Best Practices Guide
Best Practices Guide
Welcome to Dashbot!
Below you'll discover our recommended workflows for understanding your conversational data, interpreting reports and insights, and prioritizing bot optimization.
This guide highlights best practices and reports to make the most out of Dashbot for analytics and optimization. Each section outlines key workflows and includes linked videos to walk you through these workflows step-by-step. If you have any questions, comments, or suggestions, please feel free to email us at email@example.com.
Pro tip: As you review the reports on your platform, use the ‘Favorites’ feature to easily access these reports in the future. Learn more in this short video.
Review the following reports for a holistic view of your bot's health:
- Engagement: Understand how long a user engages with the bot. Depending on your bot’s use case, you are looking to increase (i.e. engagement bot) or decrease (i.e. customer support bot) your user engagement time. Review this report under ‘Activity’.
- Retention: Analyze your daily users and their return rates. Review this report under ‘Activity’.
Review the following reports to monitor intent accuracy and intent health for NLP models:
- Intents In: Review the frequency of your success intents and the confidence score for each intent, found under ‘NLP Optimization’.
- Not Handled: Identify which messages trigger your bot's 'Not Handled' intent, found under ‘NLP Optimization’. Check out these reports in action.
Conversation paths provide a visual representation of the conversation flow.
This 1-minute video discusses how to use Conversation Paths to visualize your customer journey. With this interactive path, you can understand where drop-off occurs and better understand user behavior. Navigate to Behavior > Conversation Paths.
This short video will illustrate how you can measure and visualize user behavior such as conversion with customizable steps using our Funnels feature. Navigate to this feature at Behavior > Funnels.
There, you can create a new Funnel by defining a unique step in the conversation. You can configure funnel steps using either intents, messages, or custom fields from your JSON payload.
This 2-minute video reviews how to use Goals to optimize the user journey, with an example on measuring and minimizing escalation.
The first step is to identify your target conversions to build reports on key aspects of the user experience. You can select from existing goals or customize your own to gain insight on success rates and the conversation paths that led to it.
The Goals report can be used for a variety of features, with the top 3 use cases being:
- Sales Conversion
- Escalation (Transfer to Agent)
- Not Handled (Bot Failure)
Navigate to Behavior > Goals.
This 1-minute video shows how message funnels identify the ‘before’ and ‘after’ messages of a fallback response. By default, the top 10 exit messages are shown, and each one can be clicked to review in detail.
Navigate to Conversations > Exit Messages Out to help with your cause and effect analysis on improving bot failure rates.
Transcript data can be messy, but they provide key insight into user requests and are crucial for identifying new use cases. Dashbot helps simplify this process by incorporating advanced search properties to isolate the data of interest.
This 2-minute video demonstrates how you can use the Transcript Search feature to easily filter your data based on customer sentiment, intents, and more, to drill into a conversation and understand it's context.
Navigate to Conversations > Transcripts to search your bot’s historical conversations (and click 'more search options' for advanced configurations).
Phrase clusters groups together phrases with similar meaning using Dashbot's proprietary machine learning models. These are two key ways to get immediate insights from this feature:
Quickly identify Not Handled intents:
Analyze phrases that trigger a fallback intent to identify new training data, intents, and content to add to the NLP model for further training and optimization. Filter the data by your bot’s ‘Not Handled’ intent. Within each cluster, export the training phrases to create a new intent in your model or add these phrases to an existing intent.
Identify Mislabeled intents:
Identify a mishandled intent where multiple intents are grouped within a phrase cluster. Expand phrase clusters to view any mishandled intents and use these insights to improve NLP accuracy intent overlap.
Navigate to NLP Optimization > Phrase Clusters.
The following 3-minute video walks through the above two use cases.Edit this page on GitHub