You are here:
Predictive Insights in Tableau Semantics (Beta)
Predictive Insights adds real-time machine learning predictions to your semantic model as a field you can chart, filter, and reuse. Connect any AI model from Data 360, map your semantic model fields to its inputs, and get a predicted value for every record—no data exports, separate tools, or code needed.
Required Editions
| Available in: All editions supported by Data 360. See Data 360 edition availability. |
How Predictive Insights Work
Predictive Insights is built on the PREDICT function, which calls an AI
model hosted in Data 360 (formerly Einstein Studio) each time records are queried. When you
add a Predictive Insight to a semantic model, you select an AI model and map your semantic
model fields to that model's required inputs. For every record a query returns, Tableau
Semantics passes the mapped field values to the AI model and returns the prediction as a
field—just like any other field in your model.
For example, if your team has a trained model that predicts a case's probability to escalate, you can map the relevant case fields to the model's inputs and surface that escalation probability alongside other case metrics in any dashboard or report that reads from your semantic model.
Predictive Insights doesn't train models—it generates predictions from existing ones. Models can be Salesforce-provided or your own (bring-your-own models). Any model registered in Data 360 is available to select.
The PREDICT function requires Flex Credits to run unmetered, regardless of
whether you have the Tableau Next Unmetered SKU. Contact your Salesforce account executive
to understand how PREDICT affects credit consumption.
What a Predictive Insight Returns
A Predictive Insight always returns a predicted value. If the AI model supports explainability and you turn it on, the field also returns the top predictors and recommendations behind each prediction.
- Predicted Value: The AI model's predicted value for the record. The data type comes from the selected AI model—Number returns as a Measure, Text returns as a Dimension. Change the field type if needed; an invalid definition fails validation at query time.
- Insight Type (Dimension): Indicates whether the explainability value is a top predictor or a recommendation. Optional.
- Insight (Dimension): The predictor or recommendation value. Optional.
- Impact Score (Measure): The impact score for the insight, when the AI model defines it. Optional.
Not all AI models define top predictors and recommendations. If the AI model has none,
queries still succeed and return the predicted value, but the visualization shows "no data"
for insights. If you don't request top predictors or recommendations, the default value is
0.
Limitations
- Predictions for more than approximately 1,000 records can exceed the 30-second query timeout and fail. To improve performance, aggregate the inputs or apply a row-level filter to reduce the number of records.
- Models with more than one prediction output aren't supported.
- The
PREDICTfunction isn't supported for semantic models created from a Published Data Source (PDS). The Predictive Insights option is hidden for these semantic models. - Nested PREDICT functions aren't supported. For example, PREDICT(PREDICT([Model Name], Field1, Field2), FieldX) returns an error.
- Combining the PREDICT function with Level of Detail (LOD) calculations isn't supported.
- Add a Predictive Insight to a Semantic Model (Beta)
Add a Predictive Insight to surface an AI model's prediction as a field in your semantic model. For each record your model returns, Tableau Semantics passes the mapped field values to the AI model and returns the predicted value in real time.
