Top Tips for Using Einstein Copilot in Tableau
I had the opportunity to test out Einstein Copilot, here are my top tips for using the AI

Top Tips for Using Einstein Copilot in Tableau

Einstein Copilot is Tableau's new Generative AI power chart-building tool and is now generally available, as of 18. Traditionally a user would need to drag and drop dimensions and measure onto a canvas to build a chart but now they can use a prompt. This current version allows you to create simple charts, so it can be a great way for new users to start understanding your data.


If you've used other generative AI tools, like ChatGPT, you'll know that what you can achieve depends on how well you write prompts. This is a guide to getting consistent results with Einstein Copilot as of its first release, based on my usage and Tableau's help documentation.


Examples have been created from Makeover Monday 2024 Week 34 Which social media platforms are most common?

Data: https://data.world/makeovermonday/social-media-usage


Step 1: Prep Your Data for Einstein Copilot

Einstein Copilot scans your columns and uses their names when generating responses to your prompts. To get consistent results:


Bring Only the Data You Need

Filter out the unnecessary data to keep things simple. For instance, I only needed age data in the dimension, so I created a data source filter to bring less data into Tableau.


Remove or Hide Unnecessary Columns

If there are columns that aren't needed, hide them. For example, I created a new column to clean up text from a source column and then hid the original to prevent Einstein Copilot from using it.


Rename Columns with Descriptive Names

Descriptive names make it easier for Einstein Copilot to understand your prompts. In my case, I renamed the field "Category" to "Age Bands" for clarity.


Step 2: Exploring and Analyse Your Data With Einstein Copilot

Einstein Copilot makes it easy to start chart creation using its Suggestions feature. When you load it, Einstein will scan your data and suggest three starting points for your analysis. With a click, Tableau will build the chart for you.


These suggestions may not be perfect. In my example, two of the three weren’t very insightful, but the third gave me something I could work with.


Here’s what to look out for:

  • How Tableau Places Dimensions and Measures: Check if the data is arranged logically.
  • What Values Are Being Returned: Ensure they make sense in the context of your analysis.
  • Which Chart Type Is Used: Sometimes you might need a bar chart instead of a line chart, which can easily be adjusted in the Marks card.


For new users, this is a great way to learn how different configurations of dimensions and measures can create specific chart types, like a pie chart. But it’s also important to double-check the accuracy of the data shown. For instance, if you see a percentage over 300%, it’s a sign that there might be an extra dimension that needs to be included. And remember, even if Einstein suggests a line chart, you can always change it to a bar chart by selecting the appropriate option in the Marks card.


Beyond the suggested prompts, you can ask Einstein Copilot to create charts, perform calculations, and describe calculated fields. Here’s how to get the best results:

  • Prompt with Intent: State an action, e.g., "build a bar chart...", "create a calc to..."
  • Be Specific: Name the fields and how to use them, e.g., "top products by [measure] descending"
  • Work Step-by-Step: Simple single-action instructions e.g., "break this down by [region]"


Charts will be created immediately on the canvas. However, calculations open up the calculated field window, allowing you to verify the calculation is as expected before proceeding.


Example prompts I used:

  • "Build a bar chart of [platform] by [age bands] by [percentage]"
  • "Create a calc to show [percentage] where [age bands]='18-29'"
  • "Sort [platform] by [created calc] descending"


Notice how these prompts are short, concise, and reference specific column names without synonyms. This step-by-step approach helps build the chart, create the calculation, and apply it to the chart.


Einstein Copilot can interpret non-English languages. For example, I prompted it in German to:

"Erstellen Sie ein Balkendiagramm, das die Plattformen nach Prozentsatz und absteigender Sortierung zeigt"


And received what I expected, a bar chart sorted in descending order. However, note that Einstein’s responses in the dialogue box will still be in English.


Step 3: Staying on Track and Providing Feedback

Like other generative AI tools, Einstein Copilot predicts the best response based on its training data. Sometimes this response isn't desirable. Here’s how to stay on track with your analysis:


Back and Duplicate.

If you had a great piece of work and it just got prompted into oblivion, fear not! Use the back button in the top left and your work will be returned. Then, duplicate the sheet. This will create a new sheet, which means a new prompt window. In doing this you remove the context that took Einstein on the wrong path, any calculations you created have been preserved and can be found in the data pane on the left.


Was it Me or the AI?

Addressing what went wrong with the last prompt can help us learn what led to the wrong action. For me, I look over the last prompt and check:

  • What was the intent? Was the right action performed?
  • Were the correct fields selected? Could a similar field be renamed?
  • Did you try to do too much? Could your prompt be simplified?


Sometimes this will lead me to words or typos in the prompt that have led Einstein astray, other times I can take the same prompt into a new window and get the correct response I expected.


Let them know.

Use the feedback buttons ("thumbs up" or "thumbs down") to tell the developers when Einstein isn't acting as expected. This then helps these issues be identified and resolved for future updates. From experience, the team are very good at incorporating feedback so do utilise the buttons for a better Einstein in the future.


Einstein Copilot is a powerful tool that lets a conversation be part of the data analysis process. By prepping your data, crafting prompts that have intent, and providing feedback, you’ll take your Tableau skills to the next level.

要查看或添加评论,请登录

社区洞察

其他会员也浏览了