The discussion on chatbots today is at a crescendo! My personal experience as I work with clients across retail, banking, ed-tech, healthcare, and insurance is that most enterprises today have chatbot in their consideration set for 2017. This is validated by research reports that tell us there is a growing market for this technology1. Still, as things stand today, I feel something is missing. Today’s chatbots, while helpful, are not completely connecting with me.
Let me give you an example: My DVR had not been recording my favorite shows for a week and no amount of rebooting the system was fixing the issue. Frustrated, I went to the website of my cable company to look for a phone number to call when a chat box popped up. Relieved, I started lamenting about my troubles, typing fast and furious into the chat box. To my dismay, the first response I got was to select the type of issue I was having from a menu of items. Clearly, this was no live person on the other side! This was a bot trying to match my issue to the canned answers available to it. When I couldn’t make a selection, I was asked to leave my account number and model number so a “live” person can get back to me with more answers. “But, I am already logged in! Can’t you just look me up and figure out my model number and fix things for me NOW?”
I think similar anecdotes have reached many businesses that have deployed chatbots. There was a recent study where a majority of businesses noted that their bots lacked “intelligence” and that was the biggest improvement these businesses would be focusing on for 2017. My personal theory is that this “intelligence” will come from feeding user context in real time to chatbots2. I am sure my level of frustration would have been a lot lower if the chatbot knew that I am logged in and from that it was able to access my information to at least have the basic knowledge of my subscription level. Additionally, it could have completely wow’ed me by letting me know that I am up for an upgrade and taken care of my problem right there and then.
I guess everyone will agree that such in-context interactions between a brand and its customers is the way to go for chatbots. As such interactions become more prevalent, the lines between customer service, marketing and customer experience will start to blur. The focus on chatbots will then be driven primarily by the need to engage the user with timely and contextual information. Then, personalization of interaction through the bots becomes a significant need for the enterprise. Let’s consider the following two scenarios below where a customer initiates a chat conversation on a website:
The first incarnation of chatbots being deployed today are engineered to provide the right answer for version A1 above. This entails extracting the meaning of the question and then matching a predefined answer to that meaning and sentiment. As long as I ask the same question as you, the answer is the same, except for the value of the answer that is extracted from a different system, which is specific to me. In this case, the value 30.
However, precedence and need dictates that the next generation of chatbot interaction will have to be personalized to each individual. This means that intent and meaning will need to be married to personal contextual data that resides in existing systems within the enterprise. This demands a context creation layer that sits behind the chatbot and provides real-time personalized context to each question that is asked, making the answer pertinent to the user who is asking the question.
The ZineOne engineering team took the matter in their own hands to help out consumers like me. We have integrated our real-time, event-driven context layer to Google’s API.ai providing our clients with significant acceleration with regards to the sophistication that they can handle within the bot. With this integration, the chatbot is now able to handle each question and answer specific to the individual who is asking the question (versus a one-for-all answer). The brand can run control groups and A/B testing on each answer based on personas they are defining, and can refine the answer based on machine learning from each individual behavior pattern.
In fact, we have seen that brands can build personas much faster by using chatbots than through email campaigns. The chatbot can gather the customer preferences and habits feed directly to build richer set of data for machine learning, compared to siphoning tons of feed from Facebook to do sentiment analysis. The profile information gathered from customer’s on-site activity, such as browsing for shoes on the website, can be used as a trigger to serve product recommendations and rich media content within the chatbot. Additionally, real-time activity feed can trigger chatbot interaction that can enable a brand to nudge a customer move along in their buyer journey. For example, if a selected course on the website is being abandoned, a chatbot could be initiated and incentives offered.
Overall, we feel that such contextual and in-the-moment chatbot interactions, drawing from real-time and historical customer context, can drive higher conversions and higher level of personalized engagement.