Making Your Chatbot Personal

One of the first things I wanted to do when I started to build my first chatbot was to personalise it.

To be honest, this was before I had even decided on the purpose of the chatbot, because my main objective was to figure how AI chatbots work, and what I could get them to do.

I thought a good place to start would be to greet the user and ask their name, but then it became apparent that the responses could vary wildly.

Let’s be honest, in many cases personalising a chatbot isn’t essential. It could be considered a nice touch, but not something which should prevent the user from proceeding. However, in some cases you might need to capture the user’s name along with certain pieces of other information they’ll be disclosing during the conversation with the chatbot.

In my case I wanted to personalise it, but make it optional from the user’s perspective. It was then that I realised that I’d need to handle the range of responses from the user in a number of ways.

I could (and often do) begin with a friendly greeting and question such as Hi there. I'm Carrera from Targa. Would you be happy to give me your first name?

What kind of responses to that question could I expect?

• yes
• sure
• no
• absolutely not
• John (or whatever the user's name is)
• no way
• no problem
• or a ton of other possible responses.

If the uses provides their name… for example Rachel ... we could then assign their name to a variable or entity, which we could call FirstName for example. Then the chatbot could respond with Hi Rachel, how can I help?

Simple enough, right?

That seems simple enough, but what if the user said sure or no”?

It would make no sense for us to assign the words the sure or no to the FirstName variable, otherwise the chatbot is likely to respond with Hi sure, how can I help? or Hi no, how can I help?

That, of course, would look crazy. So then what?

There are two main directions we could take here, based on either a positive or negative response from the user.

A positive response
The response sure (or any other word or phrase with a similar meaning, such as yes of course or no problem) could be taken as the user’s agreement to provide their name. But in this instance they haven't provided it yet, so you’ll need to ask for it directly, and then make use of it during the subsequent conversation with your chatbot.

A negative response
The response no (or any other word or phrase with a similar meaning, such as no way absolutely not or none of your business) could still allow you to proceed with the conversation, but without any personalisation.

So far we have anticipated the user responding to the opening question by providing their name, or either agreeing or declining to provide their name.

But what if the user responded with something entirely different, such as What are your delivery costs? or What are your hours of business on Saturday?

We now have to take an entirely different direction…

Those questions could be answered directly without any personalisation, based on the information you have applied to your chatbots knowledge base (a resource containing information about your business, products, services, FAQs, etc), or you could handle them as part of a separate path or flow.

I mainly use Voiceflow for this type of chatbot. Other platforms are available, of course, but Voiceflow provides the flexibility I need to maintain control, in the ways described here, including identifying and handling user intent.

But if the user says something like The service you guys provided was great or I want to complain about the rudeness of one of your staff, your chatbot could be set up to listen for the user's intent and take them down a specific path or flow which will allow you to control the conversation in a way appropriate to the user’s response.

Listening for their intention

When listening for the user's intent in the above two examples, this could mean detecting or identifying the opportunity to ask for a review, or allow the user to submit a complaint.

Both can be handled with or without personalisation at the user's discretion, as can an infinite number of other responses, intents, questions or request. You just need to know how to anticipate and handle them, how to capture the information they’re providing, and ask if the user would be willing to provide additional information for the purpose of getting back to them afterwards.

In fact, any capturing of personal information needs to be handled in compliance with privacy laws such as GDPR, and of course obtain the user’s approval before doing so.

So the seemingly simple matter of asking a user's name might need to be handled in a number of different ways, and that will entirely depend on the user’s response, and how the chatbot has been set up to detect, interpret and decide on what to do next.

Today's chatbots can handle most user input in a conversational manner, but they need to be trained on data, intents, and allowed a certain but controllable amount of flexibility in their ability to understand, interpret and respond.