How Human Should Your Chatbot Be?
Your Chatbot can range from uncanny valley robotic to a totally annoying blabbermouth. How do you choose its voice?
The Chatbot Human Coefficient (CHC)
Let’s create a means of measuring this conundrum, and also because a nice slice of quality jargon is really called for here to make me look good. I’m calling it the Chatbot Human Coefficient (CHC), because it sounds scientific.
The CHC is the proportion of a Bot’s interactions that are ‘humanized’, rather than delivered in purely transactional and utilitarian language. This is expressed as a numeral between 1 and 10, from a total of 10. For example, a bot with a CHC of 4 utilizes humanized and possibly personalized responses 1 out of every 2.5 interactions. A bot with a CHC of 0 never uses anything but basic and usually generated transactional responses. The higher the CHC, the more ‘voice and personality’ is built into your bot. And a bot with a CHC of 10 is totally scripted and probably pretty linear.
For more details on the CHC connect with the Chatbot Copywriter, which would be me. Here’s a not totally silly diagrammatic view of the CHC.
Okay, so let’s figure out how you choose where your bot’s CHC should lie.
Pay attention to your vertical.
A gag-a-minute, winking, stand-up comic sort of a bot is probably not a good idea if you’re crunching long-winded business data. A statistics bot should be kind of nerdy, numerical and well, bot-like. You’re looking at a chatbot human coefficient (CHC) of around 1. Occasionally, your bot will show signs of life, like an accountant does occasionally, mostly at sign-on and sign-off, but for the most part it’ll be all business all the time.
Likewise, if you’re an eCommerce Fashion bot, you might want to think about having your bot channel its inner Zoolander just a little more than the average bot. Its speech patterns and responses (being a little bitchy is probably okay at times) should match what its doing — helping you find something fashionable in an online store. And let’s face it, nobody would trust a robotic nerd with tape holding her/his glasses together to do that. Your CHC is probably about 4 or 5. You still want your Zoolander character to be able to handle the cash register, but not be able to fix it.
Finally, at the other end of the spectrum would be something like a Storytelling Alexa Skill or Role Playing Game Skill. Clearly, here, personality is critical to getting the user engaged in the story. In fact, more often than not the voice here will be ‘in character’. In this kind of application, breaking character can drain the drama and is a place where users can be lost, so limit the transactional robot responses to a minimum. You’re looking at a CHC of around 8 or 9.
How much of your bot or Skill is scripted?
Natural Language Processing (NLP) is nowhere near ready to engineer robotic responses that recognize human nuance, subtext, or all but the most basic context, so for the most part, the humanity is going to be scripted. In other words, the more your bot or Skill needs a higher CHC, the more scripted it will be.
Most bots are going to be hybrids at this point, so finding the right mix is crucial. For example, if your bot is a budgeting app like http://abe.ai you’ll want to pepper in the human touch to what is otherwise a set of transactional responses, because the humanity bit will keep users engaged. Say the data shows that you’re over budget for that month, there are many ways to design the dialogue in this context. You can state simply that the user is over-budget, or add straightforwardly that they should consider easing off next month, or trigger a scripted response with the information that perhaps gently chides the user to watch out, or maybe even a note to spend less on sneakers, or lunch for the rest of the month.
A Little Wit Goes A Long Way
Humans come in many shapes and sizes. Some are the life and soul of the party, others are more introverted, other unfortunates are just plain boring. You don’t want your bot to ‘offend’, but you also don’t want it to be so ‘nice’ and anodyne that it’s utterly forgettable. This is going to be a key decision. Big brands are likely to be very risk averse to having a bot which pushes back a little — even with a wink and a smile on its gender-free face. My guess is that a bot needs to take a few chances to be successful. And the truth is that everything it says can be analyzed, so no fair no foul. If a response is too ‘risky’, you’ll see drop-off in the logs and can respond accordingly. But remember, to use a wonderful old cliche, nothing ventured, nothing gained.
Get Advice From A Real Live Human
Sitting around a table with your team and white-boarding ‘who’ you want your bot to be should really help you answer some of the key questions about how much humanity is the right amount for your bot. If you’d like an outside consultant to join that roundtable or to answer questions on this key subject, I’m around for you. Just give me a shout. By the way my CHC is 10. And my natural language processing skills are as good as they get.
