2017 was largely marked by the rise of chatbots, conversational agents designed to answer a large volume of recurring questions from a brand’s customers automatically.
Beyond the hype, these messaging services are particularly appealing. The cost of developing a bot is far lower than building a full-fledged application. In addition, automating tasks through these bots can reduce customer service and CRM costs. Even so, chatbots appear to have certain limitations.
The chatbot, a tool with limited capabilities
Chatbots, these conversational robots powered by artificial intelligence, are gradually appearing on users’ computers and smartphones. They are supposed to hold a conversation with a human via a messaging service. The aim is for these robots to advise customers and respond to their requests. Bots detect keywords and reply to consumers using pre-recorded answers designed by humans. They are therefore not fully autonomous, as human input remains at the heart of this conversational commerce model.
Facebook Messenger was among the first platforms to open up to advertisers by allowing users to communicate directly with brands and receive immediate answers to their questions. In a publication dated 22 January 2018, researchers from the artificial intelligence laboratory FAIR explain that if chatbots fail to conduct ordinary conversations known as « small-talk », it is because they do not have a coherent personality. Indeed, they are expected to conduct a large number of conversations with different people. To address this, the researchers created the « PersonaChat » database, containing more than 160,000 dialogue lines. These are excerpts from exchanges between internet users on the Amazon Mechanical Turk platform, which will enable chatbots to hold more relevant conversations with humans. However, there are nearly 11,000 chatbots on Facebook, which would imply building vast databases like the one created by FAIR.
Machine Learning, a technology on the rise
Machine Learning, or « automatic learning », may well be behind the limitations of chatbots. This technology can indeed be used to make work easier, generate savings or boost companies’ revenue. While Machine Learning serves several purposes, it acts as a substitute for all companies that do not have the necessary language capabilities. It aims to reproduce behaviour by designing a system capable of learning from examples of problems to solve. This technology is therefore constrained by the quality of the data fed into it. But it is also this ability to reproduce behaviours without having to understand them that means companies are more likely to turn to Machine Learning than to chatbots. Finally, it should be noted that the productive side of the chatbot is not really a selling point. However, players are beginning to imagine increasingly broad use cases, such as automating internal management services, which should help reveal its full potential.
