Back in 90s, during the dot com bubble when more people started using the desktop computers at their homes. I remember when my friends started talking about computers, I thought of it as robotic pet intelligent enough to do whatever you say. This myth was quickly busted once I bought one. Chatbot story is not different than this. I am going to discuss the performance criteria of the bot that makes it efficient.
It is a fact that chatbots can test limits of your imagination since they are not bound by constraints, their behaviour is quite limited when you integrate them in any business. People expect a chatbot to be artificially e.g. if you trigger a bot with a question “I am Umair from Melbourne”, a chatbot with AI should responds with a customized greeting which may include a comment about the city or if there is any event going on in Melbourne, response may include a reference to it.
However, do you really want your chatbot to respond to a question that is out of context from the current conversation? Success of the bot is due to its ability to be used as a personal virtual assistant rather than multi skills trained bot.
Chatbots deployed into organisations only do few navigational tasks. “Alex” deployed by “Australian taxation office” gets intent from the user queries, confirms its translated intent with the user and then redirects to the that specific section. As of 2016, Alex has responded to more than a million queries from the user (https://www.businessinsider.com.au/the-ato-launched-a-siri-for-tax-and-has-called-it-alex-2016-12). This means that the real KPI of the bot is not its ability to response to each question wittingly but to be able to guide a client to the correct destination as soon as possible.
This performance criteria of a bot can be improved by training the BOT from time to time. Let’s say you have added a bot for an organization who wants to use it as a personal virtual assistant. If that organization has never implemented a bot before, you won’t be able to train the bot with correct query/response behaviour at the first place. Initial bot configuration will be on the inputs from the organization that may be just a guestimate regarding the user queries. You need to improve the bot behaviour by re training it over the period of time
Once you reconfigure your bot based on these inputs from the clients. You can improve its performance and the bot will be able resolve customer queries efficiently.