Chatbots — how do we talk to them?

It is an easy mistake to make.

In this day and age, knowing whether you are talking to a human or a robot in a chat can be difficult to figure out. For us at Convertelligence, it is very important to make sure that our clients know how crucial it is to let their customers know they are talking to an automated service.

Not only because of privacy issues and the fact that some customers might feel “tricked”, but also because of how they interact and talk to the chatbot. Humans are able to decipher and understand long passages of text and will, in most cases, have no trouble answering the problem in question. A chatbot, however, will be programmed to only understand short, to-the-point sentences.

If a chatbot receives a short story containing a complex issue, it will most likely produce an answer looking something like this; “I’m sorry, I didn’t quite understand. Would you like to try again?” (to which the customer probably will reply something like “f**k off”). To quote my colleague Petter Hohle’s article:

“In order for the chatbot to answer questions for you, the chatbot needs to understand what your customer is saying when they send a message your way. This type of understanding is called Natural Language Understanding (NLU for short), […] Humans understand natural language effortlessly (their mother tongue, at least), but with machines, it’s a different story. They lack a very important piece of the puzzle: the human brain.”

How do we solve it?

So how do we tackle this issue? It is a tough one because as chatbots become more “human” and more widespread, customers will expect more and more from them. It is expected that chatbots should be able to answer complex issues. One way is to subtly tell the customers what issues the chatbot is programmed to answer and hope that the instructions serve the purpose of guidance rather than being a hindrance.

However, it is easy enough telling people what to do, the hard bit is actually making them do it. One way we are trying to solve the issue is by gently pushing the customers in the right direction. By adding a “remember that I respond best to short sentences” or something similar, in the bot’s welcome message, we have already hinted to the customer that there is no human being at the receiving end of the conversation. Our hope is that the customer will then abandon the novella they are preparing to write and instead go for a simple “can you help me?”.

Keep it short and sweet

As humans, we mimic each other and so our aim is to make the customer mimic the chatbot’s short and precise language. This way the customer will most likely receive the answer and help they are looking for. This will please them and they will return to use the chatbot at another time (and tell all their friends about it). The company to which the chatbot belongs will also be pleased and will tell all their friends about how we (Convertelligence) make chatbots that work the way they are supposed to. All in all, by making the human users utilise the chatbots correctly by talking to them in the way that is intended, everyone is happy.

The use of chatbots is undoubtedly increasing, and even though robots and AI are on the rise, we are a long way away from making them clever enough to understand extremely long texts. However, the more used we get to chatbots and the more common they become, our hopes are that talking to/with them, using short and precise sentences will become the norm. Until then, the next time you come across a chatbot, maybe try to steer clear of the long rant about how your electricity bill was high, even though you have been on holiday in Spain and not used your washing machine for the past 3 weeks, and instead go for a simple “why is my bill so high?”.

Who knows, you might get the answer you are looking for.


Frenchie’s icon was made by Freepik from www.flaticon.com.