Chatbots, for better or worse, have become ubiquitous in our everyday lives. It’s technology in progress, as far as it goes, though too often, we mostly get the “bot” part; the “chat?” Not so much.
That’s changing for the better, thanks to “conversational AI”—artificial intelligence-driven customer service technology that has applications in everything from addressing a question related to a certain fee on a bank statement, (re-)negotiating a mortgage interest rate, processing health care claims or even handling payment arrangements.
<<< Start >>>
With conversational #AI, #chatbots get “smarter” to provide better, more “human” customer service
<<< End >>>
Conversational AI is rapidly growing and increasingly sophisticated technology, but its aims are quite simple and straightforward: productive interactions with customers and clients that are casual and informal, friendly and relaxed—human, in other words.
Yet, when was the last time you spoke to a virtual agent and felt like you were getting help to solve a problem or complete a task with someone or something who actually “listens” and truly understands?
There are many virtual agents at work in various industries and markets, and many people claim they’ve got the chatbot secret sauce nailed down.
Bots get smarter, customers happier
Still, the biggest problem is that most of these virtual agents are largely transactional—they get you so far, but once you reach more complex or contextual situations, the bot gets tripped up (and people start hanging up their phones or closing the chat window in frustration).
Take the aforementioned fee example: You could call your bank asking why you have been charged for certain transactions during your recent trip in Europe. A transactional bot may then ask you to provide credit card type, date period and the exact amount in question. If you have all this information ready and the bot understands correctly, it may then provide you with the fee details.
But there are many ways a user might ask about an unknown charge: “How come I have been charged a foreign transaction on my last statement?” Or, “I noticed an extra charge of a little over 10 bucks on my latest bill? What’s that for?” Even if we assume that a transactional bot can provide the "right" response to the above inquiries, it can easily fail on follow-up comments such as “But I don’t think anybody explained this charge to me when I opened my account.” Or, “Is there a way I could avoid this fee in the future?”
We need to find a human-centric solution that’s “conversationally intelligent”—a chatbot that not only understands how humans speak, but also senses and interprets tone and context—say, the customer is frustrated, or is travelling and pressed for time.
Our AI team at Accenture’s Liquid Studios in Silicon Valley has been working on just such a solution. Part of the goal is to establish and build around a four-rung “ladder of trust”—accuracy, transparency, consistency and empathy.
Additionally, while many chatbots are designed for a predicted straight-line user journey to fulfill any use case, we’re taking a very non-linear approach. We start by identifying a problem or need. From there, we develop a use case that ultimately shapes into a “conversation graph,” aiming to account for myriad ways a real-world, agent-customer phone discussion might play out: vague responses requiring clarification, user expressing concern, user changing their mind, contextual inquisition, expressing frustration, etc. We want it to be an organic process, much like an actual conversation, and anticipate and accommodate and bake this into our process.
Natural Language Processing (NLP) and Understanding (NLU) provide the flexibility for how someone may phrase a question. This enables, for example, a bank customer to express an inquiry somewhat vaguely—“How much money do I have right now?” or “I need to know what my balance is.” Or, the inquiry could be more specific and descriptive: “Tell me the current balance in my Platinum savings account.”
Helping in health care, other industries
Among a few specific projects, we’re working on applying our solution in the health care industry for clients who want to use AI and Robotic Process Automation (RPA) to more quickly and efficiently deal with patient questions and disputes over insurance claims, which are currently handled manually by people over the phone.
For example, through our solution, a health insurance company could upload a spreadsheet of customer information, then a conversational AI bot with domain knowledge places a call to alert someone of a claim denial and tries to resolve it by asking them to provide missing details. The solution would automate both incoming and outcoming claim resolution calls, saving a significant amount of cost and further reduce mistakes.
“Chatbot” entered the popular vernacular in the 1990s, but the underlying concept and technology originated in the 1950s and 60s (thanks to visionaries such as Alan Turing and Joseph Weizenbaum). By now, it’s clear both AI and chatbots are here to stay. It’s also clear it’s time to take the technology and make another leap forward.