A few years ago chatbots were all the rage and consequently Google, Microsoft and many other offered frameworks that can help you build chatbots.
They are all built around setting up a decision tree to steer the conversation to an endpoint and some natural language processing (NLP) capabilities. The decision tree represents the logical flow of the interaction between the user and the bot.
An order-taking chatbot in a pizzeria could, for example, first ask you what kind of pizza you want and present you with the binary choice of vegetarian or non-vegetarian. After your first choice, you’ll be presented with the next step until you finish your order. NLP capabilities can be used to help understand written text input from the user.
Decision tree systems have a long history of being used to methodically come to conclusions in various scenarios, but they are falling out of fashion. Let’s examine the reasons why.
When chatbots work
Before diving into why people are abandoning the decision tree approach, let us first look at some cases where it can work:
Triage. In medical emergencies (911 calls) the call-taker is trained to go through a predefined decision tree in order to arrive at the fastest possible conclusion regarding the patient’s condition. This setting is time critical and highly emotional so a formalized and highly optimized schema makes a lot of sense.
Non-complex interactions. Simple exchanges with our bank or insurance company such as checking a balance or reporting less complex insurance claims can be handled well with a decision tree approach.
Now to some of the reasons why the chatbot approach might be falling out of fashion:
Complexity. Often the problems that need to be tackled are too complex to be represented by a mere decision tree.
Laborious setup. It takes a lot of manual work to set up the best logical flow for the decision tree and the solution could be vulnerable to sudden changes in circumstances.
Lack of user-friendliness. Many users just want to get their questions answered and are not really interested in a long back-and-forth exchange to get the answer.
Technological advances. AI technology and more specifically deep learning based NLP has advanced so much in the last few years that we no longer need the chatbot approach. For example, the Google duplex AI makes reservations by having a natural conversation with the restaurant. During the conversation the AI negotiates the booking and makes sure that the call-taker has understood your intentions without the need for an explicit decision tree.
Chatbots can be a convenient way to solve very specific problems requiring simple solutions. However, as we’ve seen above, they have their limitations.
In our coming blog post series “The science behind the raffle-lution,” we will discuss the technological advances and research community trends that have allowed raffle.ai to build state-of-the-art AI question-answering systems for customer service.
We’ll also discuss the emerging technology that will allow us to develop ever more sophisticated products to tackle challenges that organizations will face in the future.
To find out how our AI technology is smarter than a chatbot, get a product demo today:
Applying a data-driven approach means we use training data instead of defining rules, distinguishing ourselves from traditional chatbots.
Ole Winther
Apr 15, 2020
Sign up for insights from raffle
Get the latest resources, events and webinars sent straight to your inbox. You'll learn about AI, customer service, employee engagement and much more with our knowledge-filled newsletter.
