How many times have you read an article about the best chatbot? For sure, when you dig into chatbot specialized media you find many articles on this topic. And the truth is that most of them will differ in their choices.
We’ve preferred to look beyond these options and try to find the common characteristics they all have. If you want to look at different examples, just google the last winners of Loebner Prize or the Annual ChatBottle Awards.
The problem of using buttons:
The first thing you can appreciate is best bots lack buttons for predefined options. These buttons are trendy, but the matter is that in most cases they diminish UX. Why?
We as humans don’t always talk in the same way but chatbots, even if AI powers them, they are not smart enough yet to understand all the different variations. In most cases, they only understand a normalized one. To avoid answers containing error messages, companies have opted to display bubbles with different answer options to make users’ lives easier. This makes sense because if bots fail to understand users’ requests for sure they will fail to answer them.
So why do we say predefined options reduce user experience? Very simple: options are limited.
Usually, bots that use buttons show three different options, but do we as humans propose only three choices to our interlocutors? So, if we want chatbots to perform like humans... are buttons really the right option?
Many people will say that chatbots are powered by artificial intelligence and they have their limitations. But if they are supposed to substitute humans for jobs like call-center operators or customer support agents, we need to make them much more human-like.
Looking for solutions to avoid predefined options:
- Machine learning approach: the clear majority of NLU systems use deep learning algorithms (Dialogflow, LUIS, Watson, Lex). ML seems the right solution in this era that allows your product to be used by millions of users, and therefore learn from the interactions with them. But there are still issues that remain unsolvable with this approach: for example, users don't talk to chatbots normally, as you would want, but aim to break them, and reaching high levels of accuracy turns out impossible.
- NLP middleware approach: a less known solution that puts linguistic knowledge to serve bot builders. Unlike statistical methods, it allows users to talk naturally and reaches high levels of accuracy in key tasks like intent detection. Therefore, one can expect a truly conversational experience from chatbots that benefit from this middleware. Of course, it can be integrated with any working NLU system.
As for us, we have already made an existing bot more conversational and are happy with the results. If you want to see how a NLP middleware enhanced a limited bot, download the use case below.