When I first started playing with the idea of designing a chatbot, I knew it would be very different to creating a website or app but I had completely underestimated just how different it would be.
I approached the project with a big pen, sketch pad, and a flight booking site loaded up on my screen. I took each element of its flight search form and broke it down into questions - 8 form fields became 10 questions, search fields became typed replies, dropdowns became quick reply buttons.
I found an amazing online chat bot creator tool for Facebook Messenger, called “Chat Fuel”, which I highly recommend if you’d like to experiment but don’t want to venture into coding. I used this to turn my sketched questions into a working chat bot. Then I discovered that there were more options - such as galleries and lists - that I could use to bring inspiration and suggestions into the conversation.
Then it was time to start testing out my bot. So, here’s what I learnt about designing a chatbot…
A typical website form contains multiple questions that need to be answered for a visitor to start or continue their journey. Your chatbot can only ask one question at a time and within that, the response options need to be limited or very simple to type.
Take a look at the flight booking form above, it asks a total of 8 questions. A chatbot will need to break this form into around 8 questions. However, sometimes it’s a good idea to combine questions:
At other times it may make sense to break them down if the chatbot can help to filter the options:
At first impression may seem like 8 questions will amount to a conversation the feels long and tedious, especially when it can be handled in the little flight search widget. However, consider the time it takes for a first time user to familiarise themselves with this form, they need to digest all of its visual cues: which parts are a button, which are form fields, what do the icons mean, where do I need to start, what’s mandatory and which questions are optional etc. This all amounts to a relatively lengthy thought process.
Compare deciphering the flight booking form to having a conversation with a member of staff over webchat or face-to-face. They would walk you through the process step-by-step, going at your pace and following the direction of conversation that makes most sense to you, the customer. The online form can’t do this, but a well-programmed chatbot can.
This theme applies to the majority of digital copywriting, but it’s important to emphasise this point with regards to chatbots, as it can be very tempting to add “fluff” to questions and messages that simply isn’t necessary. On the other hand, I am not suggesting that you sacrifice your bot’s personality. Let me show you some examples:
“Tell me your destination.” could be re-phrased as:
“Where are you flying to?”
“Where do you want to fly to?”
“Where would you like to fly to?”
“Which country would you like to fly to?”
“Which destination would you like to fly to?”
“Which our our 70 destinations would you like to fly to from London Luton airport?”
There are many ways to ask for the same information, make sure that you pick the most straight forward wording, that will prompt your recipient to provide the information you are looking for (e.g. country and/or destination), and that maintain’s your bot’s tone-of-voice.
These are a great way to speed up the time it takes to reply and to avoid responses that your bot can’t recognise. This is especially important when starting a conversation as it is your chance to outline what the bot can and can’t do. Here I have introduced Viva as the bot that can make bookings. She presents the options for what she can book so this informs the recipient that this is the scope of her ability.
Note that it is important to design quick responses either in order of frequency of use or business priority.
Quick responses can be shown in many formats, which I imagine will evolve over time as chatbots get more sophisticated. For the moment you can use things like lists and galleries:
The key with these is to remember that the button label that the user clicks on, will be shown as their reply, e.g.:
This means that button labels such as “Choose” or “Select this flight” don’t work for chat conversations as they become the user’s response:
When I started creating my chatbot, I didn’t program it to use typing bubbles. I jumped on my phone and started a conversation with it, Bing! I received message from Viva. I replied and then Bing! she instantly got back to me. I reacted immediately thinking - “How did she type that quickly?”. This sounds silly as of course I know she’s a bot, but this wasn’t a pondering thought, this was a knee-jerk reaction that I experienced. It occurred to me that throughout all of the conversations I’ve had over Whatsapp, Facebook messenger and other platforms, I have subconsciously learnt all of these little cues and nuances about how online conversations work.
Take a second to imagine a chat with a friend, their little bubbles are waving at you…still waving…still waaaving… you know they’re about to send you an essay, so get yourself prepped to read a paragraph. Versus, their little bubbles, flash up, disappear and there is the one-word reply that you were expecting. We’re talking about fractions of a second here but you should consider the length of the message that your chatbot is going to send and then programme the typing bubbles accordingly.
This is a relatively minor point but it helps to give your recipient the impression that the chatbot is slightly human and not just a cold robot.
When I first created my bot, I designed the following message:
Bot: “When would you like to return from Amsterdam to Luton? You can tell me an exact date, or how many nights you’d like in Amsterdam.” The problem with this was that it flowed onto 5 lines on mobile, contained two instructions and a question, which all added up to quite a lot of information to process in fractions of a second.
So instead I recommend:
This works particularly well if asking the customer a question where you would like to add supporting information. It takes advantage of the typing bubbles as the recipient will have time to read the first message, then wait to send their reply because they can see that the bot is typing a second message.
I strongly believe that they will. Not just because they’re one of the latest technologies to have industry buzz around them, but because they’re so easy to use.
BJ Fogg’s Behavioural Model outlines that behaviour change is caused by a person having 3 things in the same moment: Motivation, Ability and a Trigger. The chat interface is used by people almost all day every day, therefore the ability needed to interact with a chatbot, is already there - the user just needs to be motivated and have a trigger to get started.
The biggest challenge I see for chatbots, is that in order to replicate a service that can truly compete with a website, live chat or phone conversation, it would need to be built as a highly-maintained, new customer service channel. It would need to be fully integrated with customer databases and service standards upheld and monitored, just as with other channels.
It’s exciting to be getting involved in the establishment of a new communication channel, that could sit alongside others including voice and virtual reality. So we’ll see what happens in the not too distant future…