Ideas To Make Chatbots Powerful
Let us jump right to the point. We have all experienced chatbots, some good some bad, some powerful, others not quite. In my own experience of constructing powerful conversational designs for chatbots I have seen myself write just about a better version of an IVR manuscript to filling up a full blown sheet defining entity and intents. What I have learned from my own experiments is that you can build a chatbot in 2 minutes or you can build one in 2 years, it will still have flaws. It will still be a WIP however, that’s not necessarily and bad thing…
“You can build a chatbot in 2 minutes or you can build one in 2 years but it will still have flaws”
How do you deal with these flaws? How do you set user expectations? How should you constantly work at making your chatbots more powerful? How do you make your chatbots pass the Turing test?….Well I am no NLP expert so I really don’t know! However, what I do know is that every flawed system can be made better! After all I am just some one who designs conversations, so if you were looking for charts with algorithms and complex mathematical equations, this article I am afraid won’t do you any good. But stick around if you are interested to learn and implement some of the conversational tweaks that I have used in an attempt to make bots ‘look and feel’ intelligent
The three main challenges that I feel haunt every chatbot are:
- Too many expectations
- Small talk
- Discoverability
Now lets look at each one of those a little closely and try to find a solution in an attempt to decrease the flaws. Skeptics may leave reading this article from here on
Too many expectations
Well, we are humans after all, curious and always on the hunt for flaws. After all isn’t that the basis of our evolution? With every flaw we conquered we grew powerful and eventually unstoppable!
“Chatbots are fundamentally flawed ”
Yes, it’s true. Of course we have bots like Mitsuku that are raising the bar for intelligent bots to a whole new level but take a look at the following diagram shared by the creator of Mitsuku — Steve Worswick
I honestly don’t even know how to react to this picture! But hey it’s not every day that you build a chatbot that clocks to become a finalist for the Loebner Prize three time back to back right? but hey don’t get discouraged, there’s a bright light at the end of the tunnel and it’s letting users know exactly what is programmed to do. A common practice is to share a gist of the exact reasons as to why the bot was created. What purpose does it serve…and so on. Instead of traditionally just mentioning it as part of your bot’s greeting message, you can also try making a custom GIF or GIF’s and let it pop up as your fallback message whenever the bot is asked questions that it wasn’t trained to answer. Remember, humans have a tendency to find faults first before fully appreciating the offerings of any product. (If you want more tweaks, ideas and suggestions drop me an email at abhishek.munian@gmail.com)
Small Talk
Well, hello there!
Let’s face it, we are all hard wired to small talk, after all that’s how we break the ice, isn’t it? Well any bot for that matter is expected to small talk with its users. While you may train your bot to answer to simple user queries like hi, alright, fine and ok it’s also important to make your bot user friendly by adding in fun content (unless your bot needs to maintain a serious tone). Until we see a time when natural language processing is truly advanced and remarkable, the truth is that we will need to manually train bots to handle small talk. A few efficient means to handle queries are by
- Understanding your users and optimizing your bot to handle the most common queries. Think how your users will interact with your bot. As much as you try, your bot will be met with initial skepticism. Your bot should be able to deal with not only the most common small talk queries but also attend to them in a way that feels natural. For example, If you have designed a bot that helps travelers find the cheapest tickets online, then your bot should respond with saying ‘Hope you have a safe trip’ when the user types in a mere ‘Ok’ or a ‘thank you’.
- You can also optimize your small talk to initiate a call for action. This makes your bot look proactive which will be much appreciated by your users.
Now let us all pray that one day advances in NLP enable us to to deliver witty, meaningful and funny responses to users at random
Discoverability
Ok, you made an awesome bot. It’s fine tuned after a lot of hard work but now what? What’s next? Yes, you are right….It is to make your bot visible. Let your users know how awesome your bot is! Go all out! ATTACK!!! I have written a separate blog altogether on bot discoverability which I will soon publish but let me quickly highlight a few best practices
- Web plugins on your native website that redirects your users to Facebook messenger
- The use of QR codes in banners, posters, all your promotional activities online, offline and even your Facebook page linked to your bot
- Creating a buzz on social media
While I have made a wild attempt in this blog to highlight certain aspects of chatbots that once worked upon make them powerful and stand out from the rest of the clutter, there’s still much more to be explored. In my upcoming blogs, I will uncover more methods, best practices and tips to help you make your bots a phenomenal success.
