You are here:
Considerations for Setting Up Einstein Prediction Builder
Before you start building predictions, here are a few things to keep in mind.
Supported Data Types
Einstein Prediction Builder can make predictions for these types of fields.
- Checkbox
- Specially-constructed formula fields
- Numeric (Beta)Note This feature is a Beta Service. Customer may opt to try such Beta Service in its sole discretion. Any use of the Beta Service is subject to the applicable Beta Services Terms provided at Agreements and Terms.
Supported Objects
Einstein Prediction Builder works with all custom objects and supports these standard objects.
- Account
- AiVisitSummary
- Asset
- AssignedResource
- Campaign
- Case
- Contact
- Contract
- ContractLineItem
- Entitlement
- Lead
- LinkedArticle
- LiveChatTranscript
- Opportunity
- Order
- OrderItem
- OpportunityLineItem
- Product2
- ProductConsumed
- Quote
- QuoteLineItem
- ServiceAppointment
- ServiceContract
- WorkOrder
- WorkOrderLineItem
- WorkType
- User
Limits
The Details tab on the prediction scorecard shows various predictors for your model. It shows information about each predictor, such as the impact, correlation, and importance or weight. If your prediction model contains more than 100 predictors, it’s possible that you won't see them all. The scorecard shows you the top 100 predictors ranked by impact and by correlation. If your model has at least 100 predictors, the number displayed in the scorecard is likely to be from 100 through 200.
To build a successful prediction, your dataset must have enough records. Also, if your prediction object has segments, Einstein checks to see if each segment has enough records.
Here are the minimum data requirements.
| Type of Dataset | Minimum Number of Records |
|---|---|
| Overall dataset (or segment if dataset is segmented) | 400 |
| Example records (example set) | 400 |
| True and false values (binary fields only) | 100 per value |
Predictions in Sandbox
Predictions created in production aren’t copied to sandbox orgs. To use a prediction in sandbox, copy your licenses from production, and build a new prediction. For best results, use a recently refreshed, full sandbox.

