You are here:
Considerations for Choosing Fields to Predict
When you create your Einstein classification model, you select the fields that Einstein learns from and the fields Einstein predicts. Follow our guidelines when choosing fields.
Required Editions
| View supported editions. |
Input Field Requirements
For Einstein Case Classification, you can include up to 30 input fields to train your model. Einstein Case Wrap-Up only supports subject and description as input fields. Einstein Classification supports these field types.
- String (TextArea or TextArea (Long))
- Picklist
- Lookup
At least one string-type input field is required for model training. The fields you select to train on affect the quality of Einstein’s predictions. We recommend you choose input fields that are relevant to the fields you want to predict. For example, if you want Einstein to predict the department for a case, include closed cases with accurate department data for training. More isn’t always better, and too many input fields can create extra noise, which negatively impacts Einstein’s predictions.
Field Volume Requirements
To predict a field’s value, Einstein needs closed cases with a value in that field. The more closed cases, the better.
- High: > 10,000
- Medium: 1,000–10,000
- Low: 400–1,000
- Not enough: < 400
If a field isn’t used or isn't used as part of your setup, don’t include it. We recommend choosing fields that are filled when the case is closed.
Field Type Requirements
When you add fields to your classification model, you can select standard and custom picklist, checkbox, and lookup fields.
- Picklist fields included in your model must contain at least two values. If your recently closed cases use only one value of a particular picklist field, the model fails.
- If a picklist or lookup field has more than 1,000 possible values, Einstein considers only the first 1,000 values. In general, fields with this many values result in poor prediction accuracy because of service rep errors on closed cases or a lack of example cases with certain values.
- Only single-select (not multi-select) picklists are supported.
- Other field types don’t appear in the list of available predicted, segment, or example criteria fields: AccountId, AssetId, BusinessHoursId, CommunityId, ConnectionReceivedId, ConnectionSentId, ContactId, CreatedById, EntitlementId, FeedItemId, Id, IsClosed, IsClosedOnCreate, IsDeleted, LastModifiedById, OwnerId, ParentId, QuestionId, RuleFilterId, SourceId, Status, UserRecordAccessId. Origin isn’t available as a predicted field, but you can use it in segment or example criteria.
- Formula fields aren’t supported as segmentation or training filter fields.
- Supported lookup fields include: Account, Address, Asset, Campaign, Case, ConsumptionSchedule, Contact, ContactPointAddress, Contract, ContractLineItem, DandBCompany, Document, DuplicateRecordSet, Entitlement, EntitlementContact, EnvironmentHubMember, Goal, GoalLink, Idea, IdeaTheme, Individual, Lead, LiveAgentSession, LiveChatTranscript, LiveChatVisitor, Location, Macro, Metric, MetricDataLink, OperatingHours, Opportunity, Order, Pricebook2, Product2, ProfileSkill, ProfileSkillEndorsement, ProfileSkillUser, PushTopic, Question, QuickText, Quote, SOSSession, ServiceContract, ServiceTerritory, SignupRequest, SocialPost, Solution, StreamingChannel, WorkCoaching, WorkFeedback, WorkFeedbackQuestion, WorkFeedbackQuestionSet, WorkFeedbackRequest, WorkGoal, WorkGoalCollaborator, WorkGoalLink, WorkOrder, WorkPerformanceCycle, WorkReward, WorkRewardFund, and WorkRewardFundType
- To make predictions for lookups to a custom object, the object's status must be set to Deployed.
Controlling Fields
A controlling field determines which options are available in another field.
- If you add a field with a controlling field to your model, add the controlling field to give Einstein context. For example, when a service rep selects a location field that limits the currency field options, include both fields in your model.
- If the controlling field isn’t available to add to your model, Einstein can still predict the dependent field.
- If Einstein’s predicted value for the dependent field isn’t compatible with the controlling field’s value, the prediction is hidden to preserve data quality.
Field Access and Page Layouts
- Make sure that the fields you choose are added to your console case page layout. Service reps can’t see predictions for a field that isn't included in the layout.
- If you narrow the scope of your model by defining a segment or an example case set, make sure that the user building the model has access to the fields used in the filter criteria. With field access, you’re warned when not enough data exists to build the model.
- If you include the Escalated or Language fields in a model, give the service reps using that classification app edit access to those fields. Escalated and Language are read only by default.
The Language Field
When you add the Service Cloud Einstein license, the Language field is available on cases. If you add that field to your model and give service reps access to it, Einstein can predict language on your cases. For example, set up Einstein Case Classification to detect languages, and use Einstein Case Routing to assign cases to service reps with those language skills.
- To predict a case’s language, Einstein doesn’t consider closed-case data. For example, even if no Portuguese cases are among your closed cases, Einstein can still determine that a case’s language is Portuguese.
- To see recommendations for languages that you expect in your cases, enable all dialects for those languages in Setup. For example, enable both Portuguese (Brazil) and Portuguese (European) so that no matter which dialect the case is in, service reps can see Einstein’s predictions.
- After you build your model, you can tell Einstein to select or save the best value for the Language field, but the confidence threshold for those settings isn’t adjustable. Einstein’s predictions are more accurate when each case is in a single language.
