Completing the Chatbot Puzzle

Increasingly new prospects and customers alike are asking if Attivio is able to improve the performance of their chatbots – similar to the manner in which Attivio helps their live agents and self-support portals – by empowering them with better answers and insights to questions. While the short answer is yes, to better frame the shortfalls of modern chatbots and explain how specifically Attivio helps, we put together an ebook titled, Completing the Chatbot Puzzle: AI-Powered Answers.

The following is an excerpt from this ebook. To keep reading, download a free copy.

Technology-provider Spiceworks asked organizations currently deploying chatbots where their failures happen. Two of the main problems? Misunderstanding requests and misunderstanding the nuances of human dialogue. When asked what errors they encountered with chatbots, 59% of respondents reported that these two issues were problem areas. And these are just the problems experienced in asking a question. The problem is even worse if the question is understood, but the answer is irrelevant, outdated, incomplete, or just plain wrong.

47% of adult internet users polled in the US felt that chatbots had too many unhelpful responses (G2 Crowd). The problem is that many of today’s chatbots lack the ability to provide answers that are accurate, relevant, and contextual.

For the most part, today’s AI-backed chatbots are intent-driven, i.e., they attempt to interpret meaning, and discern exactly what a user is looking for. For example, a user typing “I’d like a virtual assistant” is likely looking to buy one, while a user saying “I’m thinking about a virtual assistant” is likely looking for general information.

The downsides of only intent-driven interfaces: There are several downsides to relying on intent-driven chat interfaces alone. First, the dialog flow ends up long and complicated. Just getting the opening hours of a business can become a multistep, programmatic process. Now imagine doing that for all the possible conversations that someone might have with your chatbot! A simple conversation tree can easily become a convoluted forest that’s hard to create and even more difficult to navigate.

Then there’s the problem that, once they identify the intent of a question, chatbots respond by performing a re-programmed action or give a pre-programmed answer, often just delivering an FAQ in a different interface. However, they’re delivered, these pre-programmed answers may be insufficient and out of date, often delivering just plain bad answers. And they certainly won’t be taking advantage of the latest information available within the organization.

Pre-programmed answers can’t anticipate every question: Another downside is the lack of a proper fallback strategy when those pre-programmed answers don’t quite work. There is no chatbot that will be able to handle - let alone provide - proper answers to all questions. No human can do that either. When faced with this problem most chatbots default to a simple “I don’t know that yet!” or continue to ask, “Can I help you with anything else?” Unfortunately, this type of interaction is frustrating to users who feel like they’re getting the runaround. And the problem goes well beyond user frustration. When chatbots can’t provide an answer, more calls come in to contact centers. More tickets are created. More work for support reps to resolve an issue that could well have been taken care of if the chatbot had been smart enough to provide the right answer, the right link to the right document for self-service.