. But I see chat functionality as the interface of financial chatbot technology, rather than the core product. The real potential of financial chatbots lies within their ability to “think” on behalf of a user and then deliver useful information to the user through a chat interface. So forget about chat for a minute, and focus on the “thinking” part.

The real potential of financial #chatbots lies within their ability to “think” on behalf of a user. Click To Tweet

Money Management is Hard

Let’s be honest, it takes an enormous amount of mental effort to manage cash flow, track spending habits, and plan for both short and long-term savings, and this is only made more difficult when consumers have accounts with multiple banks or third-party platforms. Even worse, most of us are irrational with our money, making short-term decisions based on emotion, which over time can sabotage our long-term financial health.

Financial chatbots have the potential to remove this cognitive load by delivering vital information and directives to consumers instantly and through familiar, conversational channels. Through the power of machine learning, financial chatbots can help identify meaningful patterns and even make predictions to help users make smarter financial choices and improve their financial health.

Pattern Recognition

For instance, say you have a user who goes to a video game arcade on a regular basis. Because this behavior is repeated over time, the bot’s algorithms learn this behavior and intelligently predict when the user is about to go play video games. This is a big deal because we can factor this behavior into a calculated projection to understand the ripple effects of the habit. A financial chatbot can then “talk” to the user about their projected financial position based on the predicted habit and associated expense. It’s still up to the user to act on the financial advice the bot provides, but chat is a powerful medium for providing actionable information and influencing behavior.

Griffin brought up a great point about the current state of chatbot interfaces:

“A chatbot-style interface works fantastically well when the tasks to be performed are very simple or — in the case of Digit for example — known ahead of time, and the messaging is mostly incoming. It doesn’t scale to advanced functionality very well, though.”

When the messaging is mostly incoming, financial chatbots act as a smart financial planner in a consumer’s pocket. This personified technology provides proactive communication and can think and act on a user’s behalf. This is what makes Digit great, not the fact that it operates over SMS text. And bots like this will soon be used across all communication channels.

Innovation is never easy; it takes time and work to develop the latest and greatest tools and even longer for them to be widely adopted by consumers. I think voice and text will beat out apps and GUI interfaces as the dominant way in which users access their financial data.

Bots are here to stay. And we’re only just beginning to unlock their potential.