Why Analytics Is the Key to Building Bots That Solve Problems

Liusha Huang, Software Engineering Manager at Wizeline, shares why using analytics is essential to training bots that solve users' problems.

I like to think of bots as children who must be taught a language. Just like kids who do not understand the difference between “good” and “bad” words, bots only know what they’re programmed to understand. Analytics provide insights, data, and context.

It’s important that bot developers view analytics as an integral part of the development process. The central tenant of good chatbots is that they continuously learn and adapt. Without this, bots cannot provide the best possible experience for the user.

“I like to think of bots as children who must be taught a language.”

Analytics provide actionable insights

Bot trainers have a lot in common with parents. They are responsible for showing bots how to respond appropriately when spoken to. They do this through natural language processing, which maps user questions to bot responses.

Analytics offers two advantages:

  • They help bot trainers make decisions based on data
  • They provide insights from the content of conversations

The chatbot we’ve been working on for a large real estate firm illustrates this well. The company requested a bot that curates the experiences of realtors and homebuyers during visits to premier homes. We use beacons, the technology that emits location information over Bluetooth, to notify visitors when they’re in specific rooms and locations, and provide detailed and relevant information (e.g., the year the house was built).

Our bot trainer works to humanize the conversation flows by anticipating typos, grammatical errors, and how chat interactions might progress. But no matter how many phrases the bot is trained with before launch, trainers must rely on analytics if they want a definitive answer to questions such as, “How is our bot better than it was a month ago?”

Analytics empower bot trainers

Conversation data helps determine where bot trainers should focus. First, analytics help trainers identify conversation breaks — instances when the conversation stops because bots misunderstand or fail to understand users.

Let’s say, for example, that a bot has been live for eight weeks, and during the first month of launch, 200 conversation breaks occur. The bot trainer can dissect the interactions to find where the user stopped engaging and decide what can be done to keep it from happening in the future.

Second, analytics empower bot trainers with data on drop-off rates — instances when the customer stops engaging with the bot. This data helps bot trainers understand how customers value each stage of the conversation and where they lose interest.

Quantified chat flows demonstrate how many people successfully complete a conversation with the bot. Trainers use them to identify where they can reroute user conversations in the future and what they should change to reduce user error.

Wizeline dashboard showing user growth, behavior, and activity

Analytics help improve the customer experience

There are multiple ways chatbots can help businesses connect with their customers. Call centers are a great example. Currently, most customer support agents have a log of the conversation history. However, the information isn’t always presented in a way that helps agents provide actionable resolutions.

I recently called a company because my account was hacked. The conversation was tedious. I had to repeat the same information when I called again. In an ideal world, the agent would say “Liusha, you’ve called five times in the past two weeks. Are you calling about your hacked account?” The bot would pull and analyze a caller’s conversation history to gather patterns during all customer support requests. This makes for a more productive and efficient interaction. It also helps guide the next conversation.

The goal is for chatbots to augment the customer service experience—to free up support agents by removing the tedious work and help them solve problems more efficiently. Analytics ensure that every interaction provides value and the chatbot continues to learn. The real value of analytics lies in the insights interpreted by real people.

Dashboard showing top messages customer write to a chatbot

Analytics define bots that mean business

Most bots do not have analytics built into them. That’s why there are so many bots on the market that don’t solve problems for users. In many cases, these bots exist for fun, but beyond entertainment, they don’t deliver the measurable results companies want to see.

A lot of companies want to build their own bot and must integrate with third parties for measurement. In our experience at Wizeline, it’s easier to build a bot than it is to train one. You not only need analytics data but also dedicated time and effort to make improvements based on the data.

If you want to drive decisions and enhance customer experiences based on actionable insights, consider a chatbot platform with robust analytics. Our platform can work with the tools your customers and employees already use. We have conversational UI and analytics experts who are happy to answer your questions. Don’t hesitate to drop us a line at

David Salak

Posted by David Salak on April 12, 2018