Predictive Insights using Einstein Prediction Builder

Learn how Einstein for Nonprofits uses EPB to generate predictive insights for your org.

Einstein for Nonprofits helps admins generate predictive insights by using the Einstein Prediction Builder (EPB) bundled package. You can obtain predictions for the likelihood of each constituent becoming a first-time donor, recurring donor, or top donor by using built-in AI tools. EPB models are single-customer models. These models are trained using data from only one customer org.

EPB models data for all constituents, except Contacts that are set to Don't Profile. The Don't Profile field is located on the Individual record that the Contact is associated with and it ensures that the user's data is not used for future modeling.

Einstein for Nonprofits analyzes uses EPB to compute scores for these three predictions:

• Likelihood to become a first time donor: Provides fundraisers with insight about supporters who have no previous donations, but can possibly be converted to donors.

• Likelihood to become a recurring donor: Provides fundraisers with insight about supporters who can probably be converted to repeat donors.​ If a contact has a recurring donation record with Status "Active" and Recurring Type "Open", then that contact is a recurring donor.

• Likelihood to become a top donor: Provides fundraisers with insight about constituents who have donated enough over the past three years to move them past the top donor threshold.​ The top donor threshold is calculated as follows:

  1. Consider all contacts who donated two years ago. Among these, find the giving threshold for the top 25% of donors from two years ago and refer to this group as the "two years ago threshold".

  2. Consider all of the contacts who donated last year. Among these, find the giving threshold for the top 25% of donors from last year and refer to this group as the "last year threshold".

  3. ​​If any of these are true for each contact, then that contact is considered a top donor:

     • A contact's giving for the last two years is greater than the two year threshold. ​

     • A contact's giving this year is greater than the last year's threshold.

     • A contact's giving last year is greater than the last year's threshold.

These predictive metrics enable you to make smart decisions by quickly evaluating a prospect’s donation history or trends. Audiences can be tailored for a particular campaign by using the latest constituent data.

EPB trains multiple single-customer models with different configurations, and uses the configuration with the best model performance metrics. Your org’s models are trained by an automated process, not by a person. The set of fields that are used as model inputs are different for each predictive field. This set of input fields was selected by Salesforce.org after careful study of opt-in customer research and ethical data considerations. The automated process determines which of the set of model input fields to use in your org’s model, and what relative influence each field has on the model’s output.

EPB has data requirements for the org before it can generate predictions. EFP requires a minimum of 400 Contact records. Of these records, at least 100 must be for a first-time donor, recurring donor, or top donor, and at least 100 records must not be associated with any of these categories.

EPB jobs display a Model Failed status if they are unable to model the data set and generate scores due to insufficient data volumes or an application error. If EPB fails to generate a prediction, then the results from Backup Models appear on Contact cards instead. Ensure that Backup Models are enabled for your org. Otherwise your Contact cards might not show predictive insights.

EPB models are updated each month. The system retrains each predictive field to consider changes in the org data. Retraining the models ensures that you always have access to the most accurate insights. These changes in the org can result in a new way to generate individual scores at the Contact level, which are automatically reflected in custom predictive insights.

You will need to set up EPB to generate predictions and display insights.