July 15, 2019

What to do When Your Chatbot is Wrong: Extracting Value From Failure

Article

Accept it. Your chatbot isn’t perfect

Chatbot failure exists. If you think your chatbot is perfect, you’re likely in for a rough ride once the user base starts to increase. While there are a lot of chatbots in the market that operate at an incredibly advanced level of Machine Learning (ML) and Natural Language Processing (NLP), even these bots are unlikely to be foolproof. Add that to the fact that the vast majority of chatbots lack these advanced ML and NLP capabilities and you should have reason to worry about your bots failing. Not acknowledging your chatbot’s shortcomings is a recipe for a UX disaster.

Don’t let your chatbot hide from its failures

So what should you do when a user asks something your chatbot doesn’t know, or your bot provides a user with an incorrect answer? That’s simple; your chatbot should be honest with the user. Your parents always told you that honesty was the best policy anyway. Acknowledging the failure not only confirms to the user that they don’t have the correct answer, but it provides an opportunity to help guide the user to the right answer (if possible) or collect valuable feedback to improve for future use.

Try to extract value from the failures

One of the most common lessons you’ll hear from successful individuals is that failure leads to success by providing an opportunity to learn and grow from your mistakes. This same lesson applies to chatbots.When a user asks the chatbot a question, and it has no clear answer to respond with don’t just say something along the lines of ‘I’m sorry I don’t have an answer for that.’ While that response does admit failure, it does not attempt to extract any value from the failure. A better response would read ‘I’m sorry, but I’m not sure how to answer that. Could you try rephrasing your question or providing more detail? If you still can’t find the answer you’re looking for feel free to navigate through the menu buttons below to get a better feel for what I know.’ This answer not only acknowledges the failure but it also provides useful tips to help the user find the right answer hopefully.

Boomtown's Poynt Bot helping a user find the right answer after not finding a potential answer.

Boomtown's Poynt Bot helping a user find the right answer after not finding a potential answer.

Allow the user to express dissatisfaction and learn from it

Another popular example of chatbot failures is when a user isn’t satisfied with the answer they’ve received. In most cases, chatbots provide a thumbs up and thumbs down button or some other method for users to voice satisfaction or dissatisfaction with an answer. When a user expresses dissatisfaction, there’s an excellent opportunity to learn how to improve. For example, when a user clicks the thumbs down button the chatbot should respond by not only apologizing but also asking for user feedback. This feedback could easily prevent a similar failure from occurring again in the future.

Boomtown's Poynt Bot asking for feedback after a user responds with a thumbs down.

Boomtown's Poynt Bot asking for feedback after a user responds with thumbs down. User feedback is only valuable when it can be analyzed efficiently to determine what changes are necessary. Displaying this feedback in the appropriate charts and diagrams can help to interpret this feedback better and assess how to best react to it.

Google's Chatbase Analytics shows you the most popular conversation flows and users' satisfaction at each chatbot response point.

Google's Chatbase analytics shows you the most popular conversation flows and users' satisfaction at each chatbot response point.[/caption]Have you used the Boomtown chatbots yet? If not, we’d love to show you a demo and prove just how valuable they can be to your support team. If you have used our chatbots already, we’d gladly welcome any feedback you may have.