The more context we know – the more relevant and personal a conversation becomes. The same applies to a conversation between a person and a chatbot with the understanding that for an effective chatbot operation it is necessary to feed it with context. However, expert opinions differ on how you come to that context.
To understand what the customer actually means or to hear the question behind the question and be able to answer it, it is necessary that chatbots are context-driven. This context can consist of many things such as the information mentioned earlier in the conversation, personal characteristics, location, type of device the customer uses, product preferences and the history of previous contact moments.
On the other hand, a chatbot cannot (yet) react like a human being towards non-verbal communication such as body language or the tone of voice of the customer. Precisely because a chatbot can only rely on what the customer literally says or asks, the background against which the conversation is conducted is even more important than with a conversation between two people.
AI versus BI
Roughly, there seems to be two channels towards the approach to chatbots; those who go for Machine Learning and the supporters of business intelligence (BI) or business rules. Incidentally, both methods use various forms of Artificial Intelligence (AI), but that is for another blog.
Those who follow the developments in the market a little will see that many suppliers of chatbots are fully committed to Machine Learning. The (self) learning ability of Machine Learning is beyond dispute and there has been great progress in this area in recent years. At the same time, there are important drawbacks to Machine Learning. In my view, the largest drawback is that an organization who wants to implement a chatbot based on Machine Learning needs to have enough adequate data. In practice, this proves to be more difficult than expected. The amount of data needed to get a chatbot working properly and self-learning is often underestimated. Only a few companies have sufficient usable data. Even then, the data available is not always suitable, for example because it is incomplete, outdated, unambiguous or inaccurate. Another major drawback of Machine Learning is that technology is still in its infancy despite all the jubilation stories you see in the media. They have great potential, but are only applicable in a limited area for the time being.
Nevertheless, as mentioned before, many major players present Machine Learning as the most amazing technology for the development of chatbots. Of course, they do this for only one reason that is to take part in the hype and to sell more. Most organizations do not recognize that Machine Learning is still a technology in development, with the necessary aspects that need fine tuning especially when it comes to adding context. They only find this out when the supplier in question fails to get the chatbot up and running properly due to too much focus on the technology instead of the end goal.
Conversation is everything
You will not be surprised that I am an advocate of chatbots that not only rely on Machine Learning, but are also based on BI and business rules. This technology, with which customers can be classified, has more than proven itself over the years in marketing applications. With modern SEO technology, context variables can now be further broken down. They are proven profiling technologies that can be endlessly varied to make the context explicit and can be perfectly applied to chatbots. Maybe it does not sound as cool as Machine Learning, but they are very effective in giving different answers to two people asking the same question to a chatbot fitting their specific context. My rule is therefore: use Business Rules for the things you know and Machine Learning for the things you would like to know. In other words, use mathematics for things certain and statistics for the rest.
In addition, the underlying technology should not matter for an organisation anyway. It only has one goal, which is to facilitate a context-driven or personal conversation with the customer. In other words, it is not about conversational AI, but about contextual conversation. Not about technology, but about getting the job done.