What is Einstein Case Classification?
Salesforce’s Einstein Case Classification is a tool that utilizes machine learning to suggest or automatically populate Case record fields. Einstein Case Classification always uses an org’s last six months of closed cases to recommend or populate picklist or checkbox field values. As an org grows and cases close, Einstein Case Classification will automatically adjust the prediction model to account for potential new patterns. Recommendations will be accessible from specific case records, where a user will be prompted to select a field value from a list of top 3 suggestions.
Some common fields to predict with Case Classification are case reason, language, escalated, and priority. Currently, Einstein Case Classification only uses the subject and description fields on closed cases to build a prediction model. Also, because Case Classification is designed to provide recommendations for new cases, once a field value has been selected for a specific case, Einstein Case Classification will no longer make recommendations for field values.
Einstein Case Classification creates and adjusts predictions based on the last six months of cases. ECC requires at least 1000 cases to be closed during the previous six months, but having around 10,000+ is the recommended amount. For all the closed cases, there must also be 100 closed cases that use each field and value being used in the prediction. It is also recommended there be less than 100 values for the field are predicted.
To build and manage an ECC model, a user is required to have all of the following permissions: Customize Application, Modify All Data, and Manage Profiles and Permission Sets.
The Einstein Case Classification User permission set is required for all users who will need access to Einstein Case Classification in your org’s service console.
The first step to building an Einstein Case Classification model is to search “Einstein Case Classification” in the quick find box from setup. Then click the toggle to enable Case Classification. The enablement can take a couple of minutes. To start a new prediction model, click the New button.
The user will be prompted to input a model name and an API name will automatically be generated from the model name. Next, the user will have an option to focus on a particular segment in the case dataset. Creating a model for a specific group of customers can help increase the accuracy of the model. When creating a segment, the user will need to select a case field, operator, type, and value. An example of segmentation would be if your company handles cases from all over the world, but you want to only focus the case classification model on cases from the United States. The segment would be defined as Case field Country that is equal to the US. Now, Einstein Case Classification will only build a prediction and display recommendations for Cases originating from the US. An essential consideration for segmentation is that the segmented dataset must meet the minimum count of records as listed above.
Another way to select specific records or a group of records is to use example sets. An example set is similar to using segmentation except that the example set does not filter what new cases will display Einstein Case Classification recommendations. Example sets only filter what records are used to build the model. A prebuilt example set for all case classification models is that the status of cases is equal to closed. An example of a user-defined example set could be if there is picklist value for a case record field that, when used, could introduce bias into the model. If we were continuing the example of the above of building a model for US cases, the user might want to build an example set that ignores records where the state is equal to Hawaii and Alaska because the user wants to specifically look at mainland US cases. Similar to use segmentation, the example set must also meet the minimum count of records as listed above.
Next, the user will select the field to predict. Up to 10 fields can be chosen for a single model, and all the fields available to choose are either picklist or checkbox fields on the case record.
Unlike other Einstein products, the user at this time is not able to select what fields the prediction is based on. All predictions are based entirely off the subject and description. Einstein Case Classification uses a text hash analysis to find patterns in the subject and description fields to power the model. By using a text hash method vs. natural language processing, Einstein Case Classification searches the subject and description fields for common words or themes. It quantifies how they are used rather than focusing on what the subject and description are fully conveying.
Finally, the user will hit Finish, and the model will be ready for building. Select the model name and then click the Build button. This can take a couple of hours, so building overnight is encouraged not to lose time waiting for the model.
Review Model/Activate Model
After the Einstein Case Classification model has successfully been built, it is ready for review. Select the model name and then select the Setup tab next to the Overview tab. Scrolling down, a table for the Fields to Predict is visible with an option on the right side to review under actions. This is where confidence can be set for each predicted field. The user also has the option to enable select best value and automate value.
The user can choose not to enable either Select Best Value or Automate Value. In this case, Einstein Case Classification will display the top 3 recommendations above the rest of the picklist values.
When select best Value is enabled, Einstein Case Classification will still display the top 3 recommended value but will also have an indicator for the top value of the 3. The user has the option to select the top value, any of the other top 3 values, or ignore the recommendations and select any additional picklist value. Select Best Value is best used over Automate Value when an agent holds information that currently is not in Salesforce. Assume an org is using Einstein Case Classification to predict how long it will take an agent to resolve a case, and an agent knows one particular account always runs long with case resolution times. The agent would be able to read Einstein recommendations on how long it will take to resolve the case and then manually select an amount of time slightly longer than the top prediction.
When Automate Value is enabled, the predicted field will automatically be populated based on the model’s best recommendation. Automate Value is best for orgs focusing on speed or that have high levels of trust in the prediction models. Automate Value will automatically populate a field on the case immediately upon case creation; there is no need to edit any fields. Using the example above for predicting the time to resolve a case, Automate Value would be the better option if agents have enough on their plate to resolve cases efficiently to efficiently resolve cases and shouldn’t spend time thinking about how much time is being forecasted to resolve.
On the Einstein Case Classification home page, under settings, for Automate Values As, a choice must be made between automated process user or custom Einstein user. The custom einstein user can be any user in the org. The Automated user is the user displayed in reports on automated field values and identify who made changes to them.
Confidence levels for Select Best Value and Automate Value can be between 50% and 100%. The confidence level impacts the percent of time Einstein Case Classification likely won’t select a value, percent of the time, it’s likely incorrect, and percent of the time, it’s likely correct. When selecting a confidence level, aim for a confidence level that predicts the correct value equal to or greater than 75% while minimizing both incorrect values and times Einstein Case Classification won’t select a value at all. If Select Base Value and Automate Value are both enabled at the same time, the confidence level for Automate Value must be greater than Select Best Value.
After reviewing the fields to predict, the model is ready for activation. Click on the button to Activate, and the model will be active and running almost immediately.
Einstein Case Classification recommendations can be viewed in the Service Console on Case records. First, create a new case detail component to update a record action layout. Add the fields to predict the layout along with any other fields an agent might update at the same time. Then open a Case page in the Lightning App Builder and edit the page. Select the case details component just created and add it to the page. When Einstein Case Classification has a recommendation, a prompt will appear at the top of the detail’s component that says, “Einstein Recommendations Available Einstein Case Classification recommendations can be viewed in the Service Console on Case records. First, create a new Case Detail component to update a record action layout. Add the fields to predict the layout along with any other fields an agent might update at the same time. Then open a Case page in the Lightning App Builder and edit the page. Select the Case Details component just created and add it to the page. When Einstein Case Classification has a recommendation, a prompt will appear at the top of the detail’s component that says, “Einstein Recommendations Available.”
After the activation of an Einstein Case Classification model, the performance dashboards become available to view the performance of predictions compared to the final field value saved in the case. One chart shows how effective the top recommendations are, and the other chart indicates how often one of the top three recommendations are selected as the final value. Only metrics for closed cases will appear in the dashboards. The dashboards are located by selecting a specific model in the Einstein Case Classification home page.
Einstein Case Classification is a powerful tool that can work alongside other Einstein tools. Einstein Prediction Builder is a tool that can be paired with case classification in different ways. Einstein Case Classification can be used to auto-populate case record fields, and an Einstein Prediction Builder can take the results of the automated values to supply scores or specific numbers. For example, since case classification can supply a picklist value, the picklist values for a given field can be a range of numbers. By using a prediction builder, the user can get a more accurate number from the range offered by case classification along with other case field values, and something case classification cannot do on its own. Another example is using a prediction builder to supply values to test and verify the results from case classification. This is useful when there is a lot of doubt that case subject and description alone will not produce valid results.