AI Accelerator FAQs

Use Case Model FAQs

No unused models exist in your org. Create an Insights & Predictions model, deploy the predictive model, and then retry. For information about creating a model, see Create a Model. For information about deploying a model, see Deploy Models.

What happens if I delete a model that’s referenced in a use case?

If you delete a model that’s referenced in a use case, the use case becomes invalid. You have to remove the reference to the model from the use case and add another model.

If you delete a model that’s referenced in a use case, the use case becomes invalid. You have to remove the reference to the model from the use case and add another model.

See Create Einstein Discovery Model

Feature Extractor FAQs

Is configuring a feature extractor required?

No. Configure a feature extractor only when the features required by your model aren’t present in Salesforce records and are to be computed.

• If you create a use case, configure a feature extractor.
• For your use case that’s automatically created after enabling the use case setting:
  • If there’s no default feature extractor for autogenerated use cases, configure a feature extractor.
  • If the default feature extractor partially suits your requirements, extend the extractor.
  • If the default feature extractor doesn’t suit your requirements, use another feature extractor.

Find out more about the feature extractors for your use case in your use case’s documentation.

Why is there a cyclic reference between a use case model and a feature extractor?

Use case model is the parent and feature extractor is the child. A model can have multiple feature extractors out of which there can be only a single default feature extractor.

When is the default feature extractor used?

If you enable feature extraction but don’t specify a feature extractor ID for the scorecard, the default feature extractor is used.

Saving Computed Features and Prediction Results FAQs

Is configuring primary and secondary response objects required?

No. Configure primary and secondary response objects only when you want to save computed features and prediction results.

Which objects can be selected as primary and secondary response objects?

You can select any standard or custom object as the primary and secondary response object. For the primary response object, make sure that you select the object in which the model is deployed.

What are the different combinations in which I can configure response objects to save computed features and prediction results?

You can configure response objects in these combinations:

• Only the primary response object.
• Only the secondary response object.
• Both the primary and secondary response objects.

What's the purpose of the secondary response object? Can’t we store all the saved values in the primary response object?

You can store all the saved computed features and prediction results in the primary response object and don’t need a secondary response object. If you don’t want to save computed features and predictions results in a standard object’s required fields, configure a custom object without any required fields as the secondary response object.

Which fields of the primary and secondary response objects are to be mapped?

Map all the required fields of the primary and secondary response objects to the fields that store computed features or prediction results. If all the required fields of these objects aren’t mapped, computed features and prediction results aren’t saved.

What are the considerations to keep in mind when saving computed features and prediction results?

Keep these considerations in mind when saving computed features and prediction results:

• You can save insights about the prediction score and suggestions to improve the prediction score only if you save the prediction score.
• If you map fields to save features that are already available in Salesforce records, the field mapping is ignored and the features aren’t saved. Only new features that are computed at runtime are saved based on the field mapping.
• To save insights, you can map fields for the maximum number of insights. For example, if the Maximum Insights value is 3, map fields for the names and values of up to 3 insights. If you map fields for one insight, the insight with the highest impact is saved.
• To save suggestions, you can map fields for the maximum number of suggestions. For example, if the Maximum Suggestions value is 5, map fields for the names and values of up to 3 suggestions. If you map fields for two suggestions, the top two suggestions with the highest impact are saved.

I mapped fields for the maximum number of suggestions. However, the suggestions aren't saved due to invalid field mappings. How do I fix this issue?

In Lightning App Builder of the record page to which you added the scorecard, go to the configuration panel of the Einstein Predictions Using AI Accelerator scorecard. Make sure that the number of improvements to show is less than or equal to the maximum suggestions defined for your use case in the AI Accelerator Setup page. For example, if the maximum suggestions are 2, make sure that number of improvements to show is less than or equal to 2.

Scorecard FAQs

Can I add the scorecard to a Lightning page?

You can add the scorecard to the record pages of the primary response object selected in the AI Accelerator Setup page. This object is the same object on which the model is deployed. For example, if your use case is for Communication Cloud, you can add the scorecard to the account and customer interaction record pages.

What are the modes in which the scorecard runs the request to get predictions?

To get predictions, the scorecard runs the request in synchronous mode.