If Your Chatbot Isn't Working, Try Talla's Reasoning and Understanding Bot

Posted by Rob May on Jun 13, 2019 12:17:22 PM

A big source of business for us at Talla is failed chatbot projects. Building chatbots for customer support is hard because most of them are scripted. You have to go in and do some regular expression matching and then label language intents, and then map those to a workflow. It's a lot of work to create an interaction, and then the bot doesn't really learn much once it is deployed. As the world changes and new support issues arise, you have to go do a bunch of work to add those new issues.

Talla doesn't work that way. While the interface for Talla can be a chatbot, Talla is a workflow tool that actually works to reason about your content and understand what someone is asking. We do this by continuously learning from every interaction with a customer or a support agent.

There are 3 big benefits to this.

1. It isn't as much work. Talla gets updated with minor changes to existing workflows by observing and interacting with questions and answers. There is no script to map out.

2. It is more flexible. Talla's understanding engine can learn word and phrase similarities. It can learn the vernacular for your business. It can learn if "arm" means the human appendage, a chip in a device, or an Adjustable Rate Mortgage. We can build natural language understanding models for your business, based on the language you and your customers use, which is particularly helpful when your industry differs a bit from colloquial English.

3. It is more accurate. Because Talla is constantly learning from each interaction, and the natural language models are constantly re-trained, Talla is more up to date than the scripted or intent-based chatbots so popular in support orgs these days.

If you are far down the support chatbot path, your story may sound like this. You started out with the goal of building your own support chatbot. You used Amazon Lex or IBM Watson for the NLP, and build the rest with your own team. But training took too long, performance wasn't what you hoped, and the edge cases killed you. So then you bought a scripted bot from one of the numerous support chatbot vendors. You spend several weeks designing all your flows and it tested really well, but when it was exposed to real users, it broke more often than not.

If that sounds like your experience, it is entirely par for the course with support chatbots. So maybe it is time to move beyond chatbots to a bot that reasons and understands. Intelligent support automation isn't just a NLP problem. It requires a full workflow tool that is flexible, learns continuously, is available on top of any system you want to use, and is highly accurate. That is what we built at Talla, so when everything else you have tried has failed, you should give us a try. Our customers rave about Talla's performance. You can sign up for a demo here.

Topics: Customer Support, chatbot