This article is part of a series. Be sure to check out all episodes here:
Already read these? Super. Let's dive right in then.
In episode 5 of this series, we covered step 1 of our discovery phase in bot development (how to define for whom to build a bot).
Let's repeat the steps here quickly:
- For whom are we building a bot? (link to episode 5)
- Who is our bot?
- How do we define what our bot does?
- How do we test and evaluate our bot?
In this episode we'll cover how to define who your bot is.
Who is your bot?
Whilst trying out a bunch of bots, you'll frequently run into bots that are cold hard methodic machines. Others tend to be airheads, and then there's some stuff in the middle. I've seen some great proof-of-concept video's that are promising, but no real awesomeness yet. Sometimes you'll find bots where you're awestruck of how natural and human they interact. That is, until you figure out they weren't "bots" at all but you were handed over to a real human from the first sentence you uttered, since it didn't match up with whatever the bot needed to cover. It would be nice to let me know guys. Really. Having to ask "Are you a human?" in thread 9 after having banged your head on the desk because you were so jealous of how well this bot was behaving, isn't exactly fun.
So there is a real challenge here. In fact, we believe one of the hardest challenges in bot development isn't necessarily technical but rather UX related.
One of those UX challenges is defining your bot personality, which turns out to be tricky. What is it's personality? How does it walk and talk? Is it in line with your brand and what people expect of your brand? Or should we diverge, since we're presenting ourselves in a different channel? Where is the middle ground between it's 'bot-ness' and the ability to have a normal human conversation without landing in the uncanny valley where it just gets people confused?
Fascinating stuff. So let's try out how we define a "bot" personality.
Before we start writing out any dialogue for our bot, we need to define it's personality. Because in the end, any dialogue should adhere to this personality.
So how do we go about this? Turns out, there's a methodology for this.
You can follow along if you like. I've created a template for you to use in combination with the steps below.
Step 1: define personality traits
As it turns out, one of my collegues has a degree in psychology. Handy. So we had a chat about this subject. Turns out, there are quite some personality traits. Don't believe me? Try this one on for size: http://ideonomy.mit.edu/essays/traits.html
Yeah. I know.
Then I included our strategy team in the discussion. I knew these guys had good results with, for example, our client Melexis in a "brandhouse" project.
They have methodologies for defining a "brand hero" based on brand archetypes (like this). This isn't exactly a one on one with defining a bot personality, but whatever personality you end up with should be in line with the brand, and it was good input for the next exercise.
In the end we gathered a bunch of persona's (real and fictive people) with a strong personality and we started a personality scaling exercise.
Step 2: scaling the personalities
Once you have a whole bunch of persona's you quickly go through a scaling exercise.
Choose a bunch of personality traits you think are important for your bot, and map the persona's on a personality scale.
Decide collectively where you'd like your bot personality to land on the scale, based on the insights you collected from your previous discovery phases (your brand hero, target audiences, bot biotope,...) You could research and finetune this decision into oblivion here, but for now, just keep this exercice fluid and moving.
One such scale would look like this:
Now iterate this exercise across a multitude of relevant personality scales, repeating the proces, and put the end result in the template provided above.
Step 3: visualize the personality
Once you have your scales. All you have to do is map out the personality on a spidergraph. You might need to define multiple personality graphs, based on multiple channels and audiences.
The end result can look like this:
As an example I pitched the Carrot Fit app (http://www.meetcarrot.com/fit/ not a bot, I know) to a generic fictive fitness app. As you can see, you could have 2 totally separate personalities for a bot with the same function: getting you fit. Personally, I'm much more motivated by the Carrot app then by the generic serious "coaching" type of app. One more reason to have a good understanding of your audience and what motivates them.
Step 4: Tone of voice
Defining the tone of voice is a more straightforward exercise any good copywriter can accomplish. They can learn a lot about all of the material you created previously (user journey map, empathy map, the personality graphs,...) but they themselves can gather additional datapoints that improves the process of defining the correct tone of voice. A copywriter will generally (within the context of previous gathered material):
- Talk with your brand representatives to get to know more about the brand and it's values. You'll be surprised how people, even within your own company, will have a different opinion on what they think your brand stands for.
- Gather examples of communication your brand has had so far, and do this over different channels. How do you address people on your site, on social channels, in printed media,...?
- Try to detect style and colloquials used throughout the communication tree. "Thank you very much for your time" vs "Thx a bunch" is quite different as you can see.
- Scour interactions your brand had with "real people" and learn from those. See if your support desk has communication logs, or recordings. Hunt down your CRM people and see what the "input" is they get from different channels (think entries of your site's "contact form" for example).
- Go through keyword research to detect how the outside world is communicating around the topics/subjects the bot will be handeling.
So it goes without saying that a copywriter is a vital piece of the puzzle if you want to venture out into bot development. Go get one of those and include them in the process! They're vital for the next phase too (see next article in this series) when we start defining what our bot will be doing.
step 5: name the bot
Don't forget to define what name your bot will go by. Naming your bot is not only important for further communication and promotion of the bot, but it helps development. It avoids everyone adressing it as "the bot" which is cold and without personality. This will creep in the project. Naming the bot will automatically get a higher level of involvement by everyone and if chosen correctly, will help every implementation of the bot to contain more personality.
With audio assistants this name will become even more important. Often you'll use the name as "trigger" word, which will in turn activate your bot.
A short example. This whole article series is a written down version of an interactive presentation we created for our customers to introduce them to the world of bots and bot creation. We didn't want to show them some airy presentation about bots, but we know that showing actual working bots and the "behind the scenes" of the bots is much more interesting for an audience. So we decided to incorporate an actual bot in the presentation that takes control of certain parts in the presentation and acts as a sidekick.
The presentation was called "Mobile Lunch" since we covered not only bots but a future roadmap of mobile as well, and it was in combo with a lunch, so we called the bot simply "lunchbot".
Lunchbot is created on API.AI and works on Google Home. So during the presentation I could demonstrate certain aspects by simply asking Lunchbot some stuff, and it would respond. To be able to do this, you need to ask Google Home to 'trigger' the bot so you can start using it. The conversation would go like this:
- Me: "Hey Google, Let me talk to Lunchbot"
- Google: "Sure, here's Lunchbot"
- Lunchbot: "Hi, I'm lunchbot! You in a pickle?"
- Me: "Could you explain to the audience what the difference is between AI and bots"
And it would start explaining, whilst I just toggle the accompanying slides.
So you'll need a short name that is quite distinct, especially for speech.
Step 6: visualize the bot character
This step is kind of optional, though it can help to put a "face" on your bot to further help you in developing the bot.
At some point your bot will have a physical representation. It might be in your promotional campaign for the bot, it might be as "avatar" in messenger or your website, or a promotional video introducing the bot. Chances are you'll have to visualize it.
You can go different routes here. We had a cartoonist on staff, but any designer with experience in character design can help out.
For our Reference bot I just had an open chat with the designer and I described the bot and its personality to him. I had a rough mockup of how I saw our bot evolving through different biotopes.
Stunning. I know.
After some iterations the first drawings came through, and we finalized them by pouring it all in vectors. Check out the gif below to see the final result. Mind you, we didn't go all the way here since this bot is just a learning project. Ideally you have your character created in full 3D so you can render it in any pose you need.
The deliverables from the previous article (empathy mapping, user journey map, insights,...) together with those from this article (personality graph, tone of voice, visual character) will give you a strong foothold along every next step of the development of your bot.
So now it's time to define what our bot actually will be doing. We'll cover this in the next article of this series.
This article is part of the Bot Chronicles series, a series documenting the Bot R&D of The Reference. Read up on previous episodes here: