How developers can improve their bots by listening to their users

Dashbot.io recently surpassed 50 million messages processed. We thank all our customers for the continued support — we greatly appreciate it! In thanking our customers, it got us thinking about how often bot users thank the bots as well.

The beauty of conversational interfaces is not only do people say what they want from the bot, they also say what they think about the bot. While we started out looking at the positive feedback, in fairness, we also considered the negative, as it provides an opportunity to improve the bot too.

[aditude-amp id="flyingcarpet" targeting='{"env":"staging","page_type":"article","post_id":2065283,"post_type":"guest","post_chan":"none","tags":null,"ai":false,"category":"none","all_categories":"bots,business,","session":"D"}']

That love/hate relationship

Have you told your bot lately that you love it? About 12 percent of Facebook bots on our platform and 18 percent of Slack bots received an “I love you” type of message from users.

In fairness, 11 percent of Facebook bots and 12 percent of Slack bots also received an “I hate you” type of message. In many cases the bots received both types of messages — a bit of a love/hate relationship: 10 percent of Facebook and 7 percent of the Slack bots have received both an “I love you” and an “I hate you” message.

If you expand the types of messages beyond “I love you” and “I hate you” to include other, less vehement positive and negative feedback  —  like “thank you” on the positive side and “you suck” on the negative side, the difference between positive and negative becomes more pronounced. For example, 34 percent of Facebook bots received a positive compliment (variations of I love you and thanks / thank you) versus 16 percent receiving a negative message. For Slack, the difference is even greater  —  41 percent received a positive comment and 16 percent received a negative one.

Related to these positive and negative messages is user sentiment, a metric that provides an indication of the users’ feelings when interacting with the bot.

Looking across all the Slack and Facebook bots on our platform, the average sentiment per bot overall is positive. Users generally like interacting with the bots.

What your users are telling you

To take action on this data, you need to know what led to the positive or negative messages and sentiment. Our Conversational Analytics lets you navigate the flow of messages through your bot to see what led to the message sent in.

For example, in the image below, we can see when users enter “stop” into our Slack trivia game, what was happening in the bot beforehand, and what happened after, and we can continue to navigate the flow in both directions.

This is useful for identifying where the bot is breaking down  —  if you look at the negative messages, or the “I don’t know” or other error messages, you can see what users entered beforehand. With this information, you can improve your bot to better handle the messages users send in.

[aditude-amp id="medium1" targeting='{"env":"staging","page_type":"article","post_id":2065283,"post_type":"guest","post_chan":"none","tags":null,"ai":false,"category":"none","all_categories":"bots,business,","session":"D"}']

While following the flow is incredibly useful in identifying where the bot is breaking down, it’s also helpful to look at the full transcripts; it may not be just the preceding messages that led to the negative user response, but the conversation itself. Of course Dashbot.io provides the full transcripts, as well as functionality to search all transcripts for a particular message. From the transcripts, you can jump back into the message funnel to navigate the flow of messages.

In addition to the transcripts and message funnel, Dashbot.io shows the top messages in to, and out from, the bot. This is an additional way to understand how users interact with your bot, as well as a jumping off point into the funnel. The top messages can be filtered based on content type too — images, stickers, buttons, etc. — as seen in the image below:

How to improve your bot

It’s important to look at the messages users are sending in to your bot to make sure you’re handling them appropriately. As mentioned, users will tell you what they want from your bot as well as what they think of your bot. Sometimes it’s not very pleasant, but feedback gives you an opportunity to improve your bot to increase overall user engagement. Here are three ways to make your bot better.

1. Fix what is broken

Examine how users are interacting with your bot and identify the areas where the bot is breaking down and causing users to get upset or abandon the bot. Armed with this information, you can improve the bot experience.

[aditude-amp id="medium2" targeting='{"env":"staging","page_type":"article","post_id":2065283,"post_type":"guest","post_chan":"none","tags":null,"ai":false,"category":"none","all_categories":"bots,business,","session":"D"}']

2. Add support for unhandled messages

Look at the messages users are sending in to your bot, especially the rich media messages. Is your bot set up to handle these messages? In our experience, customers who added support for rich media messages were able to increase user engagement overall.

3. Create a personality for your bot

Bots are conversational  —  like any conversation, it should be engaging. This is a perfect opportunity to create a personality for your bot. Our customers found that by supporting different media types, including images, video, and audio, they were able to develop a personality for the bot, which increased overall user engagement.

An extended form of this article appeared originally on Chatbots Magazine.