Effective training is often said to be one of the keys to getting a Conversational AI solution working in a truly helpful way. But before we can even move to AI training, we have to get the core structure and topics mapped out in a manner that is both logical and easily manageable even years later when the number of topics might have increased tenfold.
The process of bot building and content creation starts from a thought process: What kind of use cases are we trying solve with the use of automation? Who are the persons at the other side of the screen and what kind of conversations are we going to have? Backlogs of chat messages or email data can often be a valuable insight to the thought processes of the customers. Analyzing already existing chat data through the angle of your business Key Performance Metrics (KPIs) can also show which topics need the most attention.
It is important to decide which topics are important for the user and the business goals. But not less important is also to understand the overall solvability of every topic. For this reason, we have developed an evaluation tool called Topic Rating Matrix (TRM). It is a simple but very useful tool that helps to figure out what should be added to the virtual assistant’s repertoire and what should be avoided.
In a telecom use case, for example, many customers might have problems with mobile connection, but there can be a number of reasons for poor connection. So the occurrence is high but the solvability is tricky. In a nutshell, one should aim to include topics that have high occurrence and high to medium solvability. You can read more about TRM here.
In addition to that, we recommend to follow the 80/20 rule, when creating a new virtual assistant. According to the rule, a typical Conversational AI solution should automate and prioritize 20% of total topics that make up 80% or most of the chat volume. AI-powered virtual assistants are not meant to replace human colleagues and they definitely cannot know answers to wide variety of topics as humans can today (at least not yet). But this is not a problem, bots have proven to provide significant value when they’re responsible for narrowly specified tasks.
After that, it does not hurt to move thought processes to the concept of how the content should be structured. This is important mainly due to organizational matters. In the beginning, there might be 30-40 topics, but as the bot is a constantly evolving, the topics count might increase to 300-400 in about a few years. Usually, the topics are structured either in a flowchart or a folder style structure. A flowchart has its perks in data visualization, but it gets increasingly difficult to get a good visual overview of the whole structure. A folder style structure does not have the same visualization qualities, but can host a lot more topics while keeping it visually clean:
When you have a good overview of the topics that should be prioritized, then we advise to make menu out of them. Menu buttons you can host nested flows underneath them. For example, a telecom company will probably have a main menu button called Invoices which will direct to the second tier menu. On the second tier, there might be Amount Due; Invoice Settings, Credit Limit et cetera.
A good rule of thumb is to include about 3-8 main topic groups, which will form the main menu (and sub-actions under every topic). This will develop a solid foundation of high priority core topics that are vital for the business. This method of limiting your options will be also helpful later, when one might want to start include topics that are already similar to the existing ones. It is important to avoid duplicate topics because ideally you want the content to be specific and unambiguous. If the topics are overlapping and similar in their nature or wording, then the algorithms may get confused and the bot may start to give wrong answers. So we advise to think it through – is the new topic different enough to justify creating 2 different intents?
After mapping out the main content, it is strongly advised to teach the virtual assistant some general knowledge and social intelligence as well. After all, chatbots do resemble a real person so the actual users will value the experience higher when they can communicate as an equal. Some of the chit-chat topics could include answers for weather enquiries, jokes, greetings, rude comments and so on. Those topics will also reflect the bot personality, which can be a great way to show off the values and the culture of a company.
Finally, if all of the above is completed, one could further enhance the user experience with visual content. The exact possibilities differ from platform to platform, as every company has their own interface and the limitations are bound within it. A web browser bot will have different capabilities than a WhatsApp or a Messenger bot, but the general idea stays the same – Visual content is good and preferred over simple plain text.
The most important visual content piece of a virtual assistant is probably the avatar. This will be constantly seen by every user of the chatbot, so think what image should be represented to the customers? You can also give it a name or just call it a virtual assistant.
We suggest to use spacing between chat bubbles and keep the individual chat bubbles or paragraphs as short and straight to the point as possible. A typical person is already used to chatting in short messages, so there is no use in going against the trend.
In addition to that, when providing information, use emojis to create visual anchors. An average person will more likely give their attention to the visual clues of emojis rather than a big blob of a text.
Furthermore, it is advised to ask the customer to give feedback after the conversation has ended. This could be, for example, a simple question called Was it helpful? or thumbs up/down buttons. This is only not useful for the company but also for the users themselves. If a virtual assistant user does not get a solution to their problems right away, then he/she will probably be inclined to give feedback (so that the same problem will not occur in the future).
After all, bots are all about communication that should be efficient but also light. So make a plan, structure and visualize the content and keep it concise!