As we accelerate towards a more automated future, conversational applications are getting smarter. They’re taking on more challenging use cases, and playing an increasingly profound role in how organisations private or public, can be more in tune with the communities and customers they serve. The competition for your attention is heating up, with retention, session length, and organic growth of the user base, some of the key battleground metrics. For example, user retention is a key metric because it “more users are coming back to your bot for more” and that’s a good thing. One reason could be sees things the way you do, understands your concerns, and makes you tell the next person you see….
choose your platform
The first decision you need to make is the platform you think will serve your humanising work best. I’m taking a much closer look at Rasa, a fast growing Conversational AI platform, with excellent AI, Natural Language Understanding, and a very supportive team and development community. Rasa is fast becoming a leader in the field of conversational AI.
Coded from the command line, RASA’s approach is to push a basic bot out to testers ASAP, then begin cycle after cycle of fast-moving ‘Conversation Data Development’ making adjustments to the NLU and various components and then testing again. At RASA, “the happy path is a myth”, because in real life users never follow them. Wow. I was kind of spooked, until I studied the docs. And then it all started to make sense. RasaCore and Dialogue Elements offer countless creative ways to establish consistent character/dialogue throughout the chatbot. Who needs nodes?
The ability to work with the end-to-end conversations is a golden opportunity to get the character really connecting with users on a profound level. The best way to make Rasa work as as a humanising platform is to separate out a writer and data analyst team (WDA) from the core, which will create the generic bot.
The WDA would drive the humanisation element of the project via three main documents;
A Bot Character ‘Bible’ (BCB) written by the creative writer, which clearly describes the backstory, personality traits, and dialogue style for the humanised bot.
A Creative Input Template (CIT) written for the writer with a unique ID for each dialogue input into Rasa Core and dialogue elements.
Finally, a Happy Path Playbook (HPP) of the best end-to end conversations to aim for during the Conversation Data Development Process.
How to get the dialogue skills you need
My cooking is terrible, but even I know the end product depends on the quality and proportions of the ingredients you use. And for a humanised chatbot the number one, non-negotiable must-have is.
HIGH quality dialogue
It really doesn’t have to be Shakespeare, but at least one member of your team MUST HAVE the ability to consistently and quickly write meaningful, entertaining, near-professional dialogue, which clearly expresses human personality traits as it fulfills the app’s use case.
Good dialogue writing is part of a TV writer or screenwriter’s standard skill-set, but a rarity among Conversation Designers and Copywriters. And with Humanisation just taking off, the pool of converted screenwriters or people starting out in the field is very thin. The job also requires a person who’s got “left-brain and right-brain skills”, because the dialogue is specific to its automated medium. A superb dialogue writer zero interest in tech, will fail just as spectacularly as handing the dialogue ‘chore’ to the junior devs.
THE RIGHT AMOUNT OF HUMAN
Now you’ve got your writer(s), to write humanised dialogue, the next question is how much human do you need?
How much human? I know it sounds odd, but there’s a very simple, and easy way to get close to the ‘optimum level of humanisation’ for your bot project based on its subject matter, sector or domain, using the Bot-Human Dialogue Index (BHDI).
THE BOT-HUMAN DIALOGUE INDEX
Here’s the Explainer GIF (Full Disclosure: I created the BHDI).
The BHDI enables us to confidently state a recognizable number as an answer to the seemingly imponderable, “how human do you want your app’s machine conversation to be?” Without anything to measure against, the question would seem unanswerable. But the BHDI adds two variables to finding the right amount of human as easy as pie.
setting a start point for discussion.
Both the variables are measured against subjective, broadly-held, non-contentious social assumptions. The first variable sets and describes the end points of the Bot Humanisation Spectrum as “Monotone Robotic to Extreme Extroversion”. The second variable establishes different subjective, broadly-held, non-contentious social assumptions, this time, giving an assessment of the optimal level of humanisation for an app in a specific domain or sector.
to err is human - bots like us.
BHDI also accounts for the “competence” that users would expect. A pure robot with a BHDI of 0 would be expected to make no mistakes and to have “high competence:”, which Stanford researchers actually saw as a turn-off for users in their test groups, who gravitated towards “low competence” bots that make mistakes – just like us. In terms of our test bots, it wouldn’t be a stretch to say that the Fashion Startup Convo App (“RoboDude”) might be more forgetful more often than say a by-the-book Government Convo App, who only makes the odd mistake, and might be rather embarrassed about it too.
What's your project's BHDI?
Ok, now think of your own project in these terms and give it a BHDI based on the various criteria above. Have your entire team make a similar assessment, and the range should be tight enough to give you a strong starting point for the other parts of the personality build.
building bot personalities
A Conversational App’s personality traits are expressed through its dialogue. The two are inextricably linked. Personality traits in conversational apps can also be seen as deviations from the generic. A generic personality is easily to find. Just try out a new machine conversation and then the next day see if you can recall anything distinctive about its personality. If you can’t think of anything, then the app you used the previous day almost certainly has a neutral or generic personality.
MATCHING PERSONALITY TO BOT DOMAIN
pick the pair of personality traits that matches
each CONVO-APP best.
This one requires a more granular take, but is otherwise an extension of the BHDI process, using the same mind-muscles. If you’re a developer or engineer and you nail them after a single view of the 10s video image, we need to talk.
A great avatar
A well-designed avatar with plenty of personality and character adds important sensory depth to the bot humanisation process, reinforcing the bot’s believability both to new and regular users. As an added bonus, digital characters are very strong drivers of brand retention over time. Here are some tips.
If you’ve made the right choices, and your writing is strong, your users. could gradually put aside that they’re talking to a computer program, and believe in your conversational app to the point where it can be seen as dependable friend, ally, guide or companion. Reaching the dizzy heights of true believability isn’t easy but it’s definitely worth it.
If you’re putting together a humanisation project using Rasa, and you need some input on any aspect of it, from team organisation to dialogue input, schedule a conversation below.
Humanising bots is a new and miniscule field, so please feel free to comment and ask questions on the Linkedin posting of this article if you want to learn more about any aspect of it.