Artemy is a CEO of Data Monsters, a Palo Alto based research lab and consulting company. Prior to Data Monsters Artemy founded a business intelligence startup, which raised $6M of venture capital and two years later was sold to a nationwide system integrator. Artemy is an expert in computational social science, knowledge mining, chaos theory.
Why Chatbots Fail, And How To Fix Them
Chatbots look damn smart at demonstrations when presenters follow the pre-designed scripts. But chatbots fail when real users come. Real users talk in an unexpected manner, change topics and so on.
Bots still have very few success stories, with very limited number of use cases. The technology did not take off. In March 2017 Facebook recommended replacing conversational experience with a three-level menu navigation. Another leader, Amazon Alexa has only one frequent use case: “Alexa, play a song”. Everything else does not stick.
The frequency of users’ requests follows the statistical distribution with the long tail. In order to keep conversation a good chatbot should be able to understand thousands topics, not dozens. That requires huge knowledge bases.
We analyzed thousands of chatbot logs and observed a significant probability of missunderstanding that multiplies with every next phrase. 10-30% of users say something which the chatbot is not prepared and trained for. Almost every long conversation frustrates the user. Retention rate is 3-5 times lower for bots than for mobile apps, which is a disaster.
We want to discuss these problems and offer technical solutions in order to improve experience, create knowledge bases faster and build useful self-learning chatbots.
