Oh my chatbots…

After hearing about this hype, I thought it was just another idea to quickly launch new apps with little money. You can deliver almost any feature through a chatbot dialog, saving hours of time you’d otherwise need to design a set of usable screens. No need to conceptualize a structured user flow, as long as the user flows with the bot, chatting along while booking flights and shopping for shoes. The only problem is: chatbots usually are annoying. They fail, they make you try and wait. 

It’s not their fault.

When you look at how chatbots were introduced – as all-knowing FAQ engines — you get the idea of what went wrong. To fulfill that role, an enormous amount of (industry-specific) data and training are needed. And in many cases that didn’t happen. Instead, companies deployed chatbots using off-the-shelf frameworks with built-in Natural Language Processing (NLP) tools, assuming it might be just another small step to add specific needs with some training and a solid data set. However this step is anything but easy or small and companies have underestimated just how much work it is to implement dynamic, useful chatbots.

Users got presented with chatbots that were simply not equipped to deal with the many possible inquiries. When the bot failed, users got frustrated and shifted away.

C’mon, now give it a chance.

That’s why, when I was asked to design a restaurant recommender app based on a chatbot, I almost visibly rolled my eyes before putting on a big smile and half-heartedly exclaiming “yey”.

However I did not really know a thing about chatbots and how they may become useful some day. What I found intriguing was the idea of a little companion which will learn how to sift through a large amount of data to provide you with the most relevant results.

Because learning works so well through interaction, I figured that an environment of permanent thought-sharing between man and machine would be a good opportunity to train the machine efficiently. So I let myself in on the chat idea and happily started with speech bubbles and input fields.

@