You've made great progress in building a Chatbot, Intelligent Virtual Assistant (IVA), IVR or other AI-driven self-service solution.
And you arrive at a point where you can say...
I Finally Understand My Customers
You invested in some Natural Language Understanding (NLU) tooling, fed some of your user-generated data into it, run an auto-cluster algorithm to bucketize your data, gotten your subject-matter experts involved to ensure a proper understanding of what popped out, tweaked the machine learning knobs and levers a bit, and presto!
Now you have a cool intent classifier.
Now, 90% of the time, if your customer says something, types something, tweets something, or whatever, you’ll know what they want.
Congratulations! ... That was the easy part.
What? That Was Easy?
Well then what’s next?
How about building a nifty chatbot on your website to handle all those intents! But wait, there’s 300 intents. Or maybe 3,000 . . . Hmmm, that seems like a lot of work. And actually, a lot of those intents might be handled differently depending on context.
Like if a user types “Why did my payment amount increase this month?”, first we’ll have to check their account details actually, first we’ll have to make sure they’re logged in so we can see their account details! Once that’s done we have to bring them back to where they left off. Then, we’ll have to figure out what payment they’re talking about, check for special offers that expired, check if they placed any unusual orders, and so on and so forth.
Again, a lot of work.
Crafting Chatbot Responses
Next, you have to actually craft responses for all those possibilities with a helpful, friendly persona. But you don’t want to piss off your customers by being too perky, or piss others off by being too casual.
This is the face of your brand, after all.
OK, so you’ve built out some responses with a bunch of conditionals to handle all the cases and sub-cases, created some language to use as a frame to hold the variable content, and now you have designed a somewhat clunky, but functional experience. For one intent.
Let’s hope there aren’t any follow-up questions based on the information you just provided. We all know how well most conversational systems understand contextual complexity!
Me: Computer, what year was George Washington born?
Computer: George Washington was born in 1732.
Me: And what year was his wife born?
Me: <sigh> Computer, and what year was his wife born?
Computer: Sorry, I don’t know that one.
Some customers might not want to use the chatbot though. Heck, maybe their internet connection is out so they can’t access their account at all!
What if they call in instead?
Chatbot vs Phone Call
Wouldn’t it be ideal to leverage all that work you did with your chatbot over the phone, too?
But you can’t present the user with a picture of their bill over the phone, or open up a webpage with helpful information they can read at their leisure. Nope, all of a sudden the interaction is time-bounded to the length of the phone call, and as soon as your phone system says something, it’s gone forever, since it’s sound, not text. Yes, you can offer to repeat things, or tell the caller to go get a pen to write everything down, but that’s a lot of burden on your customer, isn’t it?
So the phone is a completely different animal from a chatbot.
Chatbot vs Mobile vs Alexa
What about a chatbot on a mobile device, with limited screen space?
What about a home assistant, which might be a multimodal interaction including tapping, typing, and talking?
What about a 3 rd-party chat app conversation? Are their security concerns to send user data through non-company servers?
Ugh, that’s a lot to think about beyond figuring out your customers’ needs. You’re probably wistfully recalling how easy that actually was.
So what can you do?
Just like anything else of quality in this world, the best solutions come out of thoughtful, careful design keeping your users’ best interests at heart.
It’s an interplay between tools, techniques, subject matter expertise, design best practices in a variety of channels, and a deep understanding of the business problems you’re trying to solve.
Some developers of tools make claims such as “Design once, deploy anywhere!” which while technically true, may not result in quite what you were hoping for . . .
Three Part Chatbot
In this blog post we looked at the three parts of building a great Chatbot experience:
Understanding the customer. An NLU engine (AKA, intent classifier) provides a way to understand what the customer says, what their intent is and what they are trying to get done.
Doing a task. Act. Each intent has at least one action/response associated with it. Since the task the Chatbot does is based on both intent and context, sometimes one intent can have multiple actions.
Creating an experience. The Chatbot interaction needs to create an engaging conversational experience that is across channels, personalized and branded. The Chatbot needs to be "humanized."
All three parts need to come together to create an automation system that works. To get those parts built and working you've got two choices:
- If you're confident your team has all three parts of the Chatbot covered, then you could I'll do it yourself, or
- If creating a Chatbot that can achieve these goals is daunting, then you can use a managed Chatbot service where experts (like me) take care of all the details that these system present.
My friends, this is not easy work... but we're here to help.