Whether or not you knew it, you’ve likely interfaced with a chatbot, or bot, in the past year. Gartner predicts that “By 2020, the average person will have more conversations with bots than with their spouse.” That may or may not be a lot of chatting (Depending on how much you talk with your spouse), and you’ll see them in every industry. You might talk to a chatbot to get new cable, or to tell the energy company you’re moving, to answer questions about your health, or even to get a new job. Bots are already everywhere, and their presence is only going to increase. However there is a difference between a “chatbot” and a “bot” – what is that difference?
Before we dive in, I want to talk about why this matters.
While chatbots are slowly entering the functional mainstream, generally around customer service opportunities, they have certainly arrived as a hot topic. Much in the same way marketers couldn’t stop talking about Snapchat, Virtual Reality or Augmented Reality when they were introduced, chatbots have absolutely entered the marketer’s lexicon, if not playbook. When the conversation turns to new, cutting-edge marketing tactics, these things aren’t always ready for prime time or right for your audience.
Chatbots, however, as a source of providing information or enriching a customer’s experience, are nearly there. You just need to understand the fundamentals of what’s in play to make it work and apply it, strategically, to a communication structure that serves your audience’s need.
The important thing to understand, and the reason we’re talking about how they work and serve actual functional needs for marketers and user experience strategies, is that, as a solution, this is not a distant dream. As you read this, think about the practical applications. The FAQ, Contact Us or About Us pages on your website. The call-to-action (CTA) to schedule an appointment. The basic search functions. Do you still need to have those pages as long narratives or is the information better served to the audience with a simple chat function? Consider an image campaign that stokes curiosity in the target audience and provides the answers they need in order to understand how your cardiology, smoking cessation or cancer screening campaign can help them. Want one more example? How about a chatbot that tells you stories about the art you’re viewing at the museum, providing a guided walking tour or even just giving you the best path to the finding the Mona Lisa at the Louvre.
Before you throw it out in your next planning meeting, you must understand the problem you’re trying to solve or the opportunity you’re trying to present your audience. Think strategically about it before picking the shiny new object. But, understand that if it’s a viable solution, if there’s a way a chatbot can actually help improve your audience’s experience … it’s doable. It’s software, just like any other computer program. In fact, we built one at Core just because we got curious. More on that later.
For now …
Let’s look at what the definition of a chatbot is:
With that definition in mind, a lot of perceived “chatbots” are actually just “bots”, where they are only there to answer a quick question or give you direct info without much of a conversation. Bots are the things you talk to when you call the cable company before getting redirected to an agent (because the bot doesn’t get it), or the phone call you get in the middle of the day from the number you don’t recognize (seriously, put me on the “do not call list” please!), Bots are … dumb. Chatbots themselves are more complex than that. They present the perception that a real conversation is happening between you and it. A chatbot on a healthcare website might respond to your question by asking you another related question, keeping the conversation going based on your responses, and “feeling” more human. You might even catch yourself saying “thanks” at the end of the conversation if it’s done its job correctly.
True chatbots leverage the latest and greatest in A.I., and take A LOTof data to understand what’s being asked, and to give an appropriate response. This can take a lot of time and money to get right, which is why in our day-to-day we see more bots than we do true chatbots. There are competitions every year for who has the best chatbot. Mitsuku is 2018’s winner, and still has a long way to go before being truly comparable to human conversation. Where does that data come from? It’s helpful to understand that chatbots are built on the backbone of language, and that doesn’t necessarily mean language relevant to your business or customers. It means words, context, structure and so much more.
Fortunately, for you to develop a chatbot as a solution to serve your customers, it doesn’t have to happen from the ground up. When you start, the bot needs to understand that a question is being asked, then it needs to understand what the intent of that question is. Are you asking about cars, or boats or… animals? Then, it needs to be able to take that understanding and give you an answer relating to that question. Well, that’s a lot of things! Fortunately some of this has been figured out using openly available data. The Standford Question Answering Dataset (SQuAD)  has collected questions posed by crowdworkers on a set of Wikipedia articles. That’s a collection of questions, segments of text, that come from corresponding wikipedia passages. That means the answer to those questions is within the wikipedia text. Using this dataset, a bot can be trained to understand questions and find answers given a set of text. That’s square one, the foundation you need to get creative with a chatbot that can serve your customers. Looking back at what a bot needs to do to understand a question, because of the SQuAD dataset, we can knock out at least 2 parts, but we still need the data for it to understand – your data.
So, how do you actually make a good bot that actually engages your consumer, helps them get the info they want, is a pleasant experience and doesn’t break the bank or take months to implement?
Use the data you already have, and let the bot be dumb, for a time.
At Core we implemented a bot that answers common questions that get asked around the office “What is the code to enter the building?”, “What is the holiday policy?”, or “What time is lunch?” – questions that have the same answer. We already have all of those answers, so it was simple enough to have the bot provide the answers it knew to certain questions. You, the inquisitive reader, might be asking yourself “But what about questions it doesn’t know about?”, for that we implemented a system where you’re able to talk back to the bot – you can ask a question and if the bot doesn’t know the answer, it asks if you do. It takes that answer and associates it with that question, so the next time it’s asked, the bot will answer with the previous answer. The bot lives in our public chat channels, so other people can also take a moment to answer the question if they know it. In short, we taught it to get it started and now it’s learning as it goes. The more it learns, the smarter it gets and the better it is at assembling answers based on all kinds of ways people would ask questions.
This leads into the next part about having a bot – maintenance.
Having a bot that doesn’t intuitively understand every single question means you have to maintain the questions it doesn’t know the answer to. So how should you go about maintenance?
Keep an eye on what questions are being asked, and which ones are being asked often. Seeing a bunch of questions that the bot doesn’t have an answer to? Start answering them! Let people get specific, and start to answer those questions. A good bot should also be able to understand the question that’s been asked, and then find questions that might be similar in nature. For example, someone might ask “Where can I find the nearest pharmacy?” – The bot might not know where that user is, but it could infer that they’re looking for a pharmacy based on “nearest pharmacy” and give the user a relevant link while asking if they could be more verbose in their ask by giving an example of a similar question. Ultimately this could lead you into getting A LOT of data, and that gets us into the territory of chatbots from the original definition. As you start to gain all of this data you can start to paint an even better picture of the type of person coming to your website. This data could help lead decisions about what type of content to promote to the front page, or even your next campaign making it easier to target what your audience is actually after.
Some years back, Core was tasked with concepting a campaign for a health system client to generate awareness about a rare, hard to diagnose and painful condition called trigeminal neuralgia. The goal was to help the public understand its symptoms to help people suffering in silence and encourage them to take action to get help with the dedicated clinic at this health system. We explored everything, from big and bold to hyper-targeted.
Fast-forward a number of years … and all you’ve just learned about chatbots:
… with a chatbot, designed to serve their needs and provide value from the health system that was positioned as an expert resource in that community.
When you understand the principles and think about how it can be helpful, instead of a “wouldn’t-it-be-cool” shiny object, that chatbot reality isn’t all that far off. For that pending reality, the proof is in the data of what consumers are looking for .