By: Merete Myhrvold, Partner Success Manager & Rosie Smithells, AI Interaction Architect, boost.ai
Focusing on writing great dialogue for a chatbot is a crucial part of an AI trainer’s job in building better customer interactions.
A great customer service experience is something that sticks with you for a long time. When you hang up the phone after talking with your bank or insurance company, you want to not only feel like you got the answers you were looking for, but that the whole process was handled in a professional and timely manner. Interacting with a chatbot should be no different. In fact, in some ways, the stakes are higher because you know that you’re not conversing with a human and therefore your level of expectation for what a machine can do is increased.
When it comes to AI chatbot training, the goal is to provide as good (or better!) a user experience than one would expect when communicating with a business via traditional customer service channels like phone or e-mail. In order to do that, AI trainers need to first understand how the various types of people that will communicate with their AI chatbot think.
When developing the official virtual agent for Slush 2019, we needed to consider the kinds of questions that the 25,000+ attendees to Europe’s biggest startup convention might ask before writing even a single line of dialogue. Everything from ticketing to programme scheduling to sauna timings needed to be taken into account and, ultimately, this gave us a strong foundation on which to begin building the chatbot.
A common misconception is that the AI trainers in charge of designing the conversation flows are required to have vast amounts of technical expertise or a Ph.D. in data science. This simply isn’t the case. At boost.ai, our chatbots are powered by conversational AI which allows us to build experiences that are on-par with talking to a human customer service rep. While a basic technical understanding can help, it’s just as important to be able to bring in a wealth of experience from different fields such as customer service, support, sales, marketing, etc.
Training and improving
We train our virtual agents by improving their cognitive knowledge. Even if we can’t anticipate each and every way a customer might ask a question, conversational AI makes it possible for a virtual agent to predict to the correct topic with a minimal amount of training sentences. This means that instead of having a high volume of training sentences, we focus on bringing a higher level of quality to each sentence that we write. To enhance the quality, we emphasize on creating training sentences with variety, therefore, understanding the needs of the end-user is important.
Writing good training data is key to building great conversations. It’s important to not have your virtual agent feel too robotic - customers can spot this a mile away! Injecting empathy and personality into dialogue is a great way to avoid this and helps to show the end-user that you recognize their situation.
Remember that training and maintaining a virtual agent is a continuous process. It requires frequent analysis of conversation data to improve the model and as the requirements of your customers evolve, so too should your virtual agent to meet them.