Setting up your data flow to use Data Conversations#

Data Conversations responses depend on the available data on your stack. This guide explains how to set up your data collection and processing to ensure you get the best responses.

Stack setup best practices#

To ensure that you get the best and most relevant insights from Data Conversations, perform the following steps:

  • Make sure your data is mapped correctly and that there are no duplicate names in the target fields you use. For more information, see Applying Data Mapping to a datastream.

  • Prepare the Data Dictionary for using Data Conversations. The target fields are used to get data for the analysis, so it is important to make them clear, unique, and descriptive. Use the following ideas to improve your Data Dictionary:

    • Make sure that the target fields have unique names that reflect their purpose.

      This helps prevent having multiple similar fields containing duplicate data, which could make your analysis less accurate.

    • Add descriptions for custom target fields.

      This provides the model with additional context about your data helping it choose more relevant data for your prompt. For more information, see Best practices for target field descriptions.

    For more information, see Creating and editing target fields.

  • Group your datastreams into workspaces depending on the use case. This way you can get separate insights about the relevant parts of your data, while still being able to get the full picture from all workspaces if needed.

  • Transfer data from all datastreams you want to get insights from into Adverity Data Storage or your own Snowflake warehouse connected to Adverity.

  • If you are loading data from Bundle datastreams into an Adverity storage, make sure that data from source datastreams is not loaded to the storage to avoid duplicates.

  • Make sure to not load data from the datastreams you use for testing into an Adverity storage. This will ensure that only the right data is available for Data Conversations. For more information about removing previously loaded data from Adverity Data Storage, see Deleting data from Adverity Data Storage.

Using Data Conversations with specific setups#

Data Conversations can help you wherever you are on your data journey. Get your data ready for Data Conversations in the following cases:

No data in Adverity Data Storage yet?

If your data in Adverity is ready for analysis but has not been loaded into Adverity Data Storage yet, enable it for the datastreams that you want to get insights from. For more information, see Loading data into Adverity Data Storage.

Do you prefer to store the data in your warehouse instead of Adverity Data Storage?

If you want to use Data Conversations but prefer to store your data on private infrastructure, set up a managed Snowflake warehouse and load your data there. For more information, see Connecting an external data warehouse.

Do you transform data outside of Adverity?

If you additionally process and transform your data outside of Adverity, Data Conversations could be useful to you in the following ways:

  • Use Data Conversations to get a deeper understanding of your data.

  • Set up a datastream in a new workspace to load your data back to an Adverity-managed storage after it has been transformed. This way you can get Data Conversations insights after the full processing.

Best practices for target field descriptions#

To ensure that Data Conversations get more context about the custom target fields you are using, add their descriptions in Data Dictionary. To create more helpful descriptions, follow these guidelines:

  • Use clear business language, avoiding technical jargon. Include internal terms and abbreviations only if they are used and understood by all potential users.

  • Include what the field is used for in its description.

  • Include relevant relationships to other fields.

  • Mention any filters, currencies, or time zones.

Example:

This ID maps to the ‘Campaign Name’ field and can be used to join ad-level data to campaign-level data.