Introduction

Predictions help anticipate likely problems before they occur so that proactive actions can be taken in time.  CRM Admins and users with access can add new predictions for any field in a module, using the Prediction AI Designer

Prediction Types & Use Case

1. Classification Prediction: This is useful for predicting one of the predefined values.  
   1. Will the customer renew? (Yes/No)
   2. Will the invoice be paid on time? (Yes/No)
   3. Will the task be completed by the due date?
   4. Will the Case be resolved within SLA time?
2. Continuous Variable Prediction: 
   1. Predict the task completion date
   2. Who should the deal be assigned to?
   3. Who should the lead be assigned to?

Enabling Prediction AI

• Install Calculus AI add-on

Prerequisites for a Prediction

• Minimum number of records - At least 50 historical records are required to make a prediction. Accuracy will be higher with more records.
• Actual values of predicted field in historical records - These historical records should have the actual values for the field to be predicted, either directly in a field or should be derivable from other fields. 

Adding a Prediction

Follow these steps to add a prediction:

1. Log in to the CRM.
2. Click the Main Menu.
3. Under Platform > open Prediction AI Designer.
4. Click Add Prediction.
5. Select the Module.
6. Select the Category. 
   1. Data classification - Predicting the category. For example, Case SLA Prediction. 
   2. Data prediction - Predicting the continuous values. For example, Task completion date.
7. Default algorithm (Advanced) - The Algorithm that is used to predict is selected by default and can be changed. 
8. Parameters - Select the parameters that you feel will influence the value. All the module fields will be shown 
   1. Note: Some of the fields in the record will influence the predicted value (for example, task size, task owner, and task start date will influence task completion date). Select whichever fields might influence the prediction.
9. Is the value to predict available in a field for historical records?
   1. If Yes - Select the field for prediction 
   2. If No - You can specify the expression to compute the predicted value. For example, predicting a Task's due date. As you directly cannot predict the task due date, you have to find the duration first, then add a formula saying that Task duration = task due date  - task start date.
10. After adding the Prediction, it will be queued. The data will then be uploaded, and the training will be completed. 
11. Click Manage.
12. Click Activate to start the predictions
    1. Note: Only Admins can perform the above steps.

Admins can also view Outliers in the data, exclude them from the training model, and retrain to improve predictions. Outliers might confuse the training model and lead to wrong predictions and hence should be excluded.