Conversational Commerce and Bots
You may have heard a lot about bots lately. But what are they, and how do they relate to conversational commerce?
First, conversational commerce is a term recently introduced by Chris Messina. We define conversational commerce as
“Conversational Commerce is a new way for people and businesses to interact in a world driven more and more by text messaging and with intelligence driven representatives using language and voice.”
In practical terms, conversational commerce utilizes either a chat messaging application, or a voice interface to talk to companies and brands. The chat and voice interactions provide both context and convenience throughout the conversation thread, enabling more efficient user interactions. Sometimes this conversation is with a human, but more often than not it is with a bot that can understand what was asked and respond appropriately. Regardless of whether chat or voice is used, the underlying bot technology that receives the request and generates the response is the same.
The goal of a bot is to make you feel like you’re chatting with a real person. While bots don’t have the breadth and depth of knowledge that a person does, they can be very useful when focused on a specific topic or function. For example, some bots are used to handle customer service requests, while others are used to help schedule meetings or work with your emails.
In reality, bots aren’t new. They’ve been around since the 1960s, when a chatbot called ELIZA was first developed at MIT. Since then there have been numerous attempts at chatbots, but none have been very successful. That is, until recently. A key reason is that artificial intelligence, machine learning and speech recognition – all important bot technologies – have recently seen dramatic improvements. A second reason is that more people use messenger apps than any other application, including social network apps. Thus, it is easy and convenient for people to use an app they already live in to converse with people, brands and services via a bot.
Much has been said about mobile application fatigue. Users already have numerous apps installed on their mobile devices that are specific to a single brand or function. Studies show that most of them are no longer used. As users become more resistant to downloading another app, companies are looking for new ways to reach them. So, instead of downloading and installing another app, many companies are providing bots that offer customers the convenience of using existing messenger apps to communicate instead.
Permute or Understand
Bots vary dramatically in both capability and usefulness. In general, they can be mapped onto a machine learning spectrum shown in the figure below. The spectrum ranges from simple bots at the low end, to sophisticated bots and assistants at the high end. As bots move up the spectrum the utterances and responses progress from permuted to full understanding based upon artificial intelligence.
At the low end, simple bots are designed for a specific function. All questions that the user can ask are permuted. For example, the bot may have been programmed to understand the phrase “What is my account balance?”, but not “How much is my account balance?”. There is no real understanding, so phrase recognition depends upon how thorough the list of phrases are. In addition, the bot’s function is simple, where each phrase corresponds to a single backend request and response. An example would be a request for a specific amount of money to be transferred from account to another.
At the high end, smart bots and assistants are capable of understanding the intent of the user without requiring every combination of words to be permuted. Rather, they use artificial intelligence which can understand language to devine the answer or action from the user’s request. Furthermore, intelligent bots continuously get smarter as they learn from conversations with users by applying machine learning techniques. This class of bots often utilizes deep learning to infer and draw conclusions from large amounts of data and notify the user of issues or patterns. An example of a smart assistant is the IBM Watson Health that can help physicians, administrators and researchers provide better health care. These types of systems require highly skilled knowledge workers and computer scientists over a significant period of time to develop them.
Between these two extremes are where many bots currently live. The bot may apply machine learning techniques to varying degrees for understanding phrases or finding the best answer to a question. For example, a news service could use a neural network classifier to determine if the user would be interested in a particular news alert, or perform sentiment analysis of a tweet and route negative ones to the customer service. Some of these bots can be easily created using the various available bot toolkits and frameworks, coupled with cognitive services, such as the IBM Watson Classifier or Retrieve and Rank services.
Most of the bots that we have developed in Emerging Technologies fall in the middle of the spectrum. We have applied semantic analysis and natural language processing, or NLP, to understand the question from the user, and then use a ranker or other NLP algorithms to determine the appropriate response. Future blog posts will describe some of our work in this area.
Since the uptake of bots is still relatively recent, there are many other issues that deal with both management and functionality of bots. Some of the challenges that we are currently investigating include:
- How do we locate what bots are available, what they do, and easily install them?
- How can we find out what can be asked within a conversational experience with a bot?
- How can language parser learn when it doesn’t understand the user’s question?
- How should interactions adapt and become personalized over time?
- How to continually learn and improve upon the interaction experience?
- How can conversations be composed into reusable building blocks?
- How can multiple conversations be combined or chained together to respond without being programmed?
In summary, conversational commerce is about establishing a new value proposition for businesses with simplicity for the user or customer at the center. Bots and assistants will play an essential role in simplifying how users can interact with businesses and services. We will no longer have to learn to use the machine, the machine will learn from listening to us, resulting in a natural conversation with the digital world around us.