Salesforce Dictionary - Free Salesforce GlossarySalesforce Dictionary
Full Einstein Case Classification entry
How-to guide

How to set up Einstein Case Classification in Salesforce

Einstein Case Classification auto-populates picklist fields on inbound Cases — Type, Reason, Priority, Sub-Type — based on patterns learned from your org's historical Case data. Reduces agent triage time and produces clean data for downstream routing. Requires Einstein for Service licensing.

By Dipojjal Chakrabarti · Editor, Salesforce DictionaryLast updated Apr 20, 2026

Einstein Case Classification auto-populates picklist fields on inbound Cases — Type, Reason, Priority, Sub-Type — based on patterns learned from your org's historical Case data. Reduces agent triage time and produces clean data for downstream routing. Requires Einstein for Service licensing.

  1. Confirm Einstein for Service licensing

    Setup → Einstein Setup. Einstein Case Classification is part of Einstein for Service add-ons.

  2. Open Setup → Einstein Case Classification

    Setup gear → Quick Find: Einstein Case → Einstein Case Classification.

  3. Click New Predictive Field

    Configure one model per picklist field you want to auto-classify.

  4. Pick the Case picklist field to predict

    Type / Reason / Priority / custom picklists. Each gets its own model.

  5. Pick training data filter

    Which historical Cases to learn from. Default: all closed Cases. Restrict to recent / specific record types if patterns shifted recently.

  6. Train the model

    Salesforce builds the model from training data. Takes minutes-to-hours depending on volume.

  7. Review accuracy stats

    Model card shows accuracy per picklist value. Below 60-70% accuracy means the model isn't reliable — refine training data or skip the field.

  8. Activate predictions on new Cases

    Once activated, new Cases auto-fill the predicted field. Agents see Einstein-suggested values with a magic-wand icon — accept or override.

Key options
Predicted Fieldremember

Picklist field to auto-classify.

Training Data Filterremember

Which historical Cases to learn from.

Confidence Thresholdremember

Minimum model confidence for predictions to populate.

Auto-Apply vs Suggestremember

Whether predictions write directly or just suggest to agents.

Gotchas
  • Model accuracy depends on training data quality. Cases with inconsistent / wrong picklist values produce a confused model — clean training data first.
  • Predictions can be wrong. Auto-Apply mode silently writes wrong values; Suggest mode lets agents override. Start with Suggest; promote to Auto-Apply only after observing accuracy.
  • Models retrain on schedule (weekly default). If your org's case patterns shift (new product launch, new support tier), accuracy can drop temporarily until retraining catches up.

See the full Einstein Case Classification entry

Einstein Case Classification includes the definition, worked example, deep dive, related terms, and a quiz.