Set up Escalation Predictor to Preempt Tickets

Support agents usually develop a gut feeling for cases that are quick to solve and cases that carry with them a risk of escalation. Such agents are worth their weight in gold. If you are a support manager and find such an agent, hire them right away. But if you cannot find such an agent, try Escalation Predictor.

Like an old hand, Escalation Predictor forewarns when a case is about to go south so you can take appropriate action to thwart an escalation.


  • Escalation Predictor works only on cases which have the following fields indexed: Owner Name, Contact Name, ID, Case Number, Created Date, Closed Date, Contact Email, Last Modified Date, Case Origin, Escalation Field, Subject, Status, Description, and Priority.
  • Before connecting your case management system with Escalation Predictor, ensure that the system has at least one escalated case.


  1. Click Marketplace from main navigation, and then Install New Marketplace Addon.

  2. Install Escalation Predictor.

Input Data

Escalation Prediction needs to access cases before it can start analyzing them. In this section, you will learn to connect the app with a case management system.

  1. Go to Marketplace and open Escalation Predictor. NOTEClosedAn admin is notified when the refresh or access token has expired.
  2. Then click Connect with a Content Source.

  3. Escalation Predictor analyzes the historical cases in your system in order to predict escalations. Select a content source.


    As of November, 2020, only Salesforce supported.

  4. To establish a correlation between cases and escalations, SearchUnify uses the IsEscalated case field or a custom field set up by Salesforce admins. Select that field and set its value to True for IsEscalated or a value that represents an escalation in your custom field.

  5. Escalation Predictor uses an advanced model which trains on cases' data. The more the data, the more accurate the model. With Start Date, you decide how much data to keep. Pick a date and click Next. Only the tickets logged after Start Date are analyzed.

Transform Data

Escalation Predictor trains on cases to find those factors that lead to escalations. Depending on your case management system, the factors can easily run into dozens. The factors frequently include a case's title, a product's category, and the purchase amount. All of these factors (or case fields) are put into one of the three available categories in the app: Textual (used for sentiment analysis), Categorical, and Numerical.

  1. Drag all the text fields, such as Subject, Case ID, Description, and Case Number into Textual. The Categorical field is for fields like Priority, Created Date, Closed Date, Contact Email, Contact Name, Owner Name. Escalation field, Last Modified Date, Case Origin, and Category. Use the Numerical field for storing number of comments exchanged, purchase amount, and other fields with numbers. NOTE. Case number is always a Textual field and both Closed Date and Created Date are always Categorical fields.

The data after connecting Salesforce with Escalation Predictor.

Classification Reports

The Classification Reports tab appears when you click Start Classification in Input Data.

Based on up to two months old analysis of data in Textual, Categorical Measures and Numerical, Classification Reports generates visual reports. If there isn't sufficient case data, then you will not see any reports. When the reports are available, review them to spot patterns. The keyword cloud based on an analysis of case messages help you identify the triggers which suggest an imminent escalation.

The analysis run on cases today might or might not be relevant after six months. It's recommended to click Refresh, each time the case volume increases. For instance, you can train Escalation Predictor when there are 100,000 cases in the database, then again when the number increases to 110,000.

Scroll down to view them all.

Scoring Data

In Scoring Data, the metrics to be used in forecasting are picked. A rule of the thumb is to check the metrics with high correlation and uncheck the ones with low correlation and then click Retrain Model.

Example: In the next image the most accurate predictor of escalation is Days to Respond, which stands at 21%. As the Days to Respond number goes up, so does the probability of an escalation. It's better to keep it checked. However, case fields that have little correlation with escalations can be removed. In the next image, Solutions Area can be unchecked.

Last updatedWednesday, March 29, 2023

Or, send us your review at