Imagine opening Facebook Messenger and writing a restaurant to make a reservation, then messaging Uber to schedule transportation, and then instructing Facebook M to tell your friends the wheres and whens of the meet-up. This future vision is a reality that marketers, start-ups, and media tech giants are scrambling to create for consumers. Dubbed as “Conversational UI and chatbots,” this topic dominated the programming and party banter at SXSW 2016.
Hailed as the “end of apps,” conversational user interface and chatbots are newer iterations of technology to “input” our intentions and information into software and technology. Specifically, conversational UI refers to the ability to enter in a string of text (or in some cases, using our voices to “tell” software what we want it to do) and chatbots are robots who are answering our questions and bantering with you (if they are smart enough) as they help us resolve our customer support issues, negotiate a reservation time that works for everyone, and helping provide the right information to the right folks to make a night out an easy affair.
Powering all of this is a layer of artificial intelligence focuses on natural language processing – a specific method of communicating with an intelligent system using the natural ways people speak (such as English). This layer is what makes chatbots so powerful for marketers. Now, instead of creating apps or services that are meant to be useful for the majority, there’s now a method of interaction, a way to give information and get information, that personalizes the experience. This also opens adoption of services wide open. Rather than being forced to search for answers, we can now ask for more than just information and commerce from our digital lives – we can begin to ask for real service.
For marketers, that’s a powerful way to build brand loyalty.
What will it take for us to deliver real service? What will it take to realize the power of chatbots the way culture has realized the power of the mobile device? How does it make the marketers job easier to create loyalty and market demand?
TYPES OF BOTS
There’s the traditional messenger bot that simply replaces the traditional user interfaces of “dropdowns” and “entry fields” and buttons with the ability to ask a question and receive an answer. These bots are trained to answer specific questions, but might have difficult complex inquiries. For example, say you would like to a pair of shoes you bought from Retailer A (let’s call them Appos.) You might ask a range of simple questions like “What’s the returns mailing address” or “What’s the return policy?” or even commands like “Please send me a return label” and the messenger bot will comply. The cases it doesn’t understand you, it can direct you to a human being. Here, it replaces low-level tasks and essentially deliver to you the information you’ve been looking for and if it doesn’t, it provides an interface for you to ask for a human being. Its purely about making it easier to connect.
This type of Bot will soon become an off-the-shelf software product that will require minimal configuration by developers and a new brand of “editorial” staff (likely from the marketing team) to help train the bot with the right responses. In some cases, there will be a need to train the bot with hundreds or thousands of unique questions in order to cover nearly all the right responses. In these cases, there are many platforms that offer a global network of individuals willing to answer these questions via Amazon’s Mechanical Turk or Upwork.
The reality is, our expectations as consumers raise as the interface gets smarter. For example, ASOS’s bot recently got in trouble for not being able to provide an outlet to the consumer to get the proper service after it struggled with being able to answer the customer’s question. The bot even goes so far as the chastise the consumer for language he used to express his frustration. This simply compounds the problem rather than providing real service, which is to meet the consumer and helping to de-escalate. That type of empathy is what we expect from person-to-person interactions.
We’ve been talking about assistant bots since Siri first made her appearance. Factor in Microsoft’s Cortana (who technically came out first but didn’t receive the marketing push Siri received), Google Now, Facebook M, there are groups with LARGE amounts of data about people to start this movement.
But, as we saw in the latest round up by the New York Times, the quality of these is largely determined by the kind of data it receives and how people are using it.
What these bots do well is help interpret the question and direct you to the right low hanging fruit. What these bots are meant to do is to really act like an assistant and do tasks for you. To do that, you have to train them. But, where does that lead us? The truth of human behavior is that while we want our assistants to do the tasks, what we really want is empathy first so that assistants understand the context of why we make the decisions we make. If I keep telling Amy from X.AI that I want to meet in the office and she keeps asking me “which office, what’s the address of the office” – she’s not doing her job. But, if Amy makes a mistake but doesn’t own up to it, that’s when her perceived value as an assistant goes down. Then, she does become a simple piece of software in my mind. In these cases, would I, as a consumer, use her less because she was missing the element of empathy? Can I hold space in my mind for the fact that she’s human-like because I can talk to her LIKE a human, but that she is not, in fact, able to understand my context?
By now, many of us have heard the story of Microsoft failed attempt at an entertainment bot named Tay in the US. If you haven’t, the short of it is here (and the long of it is here at Quartz):
Launched the previous year in Japan to great success, Tay’s Japanese equivalent, Rinna, was a bot launched on Twitter to engage and delight consumers with stories and conversation. In China, another bot XiaoIce was launched to 40 million users released on WeChat, China’s equivalent of WhatsApp, to similar success. However, when launched in the US, it was taken down after 24 hours because of a loophole in its learning algorithm that allowed the Twitter-verse to train Tay to be less than friendly or entertaining, instead issuing offensive remarks and comments
The natural response of marketers here is to want to limit the responses only to pre-canned text or images. But that limits the possibilities that AI offers us. We can train artificial intelligence to be useful as long as we are clear about the logic by which it constructs responses.
Specifically for entertainment bots, at least in American culture, what makes them entertaining is not so much to real-ness of the responses, but that the responses are funny in the context we understand. This is, for us, the most exciting part of developing entertainment and content on behalf of our brand clients. It gives us the opportunity to work with comedians, linguists, robotics folks, and brand leaders. The considerations here are often how to orchestrate the type of interactions where we are willing to suspend our disbelief. Right now, we go to theatre houses, comedy clubs, music festivals and live experiences for that type of spontaneous interactions. It is these types of human experiences that inform how we program these entertainment bots. Stay tuned for a few of our co-creations as we release them with our clients.
In all of this, there are a few things to consider when commissioning these types of digital experiences.
- Data: As you can see, there is a lot of training data that goes into teaching bots of all kinds. How your agency (or internal technical and marketing teams) go about the problem is an important consideration. Also, we need to think of training data as not simply a one-time thing. Like any learning process, bots will eventually need to be trained to understand different types of contexts. That means that data inputs will need to be constant and the models of behavior its training will also shift over time. Like development of customer service training, the same must happen for bots.
- User pathways: Also, what are the appropriate outputs beyond the tech responses. How do you provide a way for consumers to get connected to someone who can help de-escalate and get on the right track?
- Aligning with Brand Purpose: At the end of the day, its important to make sure that this bot (over all the other investments of marketing, sales, and support capital) is going to be delivering on something that’s uniquely ownable by your brand. We’re big lovers of new technology here, but we have had to say no to a bot idea or two because we could see that without the right support, training, and long-term plan, we would not be able to deliver on the brief’s main goal of relevance.
As an “new tool,” chatbots are in the messy in-between where bespoke solutions are still required because there aren’t many available “toolkits” yet. Even as those “toolkits” arise, to make these experiences ownable, its important to apply both a marketing and customer-first lens to how the chatbot experience is configured.
Ideating and building the right bot requires the ability for marketers and their agency partners to work in both a design innovation method as well as leveraging the rigour of the scientific method to get the right data, technology, and training into the process. In the meantime, we continue to experiment on the best practices of making this innovation a business reality.