Chat bots – Designing conversations [Part 3]

This article is the third in the series I’m writing about chat bots. Part 1 dealt about the use cases related to chat bots and the part 2 was about some pitfalls associated with building them. In this part let’s discuss about how we can start the process of designing a conversation.
Fancy two sets of conversations

Which one would you prefer??
I think most of us would prefer Conversation B over A. The difference between them is that the bot is seamlessly able to steer the conversation the right way without a user getting frustrated or feeling stuck. The conversations are kept interesting by showing vignettes about the topic and the voice and tone in general. This is what we would want to achieve by designing conversations and the most important part of it is the personality.

Conversations are very important human traits and when technology participates in something as intimate as that, it’s natural for us to associate a personality to that piece of technology – the chat bot. That is the precise reason personality has to play the pivotal role in the design of a chat bot.

How do we go about designing conversations?

The following steps can be taken, usually in that order

1.    Define the utility

You obviously are building a chat bot for a reason. In the first part of this series I had discussed various use cases. Most likely, you’ll have one of them as the use case but in any case, it’s empirical that the purpose is well defined before leap frogging into building the bot. Without that, we run the risk of creating a bad UX.

2.    Define the user base

Once you know what will the bot do, you must know who will use it. Just like a store will try and match their décor and even their staff to relate with their clients, a chat bot should try and reflect some of the traits of the folks who will use it. Doing this will come in handy when deciding the persona of the bot. A Pop or Hip Hop icon, could work with a young user base while Professor Dumbledore could work well as a coach for kids.

3.    Decide on a persona

You might have noticed that we have a character (or a persona) associated with the bot “Pnin” in the ‘Conversation B’ earlier. The usage of personas makes the process of designing a bot a lot easier. At the onset, if the team is aware of the persona and have some insight into the story behind it and any personality traits the bot should exhibit, instead of thinking what would I do when posed with a scenario, designers and technical folks automatically come on the same page – What would Pnin do? Out-of-the-box thinking right there!
It’s essential that the team has that clarity on what the bot is supposed to do and how it would behave before they begin on the journey of building it. The purpose and the user base of the chat bot should ideally determine the persona it represents but there are other things to it as well. If the company or a website for which the chat bot is being designed already has a brand image, that can be used as an inspiration for the persona. Maybe an existing celebrity or an animation character aptly defines the personality that your chat bot should represent. Let the creative juices flow.

4.    Chose a personality type

Use the Big Five personality traits to select which type of personality the bot persona would represent. Choosing a trait would help select words or sentences the bot would utter making it easy to design. You can also use some best practices like Agreeableness works best with virtual service agent while Neuroticism probably isn’t the best for a coaching bot.
A word of caution, this doesn’t mean that one should mimic human personality, the more it mimics, the more it will be compared to a real human and obviously will not match up.

5. Think Think Think

It’s time to bring out your imaginary childhood friends and talk to them as if it was a real conversation or imagine you’re writing the script for the next Hollywood blockbuster. Visualize the conversation. It’ll help if this is done by multiple people from the same user base as the chat bots.
A few topics of those discussions would be –

  • How many ways the conversation can start?
  • Various routes it can take? Take help from a Q&A database if it’s there.
  • Voice & tone based on the persona of the bot
  • The strategy to steer the user back in the scope of the conversation if they digress
  • Strategy to adopt when the chat bot is unable to handle the conversation

Remember to document everything as you go about it since it’s a long drawn process. A simple tool like Twine can help you with that. That’s your script for the bot.

6.    Get feedback before building

The time you’ve spent with your daughter drinking imaginary tea with her dolls comes in handy now. It’s role play time. So, before we start building the bot, it’s always better to do dry runs of the conversations. It’s best done with two team mates sitting and role playing with the script at hand and sorting out any issues or routes that were missed. A word of caution here, you will feel a bit awkward and are sure to get stares from co-workers unless you do it in a closed room. You will get better at it with time and maybe start enjoying it and who knows that hidden actor inside of you could just come out of the closet.
It’s also OK to take professional help for steps 3,4,5 and 6 – a person with a little known job role of a Conversational copywriter could do all that for you. It’s a difficult and time consuming process and creative folks can do a much better job than a technical team.

Conversational copywriting is a job role which is soon to become hot.

7.    Feedback strategy

Just like you would do bug fixes in your code when the application is deployed. Personality also needs slight adjustments as and when real people use the bot. The easiest way to do that is go through the chat transcripts and figure things out but that could become tedious as you scale. Think about automated means of finding good conversations from bad ones at the minimum. Maybe an emoji response could be a trigger or maybe a thumbs down or maybe a sequence of responses could indicate that the user is not satisfied. Finding bad (and good) experiences and using that feedback to improve the UX is very important – It’s a continuous process, an intelligent chat bot is supposed to learn and improve with time.
Once we’re thru with this 7 step process, it’s time to build your bot. Go ahead and play, the homework is all done now.
In my next articles in the series, I’ll start discussing about the implementation part of the chat bot. A Reference architecture to start with (hey! I’m an Architect, you know).
[About the author: Rishi Arora works as an Analytics Architect at IBM India Pvt Ltd and is currently working on building a Virtual Tutor.
*This article represents my personal views; I do not represent the view of my employer (IBM) in any way. This article was originally published on Linkedin.]

Sign Up for nextbigwhat newsletter

The smartest newsletter, partly written by AI.

Download, the short news app for busy professionals