You are here:
Prepare for Einstein Classification Apps
Before you set up Einstein classification apps, review the data requirements, address data problems early, and learn how to introduce these features to your support team.
Required Editions
| View supported editions. |
Get to Know the Data Requirements
To predict values on case fields accurately, Einstein needs lots of closed cases to learn from. You must meet the minimum data requirements, but the more closed cases you have, the better!
- To build a classification model, Einstein needs at least 400 closed cases, but 10,000 or more is ideal. Closed cases means all closed cases that have a created date from the past 6 months, and include at least one string type field.
- If you add filter criteria to limit which cases Einstein learns from, Einstein only counts the closed cases that match your criteria.
- To predict a field’s value, Einstein needs at least 400 closed cases with a value in that field and at least 2 unique values across fields.
- Einstein supports encrypted text fields for training only. Learn more about field encryption.
Einstein Case Wrap-Up works only with LiveChat, so turn on Chat if it’s not active already. While you can start using Einstein Case Wrap-Up without chat transcripts on your past cases, Einstein’s recommendations are more accurate when your cases have them. But don’t worry if your initial batch of closed cases lacks chat conversations. Einstein keeps learning as your service reps close chat cases.
Address Data Issues Early
During setup, you get to decide which closed cases Einstein analyzes and learns from. Einstein can learn from all recently closed cases or from a subset of cases that you define using filters. Here’s more about that. These cases serve as Einstein’s textbook, and inform every field value prediction. It’s essential that they consistently and accurately represent your business practices.
These issues can hurt the quality of the case data that Einstein analyzes and slows your setup process. Consider them when you’re setting up your predictive model.
| Data Issue | Description | How to Address It |
|---|---|---|
| Insufficient filled fields for training | If the case fields that you selected as inputs for training don’t contain enough information for a service rep to classify the case, Einstein probably can’t classify it. |
|
| Duplicate cases | Einstein assumes that your closed cases are filled out correctly. If Case 1 and Case 2 are duplicates, any differences between their field values send Einstein mixed messages about correct data patterns. |
|
| Incorrectly completed fields | Cases whose fields aren’t filled out correctly make it hard for Einstein to know what’s correct. |
|
| Overlapping field values | When a picklist field’s values are too similar or generic, it’s hard for service reps—and therefore for Einstein—to know which value is correct. For example, values like “General” or “Other” can be overused, or overlapping values like “Mobile” versus “Mobile Application” can dilute your data. |
|
| Too many field values | If a picklist field included in your model has more than 100 values, it hurts the model’s accuracy. Here’s why:
|
|
Supported Languages
With Einstein Classification Apps, you can build a model to classify cases in: Arabic, Bulgarian, Chinese-simplified, Chinese-traditional, Croatian, Czech, Danish, Dutch, English, Finnish, French, Georgian, German, Greek, Hebrew, Hungarian, Indonesian, Italian, Japanese, Korean, Norwegian, Polish, Portuguese, Romanian, Russian, Serbian, Slovak, Slovene, Spanish, Swedish, Thai, Turkish, Ukrainian, and Vietnamese.
Take Einstein Classification Apps for a Test Run
While it’s fine to try out Einstein classification apps in a sandbox org, you can't copy your Einstein configuration between sandbox and production orgs. For this reason, we recommend turning it on in your production org and giving a small group of service reps access to the feature. This approach lets you gather feedback and compare the productivity of service reps who are and aren’t using these Einstein tools. When you’re comfortable with the way the apps are working, open them up to your larger team.
To try out Einstein Case Classification and Case Wrap-Up in a sandbox, follow these steps.
- Sync licenses with your production org by refreshing your sandbox. If refreshing your sandbox isn’t an option, try matching production licenses.
- Copy enough closed cases to the sandbox to meet the data requirements described earlier. Select cases that are representative of your overall case data—you can use random selection or use all cases from a specific time period.
- Assign the Einstein user permission sets to sandbox users.
- Set Up Einstein Classification Apps in the sandbox.
The Try Einstein version lets you build one model for each app to recommend field values to your service reps. For this version of Einstein Case Classification, you can’t automate fields: Einstein can select the best value for a field but can’t automatically update the case or have Einstein Case Routing assign it to the right service rep.
The paid version of Einstein Case Classification and Case Wrap-Up is available with Service Cloud Einstein add-on license. This license allows for five models per app. For Einstein Case Classification, it also includes automatic field updates and Einstein Case Routing. To upgrade, talk to your Salesforce account executive.
