Explore Your Data with Einstein Copilot for Tableau
Note: To use this feature Tableau AI must be turned on in your site settings. For more information see Turn on Tableau AI for Your Site(Link opens in a new window).
Einstein Copilot for Tableau is a generative AI feature that helps you explore your data, create visualizations, and uncover insights with the help of a conversational assistant. Connect to a workbook or data source and use natural language to perform visual analysis. Use Einstein Copilot and the Tableau UI together to get to insights faster. With Einstein Copilot for Tableau, staring at a blank canvas becomes a thing of the past.
In the Tableau authoring experience, you can open the Einstein Copilot for Tableau conversation pane in a worksheet. Einstein Copilot can help with tasks like:
- Jumpstart your analysis: Einstein Copilot can suggest analytical questions based on your data
- Build a viz: "How many action movies did each director make?"
- Choose the best chart type for an analysis: "Show me the distribution of student's grades"
- Perform time series analysis: "What month had the largest growth in number of donors compared to the previous month?"
- Create calculated fields: "Create a field that calculates the difference between case open and closed dates and round up to weeks"
- Explain calculations: "Explain the "Days to Ship Actual" calculation"
- Filter, sort, and group data: "Only look at Saltwater and show me the fish with the smallest tank options"
Einstein Copilot creates visualizations based on your data, just like Show Me in Tableau. Chart types that Einstein Copilot can currently support include:
- Text
- Heatmap
- Bar
- Stacked bar
- Line
- Dual line
- Area
- Gantt
- Box plot
- Scatter plot
- Histogram
- Symbol map
- Filled map
- Tree map
- Pie
- Bullet
- Bubble
For more information about Tableau chart types, see Choose the Right Chart Type for Your Data(Link opens in a new window).
Einstein Copilot and Trust
Einstein Copilot for Tableau is built on top of the Einstein Trust Layer(Link opens in a new window) and inherits all of its security, governance, and Trust capabilities. As you interact with Einstein Copilot, neither your data nor your conversations that are sent to the Large Language Model (LLM) are saved to the LLM, and no customer data is ever used to train the model.
Einstein Copilot for Tableau only works with the data sources your workbook is connected to. It isn’t aware of any other Tableau data sources, so it can’t suggest relevant content, answer data lineage questions, or answer general-knowledge questions. User-defined policies for row and column level security are respected. The data a user has access to while using Einstein Copilot conforms to any row or column-level security policies that you have in place.
When you first open Einstein Copilot, it indexes your data to understand the context. This helps Einstein Copilot return relevant results based on your questions and your data source. The information it indexes includes field metadata (field captions, field descriptions, data roles, and data types) and up to 1000 unique field values if the data type is string (text).
The Einstein Trust Layer can be used to mask Personally Identifying Information (PII) before it is sent to the LLM. Using machine learning and pattern matching techniques, PII in prompts are replaced with generic tokens and then unmasked with original values in the response. For more information about how to configure PII Masking, see Select What Data to Mask(Link opens in a new window) in the Salesforce help. For an additional layer of protection, the Einstein Trust Layer guarantees that after the prompt and response are processed by the LLM, the LLM forgets both the prompt and the response.
Get started with Einstein Copilot
Einstein Copilot is available through a new conversation pane in the web authoring environment of a Tableau Cloud site. It’s only available for worksheets, so you won’t see this option in dashboards or stories. You also must have the site role of Creator or Explorer to use this feature.
Start by creating a new workbook or opening an existing one. For new workbooks, you start by connecting to your data. For faster results, we recommend using Einstein Copilot with extracts. However, you can also use Einstein Copilot with live database connections, and uploaded files with the following file types:
-
.hyper
-
.csv
-
.txt
-
.xlsx
Einstein Copilot doesn't support cubes. Also, if you're using data blending, Einstein Copilot can only be used with the primary data source.
Launch Einstein Copilot
To open the Einstein Copilot conversation pane, do the following:
-
Select the Einstein icon in the toolbar next to Show Me.
-
When the pane opens, select Got It to acknowledge the disclaimer and get started.
Build and change a viz
Einstein Copilot isn't an open-ended chatbot. It can only work within your data set and perform a specific set of actions around analysis and viz creation. To understand your data, Einstein starts by indexing the data set you’re connected to. If you’re connected to more than one data set, it will work with the selected data set shown in the Data pane. Einstein Copilot only works with the primary data source in a data blend.
Note: If you switch to a different data source while Einstein is processing your request, this can cause an error. Wait until Einstein responds before changing data sources; then make a new request using that data source.
Indexing scans the field names, data types (such as dates, text, or number fields), and samples 1000 unique field values to get a sense of what kind of data is in that field. Re-indexing occurs periodically when a change is made, such as renaming a field or creating a new calculated field.
After indexing is complete, Einstein Copilot provides some suggested questions you could ask about your data to help you get started, or type your own questions into the text box using natural language to describe what you want.
After Einstein Copilot creates the viz, you can interact and modify it, just like if an analyst built and shared a viz with you. Ask more questions with Einstein Copilot to iterate and explore your data further, or take the reins and continue your analysis on your own directly in the Tableau interface.
If the viz that Einstein created isn't what you're expecting, you can either provide additional information by typing what you want in the text box, or try selecting the Retry with Einstein and Einstein will query the LLM again and provide a new response to your request.
Use suggestions to jump start your analysis
Whether you're facing a blank canvas or are deeper into your analysis, Einstein can help by suggesting questions you can ask to dig deeper into your data. Before you begin, select the Suggestions button and Einstein will create three questions you might ask based on your data source. To see more questions, select Retry with Einstein .
Select a question, and Einstein creates the corresponding viz for you.
You can also use suggestions to dive deeper into your analysis as you interact with vizzes Einstein has created. At the bottom of the confirmation response from Einstein, select the Suggestions button and Einstein will suggest some actions you might take to discover more insights about your data. When you do this, the Suggestions button is converted to a Recreate button so that you can go back and recreate the previous viz if needed.
Simply select an option to apply the suggested action to your viz. If you don't like those suggestions, just select Retry with Einstein to see more options.
Create calculations
Building calculations in Tableau can sometimes be tricky. If you're new to Tableau, you might not know the right syntax to use or how to properly format your calculation. Einstein Copilot for Tableau can do the heavy lifting for you, keeping you in the flow of your analysis.
Simply ask Einstein to help you create a calculation using natural language to describe what you want to do. Einstein Copilot does the following:
-
Opens the Calculation Editor
-
Adds the suggested syntax
-
Names the calculation
-
Provides an explanation of the calculation in the Einstein Copilot pane to help you evaluate it for accuracy and build proficiency
Review the calculation, edit it if needed, and accept it. Einstein Copilot adds the new calculated field to the Data pane and it’s ready to use in your analysis.
There are two ways to ask for help with calculations:
-
In the conversation pane: Ask Einstein to create calculations for you as part of the flow of your conversation. Just describe your calculation using natural language, and Einstein does the rest.
-
In the Calculation Editor: Open the Calculation Editor and select the Einstein Icon. This opens the conversation pane if it isn’t already open. If it is, just type your calculation description in the text box, and Einstein Copilot adds the suggested syntax right into the open Calculation Editor.
Tips for creating calculations
To get the best results when asking Einstein to help you create a calculation, be specific. When your goal is to have Einstein write a calculation for you, use that specific language. For example, “Create a calc…”, “Write a calc…”, “Create a calculated field…”, or simply just “Calculate…”
It's also important to avoid being too generic. For example, instead of saying “Identify my most profitable products”, try “Calculate the profit ratio by product name”.
This helps Einstein understand both your intent (to create a calculation) as well as what you mean by the fields you want to calculate.
Calculation Descriptions
Understanding what a calculation does is just as important as creating the calculation itself. Einstein Copilot can help with that.
Whether it’s a calculation that Einstein Copilot created for you or one that exists as part of your data set, you can ask Einstein Copilot to explain the calculation. For example, “Explain the Business Days Between Order and Ship calculation.”
Not only can this help you verify that the calculation does what you need, this feature can also help you build proficiency in using calculated fields in your Tableau analysis.
Edit calculations
At any time you can go back to any calculation that Einstein Copilot created for you and edit it. Currently, you can manually edit the calculation in the Calculation Editor. Iterating on the calculation in Edit mode is not yet supported with Einstein Copilot.
To edit a calculation, do the following:
-
In the conversation pane, find the calculation you want to edit.
-
Click Edit.
-
In the Calculation Editor, make any changes you need, then click OK.
To ask Einstein Copilot to update an existing calculation in the Data pane, open the calculation in the Calculation Editor first.
-
Right-click or Cmd+Click (MacOS) on the field in the Data pane.
-
Select Edit…
-
Type the update you want to make in the text box in the Einstein Copilot conversation pane.
Calculation limitations
When creating calculations using Einstein Copilot for Tableau, the following functionality is not yet supported:
-
When creating a calculation, Einstein Copilot can't yet ask clarifying questions. You might need to be specific or iterate, such as specifying "change the field FIRST NAME into proper capitalization" if there is more than one field with the word "name" in its name.
-
Available calculations can vary by connection type. For example, certain date functions such as DATEPARSE(Link opens in a new window) are only supported by data extracts and a subset of possible connectors.
If your request involves a function that isn't supported by your live data connection, Einstein Copilot may create the calculation for you but it will be in an error state. To use the calculation, you'll need to take an extract of your data source first. In some cases there may be alternative functions you can request, such as using DATE instead of DATEPARSE. For more information about these alternatives, see the functions documentation(Link opens in a new window).
-
Einstein Copilot can’t create a calculation and then automatically use it in the viz. Instead do this as a two-step process. Ask for the calculation and add it to the Data pane; then ask for the viz and reference the new calculated field by name.
Working with filters
You can ask Einstein to filter your viz using natural language. While Einstein does its best to understand your request and return a solution, sometimes it might need your help to return the response that you're looking for. For example, if you have high cardinality fields, fields with a lot of values, Einstein might ask you to select the values you want.
In cases like this, you can select Add Filter in the Einstein Copilot pane, and Einstein opens the Filter dialog for you. Just select the values you want and select OK to add your filter to the viz.
Conversation history and viz recreation
You interact with Einstein Copilot in the conversation pane. The pane is where Einstein offers suggestions for analysis and maintains your conversation history for that session. If you ask a new question, the viz itself updates, but the conversation pane maintains a history of all your requests and Einstein’s responses.
It also has interactive elements such as Retry with Einstein to ask Einstein to query the LLM again with the same request and create another version of the viz or Recreate to go back to return to a previous viz without querying the LLM, maintaining the same results.
If you close and reopen the conversation pane while still in your same session, your conversation history persists. If you close your workbook, this clears the conversation with Einstein Copilot. The conversation history isn’t saved and doesn’t appear the next time you open the workbook.
Einstein Copilot is a per-sheet experience. There’s no awareness of other worksheets in your workbook and conversations can’t be shared between worksheets. If you change to a new sheet, this starts a new conversation. Additionally, if you publish your workbook using Publish As, Einstein Copilot won't retain the conversation history.
If you want Einstein to forget the context of what you’ve done so far, start a new sheet so there’s no history in the conversation pane.
Tips for getting the best results from Einstein Copilot
Einstein Copilot does its best to understand your intent and your data. But it’s still learning. Use the following tips to help Einstein Copilot do its best and deliver great results for you.
Tip | Problem | What to Do |
---|---|---|
Use clean data | Messy data is hard to analyze and Einstein Copilot won’t know how to clean and prep your data for you. |
|
Hide irrelevant fields | Similarly named fields in your data set can be confusing to Einstein Copilot. | Hide fields you don’t need and Einstein Copilot won’t use these hidden fields. |
Be careful with synonyms | Einstein does a good job of understanding common synonyms. For example, it would know that “titles” and “films” could apply to a field called “movies.” However, it doesn’t know about any of your company specific terminology or acronyms that you may use. | Refer to the specific fields you’re interested in to help Einstein Copilot be more accurate. |
Be explicit with your intent | Einstein Copilot does its best to infer your intent, but it might not always get it right. | If you know you want Einstein Copilot to generate a viz, use clues in your request like “show me,” “create a viz,” “build a bar chart.” Alternatively, if you want Einstein Copilot to create a calculation, try starting your request with the phrase “Create a calc that...” |
Specify how to evaluate “top” | Requests like “top products,” “best salespeople,” “highest quarter” can be ambiguous to Einstein Copilot and it might not know how to evaluate what “top,” “best,” or “highest is.” | State how “top” should be measured and displayed. For example, try asking Einstein Copilot to show you the “top 10 products based on profit” or "top 3 products based on sales volume". |
Use specific terminology in your requests | Einstein can scan your data source for the field names and data in those fields, but it doesn't truly understand the data the way a human can. | The more descriptive you are in your request, the better Einstein is at providing a relevant visualization. If you know your data should be evaluated with an aggregation of MEDIAN instead of AVERAGE, specify that. |
Break down complex tasks | Einstein Copilot can’t update the data model and generate a visualization as part of a single step. |
Break down your tasks into parts and iterate. For example:
|
Only asks questions about your data | Einstein Copilot isn’t a general-purpose chatbot. It can’t answer consulting questions like “how should I analyze my data?” or general questions like “which cat breed makes the best pet?” | Ask questions targeted at analyzing your data and iterating and digging deeper into the viz results. |
Be the human in the loop
As with all AI, its important to review the results you get when using Einstein Copilot. Einstein will do it’s best to understand your data and your intent when asking questions, but it might not always get it right.
For example, Einstein tries to pick a default for things like date granularity (such as year, month, or day) or aggregation (such as median or average). If you know the aggregation you want, it's better to specify this in your request.
If Einstein gets it wrong, you can either restate your request and clarify what you want, or you can interact directly with the viz. After all, you're in the standard authoring environment and have access to all of Tableau's functionality.
You can also provide feedback at any time on the results with the thumbs up or thumbs down buttons that show up with every Einstein Copilot response.
If you click the thumbs down option, provide additional feedback to help improve Einstein Copilot’s responses.
Einstein Copilot for Tableau limitations
Tableau provides powerful tools for the human analyst, and Einstein Copilot is no different. Tableau AI isn't meant to replace analysts and data explorers; it's here to give you a boost. Your participation in the process is crucial.
Types of analysis
Einstein Copilot can't yet manage consultative questions like "how should I analyze my data?" or "Is there seasonality in this data?". Instead, you need to specify what you'd like to see, such as "What are my sales over time?"
Unsupported features
Einstein doesn't have access to all of Tableau's authoring features — yet, and it is currently only available in web authoring.
Some things that Einstein Copilot for Tableau is currently unable to do include:
-
Choose a data source for you or do data modeling (such as creating joins or relationships)
-
Support data in languages other than English. You can enter your requests in other supported languages, but the response will also only be in English.
-
Change data types, field roles, or caption names
-
Format a viz. Such as adding fields to the Marks card properties. For example detail, tooltips, and size.
-
Add reference lines
-
Organize or customize fields using groups, sets, or parameters
-
Create interactivity with elements such as filter controls, parameters, and actions
-
Build dashboards
-
Analyze large data sets. Einstein Copilot will not work as well if you have 100’s or thousands of fields in your data set, especially if they are similarly named. If you try to filter against fields that are high-cardinality (many distinct values for a single field), Einstein Copilot might need you to manually perform your filter operations.
-
Block indexing refreshes after data model changes