Every researcher knows that the hardest part of writing a paper is making a killer introduction. Chatbots are no exception to this rule: everything is in the first sentence they say. In around 150 characters, their destiny is determined. A good opening will create a beautiful bond between the user and the chatbot, whereas a bad opening will leave the chatbot in a dark lonely place after only 5 seconds of conversation…
So what is it about the opening sentence? Why is it so crucial for the success of a chatbot? Well, just as it is with humans, first impressions are critical. To put it more scientifically, the opening defines the two dimensions of the conversation: width and depth. So let’s discuss these two concepts and learn how to design great conversations!
Width: keep it narrow
The first dimension of the conversation is its width, or what the chatbot knows, can discuss and handle. Developers usually refer to it as the “functional coverage” of the chatbot. In this era of Artificial Narrow Intelligence, where machines know how to handle specific tasks and solve specific problems, keeping the width of the conversation narrow helps avoiding user frustration. From the very first sentence, as the chatbot engages the user, it should clearly state what it can to do, and more importantly, what it can’t.
A chatbot saying “I’m here to help you understand your last cell phone invoice” is aiming too high, or rather too wide. It invites a very large spectrum of questions from the user. The probability of getting a query the bot can’t handle early on in the conversation is rather high. This is likely to frustrate users and prompt them to abandon the conversation quickly.
A chatbot saying “I can help you pay your last cell phone bill” is closing down the conversation width quite drastically. Nevertheless, it is stating the rules of the game very clearly and reduces the probability for a question it may not be able to handle.
Imagine, however, that at a later stage of the conversation, the user asks the chatbot something that goes beyond the declared scope, such as “could you remind me how much the last invoice was?”. If the chatbot knows how to handle such a demand, the user will have a pleasant surprise. Everyone is happy.
If, however, the chatbot does not understand such a request, the user won’t be overly frustrated. After all, the bot is primarily here to pay invoices, not to display past ones…
Length: respect the Reward Principle
The second dimension of the conversation is its length. Once again, the idea is to keep things short. Users are impatient, therefore your bot should help them quickly get to their goal. Additionally, in long conversations, the probability of diverging and expressing an idea the chatbot doesn’t understand increases.
Easy to say, but long conversations are sometimes necessary! What if a lot of information is to be collected to deliver what the user wants? This is where the Reward Principle saves the day!
Consider the following example, a chatbot that helps you calculate your mortgage entitlement. Quite useful! To provide a good service, the bot needs to gather around six different facts about the user’s real-estate project to do the math. This takes time. The Reward Principle states that the chatbot should give the user a “reward”, an added-value, after every 2-3 questions asked to keep the user’s interest. A reward is a piece of information that has value to the user at this point in the conversation. So even if he cuts the conversation short, the user leaves with something useful.
In our example, a good implementation of the Reward Principle will cut the conversation into two parts.
After three questions, the chatbot would give the user a range of sums of money, but not the exact amount he could get from the bank. The chatbot will then ask to go through another three questions to calculate the exact sum the user could borrow.
Even if the user doesn’t play along and quits, he leaves with an idea of how much his mortgage could be. Not that bad! 😉
Tip: doing something for the user is better than simply informing them
We are in a fast-paced world. People try to be as a efficient as possible. Chatbots should help!
Making a chatbot that helps you buy your train ticket is better than one that simply shows you the train timetable.
A chatbot that helps you return a product is better than one that simply goes through a goalless conversation about the store’s returns policy.
A chatbot that goes all the way to ordering takeaway food is better than one that simply displays the menu and today’s specials.
Building chatbots that deliver is better, so no more FAQ chatbots! 😉
Also published on Medium.