Last week, I wrote an essay about chatbots growing up to become ominbots, conversational entities that represent brands everywhere. Omnibots are the future of branding. But they have to be crafted carefully to avoid creeping out customers. Here, I want to address the identity, design, and tech decisions you’ll want to make when creating an omnibot.
Clippy was Microsoft’s proto digital assistant from the ’90s. Imagine a cartoon paperclip, possessed by Ned Flanders, that the entire world hated. It is a cautionary tale worth bearing in mind as you sketch your omnibot’s personality, look, and functionality.
Yes, build something charming. Build something personalized. Design something that works in all the channels. Just remember, chipper servility will be the default personality most brands go for. It will become boring. Also while it should be omnipresent, it should not be over-present.
You Better Like The Name
Your customers will be talking to your omnibot constantly. It will certainly become synonymous with your brand, but might go further. Your omnibot’s name could become your de facto brand name. Aside from this, many other factors will influence your choice of name.
It will probably need to be short, as the name will also function as a wake word. You might choose a normal, human name to make people feel comfortable. You might choose a name that is not obviously Western to suit your global markets.
Whatever you go for, just remember, you better like it. It may become more important than you think.
To Embody Or Not To Embody
So your omnibot has a name and a charming personality. Now for the design process. What will it look like? Will it have a face? Will it be photographic or animated? Should your omnibot even be a physical entity at all?
Plenty, like Alexa, manifest solely as a name and a voice. Google has gone a stage further and given Assistant a dynamic logo that is suggestive of a body.
Personally I hate it. Made of shifting Google blobs, it is a sort of underwhelming, digital puddle. It may be that GA’s physical branding is deliberately weak here. Think about how users summon Assistant. They don’t say “Hey Google Assistant,” they say “Hey Google.” Google may just want people to feel they are talking to Google. I suspect the suffix “Assistant” will go away, as Google retrains people to interact with its brand conversationally.
Such subtleties, however, are the luxury of tech titans. Most mortal brands will need to build something memorable. An embodiment of some kind, be it a cloud of light or an illustrated human physique, is a good way to be memorable. A brand may also want its omnibot to appear in advertising, on packaging, and on social, even at public events. Consider every channel you can imagine and how embodiment can and should change across them.
If you’re puzzled, the concept of place-onas is a helpful framework for working out how your bot should manifest depending on where, how, and what hardware customers are using.
Apart from the name, gender is the most obvious identity choice you will have to make for your omnibot. Here again there are many options. Ask Siri what gender it is and Siri will declare it has no gender, like “some species of cacti.”
This is a neat side step, however Siri’s voice is still human and available in the binary genders (male and female) and in different accents and languages.
Early on Alexa and Siri took hits for being female. They are also, to my ear, pretty white and middle class. The U.K. Google Assistant sounds like she is about to bully off for a lovely game of hockey.
Right now, I’m consulting on an omnibot project, and I’ve suggested to the client that letting users choose the omnibot’s identity is crucial. Offering the full rainbow of human identity is a huge challenge, but must be the goal. Unconscious bias in tech is an ugly thing. It probably won’t be as awful as this example, but is best avoided from the start.
Omnibots Are Content, Too
Offering the full rainbow could also provide a point of differentiation. Snapchat’s Bitmoji might be an unlikely beacon here. Bitmoji is not an omnibot. It’s service that allows users to create cartoon avatars of themselves, which interact with Snapchat content. They are regularly updated–just last week thousands of new haircuts were made available.
They also use data and smartphone sensors to adapt to context. For instance, when you are flying your avatar suddenly appears in a little paper airplane. This gives them a kind of living quality. Even an unembodied bot would be more engaging if it were aware of its users situation and adapted entertainingly. If users are dressing their omnibot, if they feel it reflects their life, they might care about it, too.
The Boring Tech Stuff
Aside from all the fancy AI stuff, the core tech here is boring old CRM. The ultimate goal is to build a cloud-based omnibot that remembers every conversation from every channel. Each customer must be tagged and served the right version of your omnibot. Salesforce is making this possible right now.
The Data Goldmine
If you want to rival Amazon, your brand omnibot will sooner or later make the leap onto genuine natural language processing (NLP)-powered systems.
Most small bots run on dialogue trees, simple maps of “call and respond” questions made with software like Chatfuel.
NLP is a form of AI that, according to one vendor, allows brands “to process, analyze, and gain better insights from unstructured data.” Aside from making your omnibot converse more naturally, it can unlock the data potential of bots.
Imagine all you could learn about your brand from a person who talks to millions of people about it.
A Real Voice
People recognize voices, they are as unique as thumbprints. An omnibot should have a natural voice of its own, that’s available in diverse options. The tech is out there to create it, too. Take a look at Google’s Tacotron, and forthcoming products like the “Photoshop for audio” Adobe Voco and startups like Lyrebird.
Omnibots have extraordinary possibilities because you can insert them into every new tech development, making that technology more helpful. To finish let’s look at one example: AR.