When Rule-Based Chatbots Hit the Wall: How to Overcome Their Limitations without Breaking the Bank


Most chatbots are rule-based. A rule defines that if certain keywords occur in the user’s question, a certain piece of content should be displayed. For example, “if the question includes words ‘replace’ and ‘battery,’ show the topic about replacing a battery.”

While this method is easy and relatively cheap to implement, it covers only simple use cases. It may work perfectly well if the amount of content is small and it’s not frequently updated. But what about a case when the procedure of replacing a battery is different for different product models? Or what if it’s different depending on the user’s role, and there are multiple possible roles and their combinations? You’d have to explicitly add rules for each variation and instruct the chatbot about the questions the user should be asked when information required for a precise and relevant answer is missing in the user’s question.

On top of that, every time you add new content, you have to manually add new rules. In the long run, rule-based chatbots are expensive and difficult to maintain, if the amount of content is significant and it’s frequently updated.

Another approach is to build a knowledge map of the subject domain which would automatically guide the chatbot about the questions the user should be asked, automatically identify semantic metadata of the content, and map the metadata to the knowledge map. This approach would make the chatbot smarter while reducing the maintenance efforts and costs.

In this session, we’ll discuss both approaches and see which approach works better in different use cases.