Inside the Quality-First Contact Center

How Callzilla Utilizes Artificial Intelligence in their ChatBots

Blog_CallZilla_11Artificial Intelligence (AI) has been a hot topic in the contact center industry these past few years. Even with the term floating around, it can be difficult to visualize exactly how AI can be implemented in our tools and services to improve customer experience and make our lives easier. In this article, I’ll give my personal account of how the artificial intelligence in our ChatBot service does just that.

I oversee the automation projects for Callzilla, but my experience is in Client Success and Marketing. Since I’m not technically trained in programming or coding, I need user friendly tools that do the technical work for me, so I can focus on upkeep and operational settings.

tombot ai

Our ChatBots are hosted by TomBot, a conversational AI platform. ChatBot is an important service for us because not only does it save us and our clients time and money, but as a somewhat unexpected outcome of the progress we’ve made with TomBot this past year, the bot has also proven to be more efficient than live agents at communicating with customers in some cases.

The main contributor to that improvement was the self-learning tool that TomBot developed, which is where AI comes in. In the initial stages of a new bot, I create the business rules, basic responses, and associated flows manually and input them into TomBot’s system. Flows are the combination of expressions (or anticipated customer inputs) and predetermined responses. This base is what the bot uses to learn and grow from.

Before the self-learning tool, I used to have to read through transcripts to find areas of improvement for how the bot was interacting, and then manually add new conversation points on the backend. We were to the point where we considered hiring a second person to help me do just that step, as it was a time-consuming task.

Now, the self-learning tool with AI capabilities picks out all of the untrained queries, which are customer questions or comments that I hadn’t predetermined an answer for, and then it stores them in a separate tab. If the bot is 80% or more certain that it can categorize an appropriate answer for that unrecognized input, then it will automatically add it to an existing flow. If it’s less certain, it will leave them for me to review and give me it’s best guess for what the most appropriate flow would be.

This is a huge time saver because,

  1. I don’t have to spend time reading through thousands of transcripts looking for these unrecognized interactions.
  2. I don’t have to spend my own time or mental energy looking for the most appropriate response within our hundreds of flows.
  3. The bot will group untrained queries together that it feels are similar enough to have the same response, so it cuts down on my review time in the self-learning tool.

When reviewing these untrained queries, the bot is still fairly accurate at guessing the appropriate responses for the unrecognized conversation points. Out of these guesses with less than 80% certainty, the bot still suggested the correct response 46% of the time. Of the 32% of responses that were inaccurate guesses, most of them were appropriate but a better option existed.

Approving a flow that the bot suggested

 

The new expressions are automatically added to the flow after approval

 

Once I verify that the bot has an accurate response, I click “add to flow”, and it’s automatically added as an expression under that specific flow. Expressions are the possible inputs we anticipate customers will ask. If a customer asks something that is not already listed as an expression, then the bot would categorize that as an untrained query. The more expressions that the bot adds to its flows with its self-learning tool, the “smarter” it gets, and the more efficient it becomes as assisting our customers.

Conversational Context

 

I can also view the conversational context directly in the self-learning tool if I need more information before selecting the most appropriate flow. There is a separate tab that houses all ChatBot transcripts, but I appreciate this quick context checker which helps the review process move quickly.

This Artificial Intelligence is a game changer because the bot improves its efficiency on its own, or makes it easier for me to train it in the cases where it’s not certain. The idea is that every time customers come back to use the bot, their experience improves. The goal of our bots are not just to provide customer service, but to improve the customer’s experience.

You can find more information on our ChatBot, and other automated services here.

Chatbot Case Study