There have been some spectacular chatbot failures in recent years. Most commonly, chatbot failures take the form of greatly frustrating customers by allowing them to enter loops, unable to solve the problem, and not effectively programmed to be passed off to their human counterparts. 

Bots saying unacceptable things—Scatter Lab’s Luda Lee was at first widely embraced for her straight talking style. She gained more than 750,000 users by early 2021 and logged more than 70 million chats on Facebook. However, her training regiment allowed her to learn to say offensive things and after making homophobic comments and inappropriately sharing user data, Scatter Labs is being sued by hundreds of people.

Bots without common sense—bots can easily misinterpret the nuance of publicly available information and simply not make sense. Thinking about the subtleties of learning a language, we can easily relate to this—literal interpretations often just don’t make sense.

Bots that try to do too much—In 2015, Facebook launched a conseierge service through its Messenger app. By using M, the Messenger bot, users could request anything via the conseirge service. Requests were all over the map, were constantly getting routed to human customer service agents. Without solid plan to monetize the service, the very broad nature of the bot made it too costly to deliver a service that was valuable to customers.

Bots without human partners—bots operating within environments of complex customer needs need to be able to transfer the customer to their human partners. Without this transfer, customers will become frustrated putting the brand and organization at risk of losing customers.

Want to better understand how to train your enterprise AI chatbot? Get a copy of a new guide from ThoughtFocus that helps organizations understand the training necessary to benefit from robust, AI-driven chatbot? The new guide is written for organizations interested in scaling a customer service or support team by offloading routine queries to an AI chatbot in order to manage the costs of scaling the contact or support team.