The basics of chatbots for business.
How chatbots can help us and how they can’t just yet.
With a focus on small business enablement, chatbots have been my focus over the last year.
There continues to be excitement about chatbots and the space continues to evolve despite speculation and romanticization around NLP (natural language processing) and AI (artificial intelligence). These things sound fancy but it has yet to be proven whether they carry the core value of the category or if the value lies somewhere else. At the end of this article, I’ll let you decide.
In my work on chatbots, I had the pleasure of working with Adam Geitgey whom I highly recommend reading Medium, to figure out the limits on what our bot could and could not accomplish today. I’ll share a few of these learnings.
First, and most pertinent: I bet you’re interested to know whether customer service will be possible with chatbots. Well here’s the thing… chatbots are currently capable of handling short bits of text. Like, “what is your name?” but it’s a little harder to handle long text like, “did your friend ask my mom, ‘what is your name?’ and do you know if and what she answered?” Huh!?
The first question has a specific question to specific question.
The second question has multiple turns and is way harder to process.
The fact is that most customer service requests fall into the second category: long text, multiple turns. Good for human brains. Bad for bot brains.
Another factor here is how machines communicate. They can retrieve templated responses — like we humans sometimes tend to do . Think FAQ or canned responses we may use to address to common customer questions.
The other kind is where we come up with a non-templated response on the fly. We — as humans — would likely synthesize the response based on our knowledge and the incoming question and its nuances.
If I was going to make a customer service chatbot today, I’d probably make it amusing, dumb and retrieval based. Dumb can actually be a plus because the degree of frustration a customer experiences is usually equal to their expectation of getting a particular result and not getting it. A dumb bot sets expectations, has fun, and can tackle a narrow but meaningful set of inbound requests.
The hard stuff
Domain
Bots need context. They don’t do well with endless possibilities. They’re terrible at conversations that can go in a whole heck of lot of directions. The narrower the domain, the better the bot can perform. There are ways of dealing with this where the bot itself can help the user narrow the conversation by taking control of is with buttons.
Context
Machines are lousy at it. We humans are good at it.
We can derive linguistic context — the tone, the style, formality, sarcasm, culture and more — intuitively. Some machines can do this. Most can’t.
We can physical context: location, weather, gender, environment without even thinking. Machines can’t. These factors are the little points where us humans can make connections and can even play off one another. It’s very hard to generate positive results with automation in this area when it comes to your company’s NPS score. Don’t say I didn’t warn you. Let’s check back in a year or two.
Keeping a coherent vibe
Personality is no longer bound up in pretty colors, amusing icons and branded copy. It’s something that we’ll need to start considering how to stuff into the mediums of text and voice. Branding takes on a lot more to its plate and design challenge becomes more, not less.
Intent
Machines that get intent right are great. Neural networks like Google’s Inbox reply-buttons do this well, but you need a lot of data. Intention is a little easier than context and this is largely possible today however it’s still too early to ask a bot to classify our dusty high school papers or hire some summer interns. The bot needs context data which in itself is nearly impossible to normalize.
But here’s what we can do without getting into deep learning or hardcore neural networks.
What we can do today.
Chatbots as sidekicks
Chatbots are good sidekicks because they can listen and they can share simple and clear bits of information. They’re a little like a well trained puppy. You can call it, it’ll come. You can give it a simple directive and it’ll deliver a correct result. I can’t ask my puppy to make my coffee but I can ask my puppy to go fetch a stick. This is your starting point, and you can built from here.
Chatbots as educators
Dropping the puppy analogy, chatbots can fetch all sorts of information. That means your bot can insert the relevant content into communication channels at the right time for the right people. It’s helpful when you’re broadcasting ideas and content. You can even invite reactions and engagement, which can be powerful.
Chatbots as mini-games
All good education is engaging. The beauty of chatbots is that they can be playful. They can be silly. They can ask questions and take answers. For a small business wanting to do contextual surveys, branding or lifestyle campaigns a chatbot is a great idea. Sensitivity is due of course because customers largely respond negatively to persistent pings from bots so proceeding with caution is in order.
Chatbots as providers of success metrics
If you want to know about something and you don’t feel like visiting a web location or opening a mobile app, a chatbot is the fastest way to acquire the information you need. One of the most successful chatbots today (Statsbot) has gained significant traction and responds when you ask it to go fetch your website metrics. Metrics are typically a little annoying to sift through so customers enjoy the ease of asking for only what they want and avoiding lines, charts and bars.
Great with YES and NO for services
A lot of people think chatbots are all about natural language processing when in actual fact a simple yes or no can do. After all, most of the software we use depends merely on a click or a drag or an upload. When thinking about designing a chatbot, skip the conversation and go straight to the bottom line… buttons. Click!
Compensate for dumb chatbots with superb text
When designing chatbot workflows, the hardest thing I’ve discovered is designing compelling tutorials and guidance. Not just that, but keeping copy consistent across the bot. Every new feature, every update and tweak requires at least some text to be changed and you always run the risk of a loosely branded lexicon. Our brand and our bot’s personality live in its linguistics and no longer in the visual realm of colors and shapes.
I hope you liked this article, and please give it a fave below.
