What is a chatbot and should you care? All you need to know.
You may think you know what a chatbot is and have dismissed them as a gimmick. But they might be worth another look.
I cannot express how much I hate Siri. She forever activates when I am not speaking to her. When I do ask her a question she either doesn’t hear me or misunderstands the question. Not that Alexa or any of the others are any better. They all suck.
So you can only imagine my ambivalence when everybody started talking about chatbots. What a pointless gimmick!
You may feel the same. I don’t blame you. That said, I am beginning to come around to the opinion that they have their place.
But before we get into the role of the chatbot, let’s answer the question: “what is a chatbot?”
What is a chatbot?
There are two types of chatbots. The audio based variety such as Siri, Alexa and Cortana. Then there are text based versions such as those in Slack or Facebook Messenger.
In many ways we have tried to run before we can walk with chatbots. But they still have a lot of potential.
In both situations the user is having a conversation of sorts with a computer program. That computer program is attempting to mimic human responses. Generally speaking companies use them for some form of customer support.
Text based chatbots have been around in a basic form for years. For example we’ve been able to have simple conversations using automated text messages. There have even been some examples of it happening via email.
But more recently we have seen an explosion of interest in Chatbots. This excitement reached new levels when Facebook supported them in their messenger app. Some announced that the graphic user interface was over. Soon we would all be having natural conversations with applications.
In reality chatbots have faced some major usability problems.
The problem with chatbots
In many ways we have tried to run before we can walk with chatbots. With voice based chatbots such as Siri, we have set ourselves an ambitious goal. A program capable of hearing and interpreting verbal commands.
This presents two big challenges. First the program has to understand the words users are speaking. Developers have made great strides in this area. In fact this post is being dictated. But as soon as there is background noise the ability to interpret what is being said diminishes.
The bigger problem lies not in understanding the words but interpreting their meaning. Take for example the simple act of responding in the positive. A user could say:
The list could, and does, go on.
The user is often left to guess what word or phrase the application will recognise.
In many ways this reminds me of early text based adventure games I played on my Spectrum. Was I supposed to type “hit goblin” or “attack goblin”?
This problem is not limited to voice activated chatbots. Users have to take part in the same guessing game when using text based chatbots.
The result is that instead of a natural conversation, users have to find the right phrasing. It feels like the days of mystery meat navigation all over again.
But there is potential in chatbots. We just need to be more realistic in what we can expect them to do.
Overcoming the problems with chatbots
When I first came across Adrian Zumbrunnen’s site it was a lightbulb moment for me. The idea that chatbots worked if the user interacted using dialogue options. That instead of open text entry they selected from a limited set of options.
Of course this is something video games have been doing for years. In conversation with game characters you have a series of choices of what you could say. Those choices dictate how the conversation branches and computer characters respond.
Adrian has taken this approach on his website. He gives the user options. The chatbot then responds in different ways depending on which option they select.
But Adrian doesn’t stop there. The Chatbot adapts depending on various factors. Factors such as previous visits, if it knows their name or if they arrived from another website. In fact it would be easy for the bot to adapt to everything from time of day to geographic location.
All these subtle adaptations gives the impression of intelligence. This impression is reinforced because the bot never misunderstands the user. After all the dialogue options are set.
Not that open dialogue is impossible. You can still add it into the mix, but in a much more controlled way. For example the chatbot could ask the user their name or email address. This kind of information is easier to parse and so you can be confident in your response. You can also be more confident if they enter something wrong. This allows you to respond in a more realistic way.
So where does this leave the role of chatbots? After all the chatbot on Adrian’s site is still a bit of a gimmick. A proof of concept if you will. But I do believe that in some areas Chatbots have real potential.
The role of chatbots
We are still in the early days of Chatbots. But already there are some obvious applications. Applications I have started to explore with some clients.
The most obvious application for Chatbots is to help with customer support. It could well be able to provide many answers for customers. All without the need for human interaction. But where the user does need human support it could hand off to customer support staff. That or gather the users contact information to followup later.
Chatbots also open up the possibility of guiding users through complex interactions. Interactions that need the user to complete a series of steps. For example it would be easy to imagine a Chatbot being a better way to book a holiday. Or help a student find a course to study at University. These activities need the user to make a lot of choices or enter a lot of data.
We could use Chatbots to educate a user too through a guided conversation. Whether that is helping a user learn a new skill or orientating them in the use of a new application.
But all these ideas are the tip of the iceberg. Chatbots provide a wealth of opportunities. For example guiding users through complex search queries. Or making contact forms feel more like a natural conversation.
Best of all, getting started building a chatbot is easy.
Building a chatbot
There are no shortage of services that will help you build a Chatbot without the need to code. One that seems to stand out from the crowd based on my research is Motion.
What attracts me to Motion is that you can deploy it to a variety of platforms. For example it supports Facebook, your site, text and even email.
That said, if you want to do anything complex you will need the help of a developer. For example out of the box Motion won’t remember a user when they return to your website.
The point is that getting started and experimenting is easy. Tools like Motion typically have a free account option. That means you can take your time crafting the conversation. Because when it comes to Chatbots the devil is in the detail.
The devil is in the detail
At face value designing a Chatbot conversation seems straightforward. It is only a branching conversation after all. But to make it feel natural there are many details to consider.
For a start the writing will be essential. Most of us are bad at writing personal, engaging copy online. But that is essential when creating a Chatbot.
Then there are the extras I mentioned earlier that give the impression there is a person behind the bot. Things like remembering you, noting where you came from and understanding context. For these things you will need developer support.
Finally there is still a lot of work for the designer too. Details such as chat bubbles animation and how fast the bot responds to a user. All these help shape the experience.
There are many people out there working hard to get these nuances right, but there is still so much to learn.
Still so much to learn
We are in the early days of Chatbots. I am also pretty sure they aren’t going to replace existing user interface models. But I do believe they can supplement them. They add a naturalness to an interaction that is so often lacking in digital. The trick is to avoid too much open data entry on the users part.
Originally published at boagworld.com on February 25, 2017.