Making Chatbots Talk Like People

The Sidekick AI Team | Friday October 8

Sidekick Agents trained to talk like humans, and that makes all the difference.

Today there are hundreds of chatbot companies out there offering a slick interface to build your own chatbot. Typically the user (you) end up trying to predict what your customers will say, and what the proper response will be. You can build up a massive tree of possible inputs, along with their appropriate responses, and still only cover a fraction of the possible needs your customers will have.

What you end up with is a hand-crafted experience that's really effective at serving a few select questions or use-cases. This is great if the customer expects it, but more often than not businesses try to replace their main customer service channel with one of these hand-crafted chatbots.

What's missing?

What's so great about humans, as opposed to chatbots, is that they are flexible. You don't need to train a new employee on how to respond to every possible customer need. You train them by giving them information about your company, and relying on their background of social experience to take that information and converse with your customers.

This flexibility is what's missing in today's chatbots. They don't have that social experience, and so need to be instructed on even basic responses, which leads to very stiff-feeling conversations. One of the primary reasons customers get frusterated with chatbots is this stiffness. They want to talk to someone that will understand them, not just throw back pre-written responses.

But how is Sidekick any different?

Sidekick takes a fundamentally different approach to building our agents. We never teach them individual responses, and never build massive chat trees trying to handle exponential possible conversation paths. Instead, we train large machine-learning systems on human-made data. Lot's of data. Gigabytes of data. This data covers millions of human written sentences and conversations, which helps our system to learn how people talk.

At the end of the training process, we sit down and do some chatting with our system, making sure it's up to our standards, and correcting it where it goes wrong. What comes out of this process is a system that is able to handle flexible, natural conversation and feel far more humanlike, all without requiring a single pre-scripted response.

Once a business sets up an account, they create their own Sidekick agent by simply entering unstructured information about their company in different topic sections. This informatin can be anything from FAQs to office locations to pizza prices. Our agent then takes this information, references it while a conversation is going on, and smoothly integrates it into it's responses. All of this results in a far more natural experience for the end customer than any rule-based chatbot can provide.

Head on over to our Solutions pages to chat with live Sidekick agents!