In financial services, chatbots are becoming a buzzword on par with Big Data, Machine Learning, and Artificial Intelligence. In fact, some companies market their chatbots using this holy trinity of buzzwords: “Machine Learning, Artificial Intelligence systems to leverage Big Data.” The truth is that chatbot marketing is well ahead of the technology itself. Often, we are sold this idea of how wonderful chatbots could be for our business, but in many ways, the technology falls short. So why do chatbots fail in finance?
I am not saying that chatbots can’t improve Customer Experience (CX) while decreasing inbound calls to your call center. But before undertaking a chatbot project, you need to separate the marketing from the reality when it comes to assessing chatbot technology. You also need to realize that all chatbot technologies are not equal. Here’s what we’ve found as challenges within the finance industry when it comes to designing and implementing chatbot technology.
1. Do you speak Chatbot?: Chatbots only “know” the answer to specific Questions
This is a common problem that is well documented. Here is a personal example. My Credit Card stopped working last month, so I tried the chatbot of a large bank (whom I shall not mention). Here’s our “conversation” from the beginning:
After several more tries, I was in a foul mood and had to call customer service and wait 30 minutes to speak to someone. At the end of this interaction, I, the customer, was angry and the bank wasted money serving me over the phone, one of the most expensive ways to address a customer’s problem.
It’s important to note that chatbots are made up of two pieces of technology:
Natural Language Generation (NLG): AI technology that can turn computer data into written, easy-to-read text.
Natural Language Understanding (NLU): AI technology that enables a computer to understand the written or spoken word.
Many conventional chatbot systems come from NLU companies. This means chatbots are focused on “understanding” what you say, but then they can’t respond because they can’t actually write, they only have a certain number of pre-written responses.
Conventional chatbots “respond” with pre-written responses. So even if they understand your question, they may not have an answer.
This is changing. There is a demand from companies, specifically in the finance industry, to enable chatbots to respond with dynamic, easy-to-read text. Here at Yseop, I’ve been on a team that has delivered smart chatbot systems in banking and insurance with great success.
2. Language Limitations: Chatbots Are Very English-Centric
It’s common knowledge that chatbots have difficulty understanding accents. While this is true, there is a bigger problem in today’s global world. Even as chatbot systems are able to understand more languages, conventional systems can’t communicate back in that language.
As a result, you need to have in-house experts pre-write responses in the additional languages. This is a slow, expensive, and unnecessary job because next generation chatbots are incorporating advanced Natural Language Generation software. Now, there are solutions available today that enable chatbots to understand and respond in multiple languages.
The simple reason comes from the first point: if you need to provide your response using chatbot-friendly phrasing, then why not just use the search filter on a FAQ?
Chatbots have over-promised and under-delivered, so now we have work to regain public confidence. The first step is to deploy a chatbot that can write dynamic text while applying your corporate expertise. The system doesn’t have to pass the Turing Test, but it does need to be able to ask intelligent questions, offer personalized responses, and provide a satisfactory online Customer Experience.
These types of projects are being delivered today, so it is already possible to get the benefits of a chatbot without the drawbacks, you just need to do your homework.