Chatbots are all the rage today and are deployed in all sorts of environments including FaceBook messenger, customer service websites, and voice assistants such as Siri and Google Home.
What most developers and designers focus on early on during the chatbot development cycle is training the bot to handle a wide array of questions and then also training the bot to interpret the questions correctly. For example "Hey Siri, how do I get to my next meeting?" and "Ok Google, give me directions to my 10.30am appointment" need to trigger exactly the same action on the chatbot's part. More precisely, the bot needs to break down the input using Natural Language Processing (NLP) and interpret the meaning and context of the input. For longer, more complex conversations, context needs to be maintained across the back and forth dialog between the bot and the human. This task is complex, often results in a flurry of action and effort from development teams. In the end you get a bot that is functional, well trained and can take orders well and hold a conversation.
However, putting energy in perfecting the bot conversational ability without putting energy into giving the bot a personality is a sure way of making sure the user experience will fall off the cliff because of two reasons.
First, the bot will invariably get things wrong and not understand the user in the first place. Over time the odds of success are greater, however the ability for the bot to handle 'long tail questions' at the start won't be good. How the bot catches itself when it can't interpret things is critical.
Second, the bot at launch will create some hype. It'll draw users in. Bots will a personality will ensure that the users stay longer and have fun with the experience and in turn help create a sense of virality and collect better training data whilst doing it.
At Rush, we're a few bots in and we find the below strategies and tactics useful in ensuring our bots come out with a personality that works:
- We really try to understand the brand that our clients are trying to project along with the target segments of the user base.
- We feed the above research into a personality workshop where we take user conversation scenarios and contexts and come up with some personality options.
- We test the personality with real user base as often as we can
- We try to create friendly error messages, and if it makes sense we use tactics such as humour
- Post launch we ensure effort in the personality development continues.
Somewhat like a toddler, the personality is born with largely but needs to be developed depending on the engagement the bot is having with its user base.
While creating a chatbot, many of us are mostly focused on the topic of a conversation. We think about which questions should be asked, when, and what answers make sense. Many companies forget to tailor the scenario to the specific situation with the specific customer.