The Impact Report is a powerful tool that provides insights into automation potential and helps you find training data and common topics in your historical conversational data.

This article covers:

• Automation Rate: X% Automation Potential
• Human + AI: Employee of the Month
• Automation Analysis: Your Top Topics
  • Similar Expression Suggestion
  • Matching Intents Suggestion
  • Advanced Filter & Search

Automation Rate: X% Automation Potential

This is where You can see the automation analytics based on the messages provided for the report.

The Impact Report is split between two main parts: 

• Performance data
• Training data

The estimated performance of the model (based on your historical conversational data) is displayed as different metrics at the top of the report and includes:

• Potential automation rate
• Estimated chat duration reduction
• Estimated avg. response rates after AI
• etc.  

All the data is based on the actual performance of your customer support team. We calculate the potential AI impact based on the effectiveness of the model and use a high "confidence threshold" in these calculations in order to minimize false-positive predictions. 

Human + AI: Employee of the Month

How many conversations your bot can be handling without breaking a sweat.

Automation Analysis: Your Top Topics

This is the topic visualization and data table at the bottom of the report. This is where the real value of Impact Report comes in. Here we visualize the most common topics (suggested "Intents") the AI has found in your data. 

By clicking on Explore & Train of any of the topics will take you to the list of messages from your data that our AI pre-clustered for you. Here we are confirming this pre-clustering that the AI has done for us.  

The features are:

Similar Expression Suggestion

When you start selecting messages, similar messages will start to be highlighted. As you will continue selecting more messages the similarity level continue to adjust and it will become more accurate.

Keep in mind that suggestions are only there to help you train the bot. A high similarity level does NOT mean that they need to be trained. This needs to be considered separately for each  intent.

Matching Intent Suggestion

Based on the selected messages the AI algorithm will sort the list of intents with the highest matching ones on top of the list. This is just a suggestion to help you train the bot. If the intent you are looking for was not identified as matching one, you can use search to easily find it or simply create one.

Advanced Filters & Search

We have added new ways of filtering and searching for the messages! Here is a list of functionalities with descriptions on how to use them:

Similarity Threshold Adjust this so you only see expressions above a certain similarity level. 50% is a good starting point. Confidence Threshold Adjust this so you only see the messages between the minimum and the maximum levels. Having this between 60-80% would yield the most benefit. Number of Words Adjust this so you only see messages that are between a certain word count. This is particularly useful if you have Zendesk Support Automation Engine on.

Fuzzy Fuzzy search finds messages based on text similarity - perfect for when you want to search for typos or similar spelling of a word. Semantic Semantic search looks for words with meaning similarity. Perfect for when you need to go through data in a language you don't speak!  Show Trained Messages Select this if you want to see expressions that have already been trained.  Show Noisy Messages  Select this if you want to see expressions that are considered incomplete and fragmented. by the algorithm.  Show Only First Meaningful Messages Select this if you only want to see the first message that is not greeting in a conversation.