Deploy Models

Deploy a model so that you can use it to make predictions and improvements.

Note Einstein Discovery stories are now models. We wish we could snap our fingers to update the name everywhere, but you can expect to see the previous name in a few places until we replace it.

Note Before you deploy a model:

• Consider reviewing Model Metrics to assess its quality.
• For the classification use case (binary outcomes based on text fields), consider setting the threshold level in the Threshold Evaluation for Binary Classification Use Cases.
• If you want this model to use automated prediction writeback, before you deploy it, read the deployment requirements in Display Einstein Predictions Using Automated Prediction Writeback.

To deploy a model, open it, click Deploy Model (either from the Model Overview screen or from the model tools dropdown), and then complete the following steps.

1. Select a Target Prediction Definition or Model
   Choose how you want to deploy this model: to a new prediction definition, to an existing prediction definition, or to replace an existing model.
2. Select Whether to Connect to a Salesforce Object
   If you deploy this model to a new or existing prediction definition, decide which Salesforce object, if any, you want to associate with the prediction.
3. Map Model Variables
   Define the mapping between variables in the model to fields in the Salesforce object or to columns in the supplemental dataset.
4. Configure Projected Predictions
   If the model is configured to derive a projected prediction, revise time frame settings as needed.
5. Configure Segmentation Filters
   Choose whether to use the model to get predictions on all data or on just a segment (subset) of the data. For example, you can focus on a specific product model or a group of customers. A prediction definition can contain multiple models in which each model produces predictions for a different segment.
6. Select Actionable Variables for Suggested Improvements
   The suggested improvements from Einstein Discovery are recommendations that users can take action on to improve predicted outcomes. To get suggested improvements, at least one actionable variable must be selected on the model. Actionable variables represent data factors that people can control, such as deciding which marketing campaign to use for a particular customer.
7. Customize Prediction Text
   Provide custom text for variables that appear as top predictors or suggested improvements on your Salesforce records to make them easier to understand. For example, you can have Einstein explain that a top predictor for time in the sales cycle is “Repeat Purchase” instead of “Previous Closed Won Opportunity Match is True.” You can also have Einstein suggest to “Schedule a meeting” instead of “Set Last Customer Meeting More than 30 days to False.” Your custom text for top predictors and suggested improvements is configured in the model.
8. Review Your Selections and Deploy the Model
   Review your deployment settings before deploying the model.