Einstein Classification Key Concepts

Classification

In machine learning, classification is the ability to predict labels based on underlying data and data patterns. In Einstein Classification Apps, Einstein learns from your closed-case data and “classifies” field values based on that data.

• For Einstein Case Classification, Einstein makes recommendations right after the case is created.
• For Einstein Case Wrap-Up and Einstein Voice Wrap-Up, chat service reps see them on-demand or when the conversation ends.

Model

At the heart of Einstein Case Classification and Wrap-Up are classification models. Models are trained on the closed cases' Subject and Description as well as the fields to be predicted. Einstein uses natural language processing (NLP) to identify data patterns in each model's closed cases.

If you have the Service Cloud Einstein add-on license, you can create up to five models for each classification app and customize them for different parts of your business. The Try Einstein version lets you create one model for each app.

Segments and Example Cases

When you set up a model, you get to decide what types of cases to focus on. Segments and example cases are optional filters that you can apply to your closed case data. They let you limit the closed cases that Einstein learns from and determine which cases get predictions.

If you want, you can specify filter criteria that narrows a model’s scope to a subset, or segment, of cases. Einstein learns from closed cases in the subset and makes predictions on new cases that match your segment criteria.

Use segments to focus on cases in a particular business unit. For example, use a segment to predict field values on cases in your Enterprise division. In another model, you can define a segment that represents your Consumer division’s cases. Since Einstein learns from the words that customers use to contact to you, segments can provide helpful context for Einstein's predictions.

If you want specific closed cases to serve as examples, define filter criteria that identify example cases. Einstein learns from closed cases that meet your example criteria. If you define a segment and example case criteria, your example cases are a subset of your segment.

You can use segments and example cases together to focus on certain types of cases and filter out low-quality data from your model. For example, in your Enterprise division segment, exclude cases with a certain record type or cases that use obsolete picklist field values. Keep in mind that cases from your segment and example filters must meet the minimum data requirements of 400 closed cases created in the past six months. If this requirement isn't met with cases where the Subject field is not empty, the rule is relaxed to include cases where the Subject is NULL.

Here’s a summary of these rules.

Classification Automation and Confidence

You can choose to automate field predictions. During setup, you choose case fields for Einstein to predict on cases. When you build your model, you can customize prediction preferences for each field. You set percentage-based prediction confidence thresholds to control when Einstein recommends and selects field values. For Einstein Case Classification, you can also set a threshold for when Einstein saves field values automatically.

The more automated the action, the higher the prediction confidence needed. If a prediction doesn’t meet one option’s threshold, Einstein tries the next, less automated option.

• Recommend Top Values (less automation): By default, Einstein recommends the top three field values. The service rep selects and saves a value. A confidence threshold isn’t needed.
• Select Best Value (more automation): When the prediction meets your confidence threshold, Einstein shows the field with the best value already selected. The service rep confirms and saves the value.
• Automate Value (full automation): When a prediction meets your confidence threshold, Einstein saves the best value for the field—no service rep review is needed. You can have Einstein Case Routing run your existing case assignment rules on auto-updated cases. Available only in the paid version of Einstein Case Classification.

For each field in an Einstein Case Classification model, you can turn on Select Best Value, Automate Value, or both, or use the default top three recommendations. For an Einstein Case Wrap-Up model, you can turn on Select Best Value or show the default top three recommendations for each field. You can update prediction settings at any time without rebuilding your model.