At 12pm, Mark Nelson and Kedar Doshi introduced the new integration of Einstein Analytics into the suite of Tableau products, showing impressive demonstrations of how Einstein integrates with Desktop, and Prep. This new feature will be released into Tableau products in early 2021, and Einstein Analytics will continue to operate under a new name, Tableau CRM.

The alliance between these two companies will help fill in the gap in the intersection between Data Analysis and Artificial Intelligence, allowing Tableau users to harness the power of AI directly into their dashboards and workflows. The integration is seamless, and examples were shown in minutes, emphasising the ease of use for Tableau developers.

In Tableau Desktop

In desktop, the user can connect directly to the Einstein platform by connecting to Einstein analytics server via the ‘more’ section of the connect pane in Desktop. After logging in, the user has access to the dataset, and all of the predictions defined within Einstein Data Manager. 

In the example, 10 billion records were loaded into Tableau and filtering took milliseconds to refresh, making the user experience for viewers of the dashboard fast and efficient. Viewers can filter their dashboard, and receive personalised predictions and insights based off the chosen filters. To achieve this, all you need to do is drag an ‘Einstein Discovery’ extension onto the dashboard, and it will sit embedded within the dashboard as an alternative to other visual tools. 

In Tableau Prep

If you would rather integrate predictions and predictive analysis into your prep output, that can then be brought directly into Tableau, it is possible to run Einstein analytics inside of prep, and select how you would like predict your insights. This becomes incredibly powerful when combined with Prep’s ability to view your output as you go, and easily undo any steps you wish to redo.

For collaborators, this is especially useful as it will allow people who may wish to view, or debug the flow the opportunity to walk through the workflow step by step, instead of having to decipher complex code in other data-prep languages. Instead of this being a previous (or future) step in the data lifecycle, the ability to integrate this all within 1, simple step is definitely an appealing prospect. 

Summary of the Live Demonstration

At 2pm on the European Schedule, Robert Brill provided an insightful demonstration of how we can embed Einstein Discovery into our Tableau Dashboards. This section will cover highlights from this talk, and give a high level overview of what to expect when it is released in 2021!

Highlights

  • Einstein Discovery either solves binary classification problems, or regression problems
  • All predictive tools that are to be used within Tableau must be created in the Einstein Data Manager 
  • Users can automatically or manually configure their models, with the help of Einstein Discovery
  • Users are given actionable insights throughout the design process of the model, as well as model accuracy, or performance metrics 
  • These models can be incorporated into Tableau as ‘worksheets’ or ‘parameters’, complimenting the ‘what-if’ analysis of parameters
  • Einstein Discovery works seamlessly with Tableau filters, allowing for an interactive AI experience

Part 1: Building a Model in Einstein Analytics

The first stage of implementing Einstein discovery within a Tableau dashboard begins in Einstein Data Manager, where users can either provide pre-prepared data, or transform the data by using the ‘Create Recipe’ tool. Users must define the column they are wishing to predict in this step, and in the demonstration, it was whether orders were likely to be late or early (with Superstore data). 

Once this has been created, the user must create a story. This story requires the developer to decide what they are trying to predict, and select the necessary columns they would like to use to as predictor features. At this stage, the user can decide whether they would like to manually configure the model, or allow Einstein Discovery to automatically configure the model. What is great about this step is that it allows for unlimited iterations, allowing the developer to re-visit and optimise their model over time.

Another really great feature here is that Einstein Discovery offers suggestions that may help improve the accuracy of the model, such as suggesting to convert the date field from ‘exact date’ to at a monthly granularity, which may improve the performance of the model. Developers can also view how strongly different variables were used as predictors, such as that Region was important when predicting whether an order would be early or late.

Einstein Discovery then gives us a high level overview of our model, and gives the developer drill-down abilities, to provide instant insights. For example, it may explain to that while on average first class shipped orders were usually late, in the central region, it actually outperformed other shipping methods.

Discovery also offers a ‘what could happen’ section, allowing us to test hypothesises, and create scenarios to see how our model reacts. For example, in January, in the state California, what is the likelihood that our orders will be late?

Performance Metrics

We are also able to delve deeper into the performance of our model, such as tracking the accuracy, AUC score, and the correlation between various predictors and the actual outcome. This feature greatly enhances the iterative experience of Einstein discovery, and will help quantitatively track improvements to the model over time. 

Part 2: Deploying the Model in Tableau

To deploy the model into Tableau, users must drag an extension onto the dashboard, and directly connect to Salesforce. Once they have chosen their predictive model they would like to embed within the dashboard, they need to match their columns in Einstein Discovery with the columns within Tableau. Once this is done, the insights drawn from the model will be available directly within the dashboard.  

Final Remarks

Tableau Conference-ish 2020 day one was action packed, and full of content. I was specifically excited to hear about this part of the conference, and I am really happy about Tableau integrating the abilities of Machine Learning directly into its software!