Starting a Chatbot project brings various questions to the table. While we might spend a lot of time finding the right set of components to put the project together, little time is spent on the rather non-technical component that the Chatbots is about — content. At Martian & Machine we’ve tried to find a convenient way to communicate, plan and integrate content into our Chatbot. Throughout our process of bringing the topic one step closer to our clients, we’ve had to work hard on finding understandable ways of content mapping and collaboration in order to execute successfully. Furthermore — we needed to create a process. Here’s what we’ve come up with.
Chatbots are only as good as the narrative itself, and storytelling is way more important than cutting edge tech or continuous content sheets. To put it differently, even with an enormous amount of predefined content, conversations will still lack the necessary guidance to walk new customers through the narrative and create friction fast enough. The question that arises at this point is really from an user-centric perspective — compared to chatting with an agent, would you like an open end conversation where you need to direct the potential outcome or rather a smooth, fast guided story where several options are put in front of you.
Let’s presume writing to a chatbot includes at least minimal knowledge about the entity the chatbot represents on the one hand and insights about a topic (or problem itself) on the other. Most interactions fall short at the point where we’re provoked to take the first step. That is by no coincidence, though, and no sharp UI or tutorial solves it effectively for two reasons. One, we’re unsure about how the Chatbot interacts, hence, we’re unsure about our actions. Such uncertainty leads to a prolonged dwelling time within the chat while the risk of leaving without creating friction enhances. Two, we’re unsure what the bot is about, or what he can do for us.
Getting users attention and getting them into instantaneous engagement is key. We practice to build up conversational engagement by creating topic scenarios where predefined questions and outcomes are served upfront, with a limited need of interacting manually. Essential to guiding through sections is content optimization — thoughtfully divided into small topics, bundled into sections.
Following those easy steps while preparing scenarios, we’re often questioned of a methodology or system where those topics can be monitored and scaled independently. Therefore, we decided to visualize a content map and build it from the ground up.
Assuming we start from a blank slate, nothing there but the existing brand that the Chatbot needs to serve. Most probably, we’d start with what we know or have, and build our way down into the unknown. We tried to map it out that way — it made it a much easier for clients to visualize the whole content structure, and how it affects workload and conversational depth.
Content is mapped within three main sections: existing, adapted and created content.
Topics are distributed as items across the map within three important categories: existing, adapted, and created content. It seemed to make sense since we finally could explain (or document) a reliable methodology for creating new or structuring existing content. Let’s clarify the structure a bit:
Existing content: since most if not all clients have already a brief brand history or some general outlines, why not make use if it — this refers to simple questions like ‘how many of your stores are in Chicago’ or ‘how old is your brand.’ It boils down to optimizing what’s already there into questions and answers. Furthermore, if the brand uses any technical terms or product names, it’s advisable to pin them down and describe them, in case user stumbles upon them anytime during a conversation and wants to check them out — one could also name this topic ‘lexicon.’
Adapted content: This one’s a bit trickier. It consists of existing content (existing products, services, etc..) that has been heavily modified regarding structure to fit a smooth, conversational mold. Structurally, many small topics were created out of greater categories or content in general — for instance — for an automotive brand, general categories would consist of items like ‘cars, trucks, vans..’ while further down cars would comprise ‘Sedans, Suv’s..) — you get the idea. Now, within one small topic, for instance ‘Sedan’, dedicated content can be created in various directions. Keeping topics small and separate offered greater scalability of our content map and made adding, deleting or editing separate topics efficient.
Created content: Conversations require both parties to exchange information, in a ‘human’ manner. To keep the artificial side out of the equation, a persona is created. Often, we call our Bot by human names. Encourage him to roll out a joke on some occasions. React to insults. You name it. This is the type of content that has to be created from scratch and reflect the attitude of the brand. Such type of content is split into categories like general, humor, greetings, etc.
The next goal was to determine an amount of time or questions within a topic that would lead to friction. Short attention spans are to conquer, therefore a fast drill down method needs to be applied in order to guide the user within a few questions (with selectable options) through one topic and purpose some actions once the user gets through successfully. Those actions consist mostly of applying for something, buying an item or requesting a follow-up email, etc.
If all steps are through, a Goodbye topic is initiated, and the cycle is complete. From there on, the user can either continue to chat, and open up new topics, or call it a day.
Assuming the content map works well, Bot should be ready to get into a conversation with us. Now, let’s talk about making the conversation more interesting and engaged. What’s needed there are fun, emotional bits and pieces. Something like our friends would send us. Emoji, Gifs, pictures. Anything that communicates well and within brand guidelines, of course. We use such visual content whenever we can, for instance — welcome greetings or ‘did not get that’ messages. It’s a lot less frustrating to get such responses when something goes wrong, or when a positive emotion needs to be provoked. Here’s what I mean:
Not really. But we’re getting there. Although we’re surrounded with a lot of benchmarks regarding tools and references, Chatbots are still no counterpart to human interaction. That’s due to a simple reason — every chatbot needs to be tailored to a specific use case. And that’s a good thing. Given the opportunity for a continuous dialogue, we would never cut to the point. While Chatbots can’t afford to be dumb and unspecific at the same time, it’s of utter importance to construct a narrative that makes most of the use case and gets there within a few inputs. And that’s where we’re currently at. A world with chatbots designed to do a specific task better and faster than a human — as long as it stays specific. And I don’t mind. Actually, I enjoy it.
You can also find this article in Chatbots Magazine.