During the past year, several major financial institutions announced new conversational banking services driven by artificial intelligence (AI). While semantically true—a person says something and a bot responds—it’s worth digging a bit deeper into the technology behind those services. In some cases, what you’ll find is that these bots are actually providing scripted responses using very little intelligence. You can teach your parrot to respond to the command, “Hello” with “Hi, how are you?”—but is that really a conversation?
There are indeed vendors delivering true conversational banking, by cognitive AI. Digital banking concierges can provide customers with personalized interactions driven by data and a complex understanding of banking processes. Customers will immediately notice a difference between this kind of interaction and a more scripted one. In this article, we’ll examine two ways true conversational banking differs from an interaction with a banking chatbot.
True multi-turn dialogue, unscripted responses
No matter how unsophisticated a chatbot may be, it will be able to help customers perform basic tasks. For example: Finding an ATM, hearing an account balance and even making a payment— all simple exchanges that can be scripted in minutes and delivered through a conversational interface. Chatbots can even perform tasks for humans, such as opening and closing accounts, or moving money from one account to another. But when a customer attempts to go beyond the standard request-response, or Q&A-style, conversation that chatbots are able to handle, the chatbot will be incapable of completing the request.
Here’s why: Let’s say a customer wants to find the nearest ATM in order to make a deposit. He may say, “Please help me find the nearest ATM so that I can deposit a check.” A chatbot will hear the first intent, “find the nearest ATM,” and direct the customer to the closest one. However, because the chatbot isn’t really having a conversation with the customer, it doesn’t stop to contemplate the true intention behind the request, which is, “deposit a check.” The chatbot is trained to register an intent and perform an action. The chatbot is not sophisticated enough to have back and forth exchanges during which conversations go off on tangents, new intents are introduced, and the chatbot is expected to respond with human-like intelligence.
A digital banking concierge driven by cognitive AI will process the entire request and realize that the customer could use his mobile device to deposit the check. As this is a conversation and not a scripted back-and-forth, the digital concierge will fulfill the first part of the request (locate an ATM), but it will also ask the customer if he would prefer to deposit the check via the mobile app.
Customer: “Please help me find the nearest ATM so that I can deposit a check.”
Digital Concierge: “The nearest ATM is at 211 Main Street. But I can also help you deposit a check using your mobile app. Would you like to do that?”
The technology behind this interaction allows for more complex service with the proper context. Customers who ask questions such as, “Can I apply for a mortgage?” or “Can I consolidate my student loans?” aren’t asking for a literal answer. What they really want to know is, “Will I be approved if I apply for a mortgage?” and “Will consolidating my student loans help me pay them off faster?” Scripted chatbots can’t answer more complex, personal and indirect questions, and they can’t handle multi-turn dialogue. True conversational banking driven by cognitive AI can. This brings us to our second point.
Personalization driven by data
Chatbots are programmed to answer generic questions using generic answers. Extending our example, can a chatbot tell a person whether they should actually apply for a mortgage, or that they should actually consolidate their student loans?
Customer: “Will consolidating my student loans help me pay them off faster?”
Chatbot: “Some people find that consolidating their student loans helps them pay the loans off faster because of lower interest rates. However, one should find a lower rate before making the decision to consolidate.”
That’s helpful information, but it doesn’t answer the customer’s question. A conversational banking app driven by cognitive AI will understand that the customer wants specific information relevant to his financial situation. True conversational banking will be able to read through the customer’s account history, balance and existing loans, and ask questions, make recommendations and offer data-driven advice.
Customer: “Will consolidating my student loans help me pay them off faster?”
Digital Concierge: “I would suggest waiting until interest rates decrease. Your current student loan interest rates are 3% and 3.45%, respectively. The national average consolidation rate is 4.35% and has increased each month for a year.”
Customer: “Is there any other way for me to pay them off quicker?”
Digital Concierge: “You’ve been making recurring total payments of $350 for a year and your account balance has never gone below $4,000. Would you be able to pay more each month?”
These are two very important ways that AI-driven conversational banking differs from a static scripted conversation, but there are dozens of features that separate cognitive and chatbot technology.
Here are a few other differences worth mentioning:
Observational learning: Cognitive solutions learn from historical interactions and apply that intelligence to future conversations. What’s even more compelling is cognitive AI can also learn by watching. A digital concierge will remember that a customer wants to consolidate his student loans, but it will also observe in the background how a human service agent walks the customer through the consolidation process, it will then be able to repeat that process with future callers. Chatbots do not watch or learn.
Context switching: What happens if a customer calls to consolidate his loans, and during the conversation remembers that he wants to move some of his checking balance into his savings account? True cognitive intelligence will be able to complete this transaction and return to the consolidation conversation without having to start the entire process from the beginning.
Dialogue variance: At the start of our example, the customer used generic language in place of a specific precise statement. (“Should I consolidate my student loans?” versus “Will consolidating my student loans help me pay them off faster?”) This change in structure completely alters how a chatbot would respond to the customer. But what would happen if the customer was even less precise with his phrasing? What if he said, “Can I lump my loans together?” or “I have too many loans and I hate paying them off separately is there any way I can make one single payment?” In neither of these instances did the customer reference student loans. He also never used the magic word: Consolidate. A simple chatbot wouldn’t understand what the customer was trying to accomplish; it would escalate the conversation to a human agent. A digital colleague capable of real conversational banking would use its knowledge about the customer (he makes loan payments every month) and sophisticated dialogue comprehension (“lump loans together” means “consolidate”) to at the very least ask the customer, “Are you referring to student loan consolidation?”
Be vigilant when you hear the word conversation in reference to banking services. Many banking apps today are not using true cognitive intelligence to create natural dialogue between customers and technology. During the next several years, banks that find vendors who deliver real conversational banking, capable of multi-turn dialogue, driven by true intelligence, will provide dramatically better customer service than banks that implement a low-level chatbot.
Contributors
