Leverage Einstein Prediction Builder For Your Association’s Next Membership Campaign

For associations, this serves as a great benefit, as a lot of the business questions associations ask are related to membership, certifications and events. Being able to quickly deploy models and address business questions can greatly improve your associations’ ability to attract and retain members.

Requirements for the Dataset / DataObject

For Einstein Prediction Builder to work, you need a well structured dataset that contains multiple variables and one predictive outcome. The prediction outcomes that Einstein Prediction Builder can generate is limited to a Yes/No answer or a Numerical Value. For instance, if you want to predict how much revenue a contact is likely to generate (customer lifetime value), you can set your total sales as your numerical outcome. If you want to predict which contacts are most likely to sign up for membership, you can set whether a person signed up or not as your Yes/No outcome.

The spreadsheet image below contains mock data we used to build our membership predictions, which also serves as a good example of how Einstein Prediction Builder looks at data. In the dataset, the green column, column P, contains Yes/No entries, this is our known outcome of whether someone signed up for memberships or not. We’ll be using most of the data from columns C to O to predict the unknown cells in Column P.

The general idea is to feed Einstein Prediction Builder training data that contains some of the existing outcomes, in order to predict the remaining unknown outcomes.

Do note that our known Yes/No data points need to be at least 400 entries, approximately 50% positive and 50% negative. Einstein Prediction Builder will eventually use the example data to train itself to build a suitable model to populate the prediction (unknown) data.

Building your Prediction Model

To start building the model, we will tell Einstein Prediction Builder that we are predicting Yes/No outcome and choose “No Field” as shown below so we can split the example and prediction data easily.

We will then tell Einstein which records we’d be using for “Yes”/”No” examples, in our case, it’s column P outcome_term_subscription that we mentioned above in our prerequisites. Using the Data Checker on the right hand side, we can see how big our examples are and how many records we are trying to predict at the moment.

After that, we choose which columns to contribute to your predictions, making sure we do not include any unnecessary features such as who created the record in salesforce and the email address of the contact. Just to reiterate, we are using age, income, contact_times, education to predict whether a person will sign up for membership or not.

Finally, we can click through the remaining pages and land on the status page for our model. Depending on how much data you are using to train your model, the time it takes to generate a model could range from 30 minutes to 24 hours.

Once the model is completed, we are able to obtain a scorecard on how well the model did. In our example, we can see that our prediction quality is around 74 points, which means that our model will be 74% accurate in predicting future outcomes. The top predictor side bar graph shows that the variables that most impact the model include: whether a person signed up for previous memberships or not, whether they have a housing loan and how many times they were contacted this year.

To dig into more details, we can click on the predictors tab and see that housing_loan field has a strong impact but negative correlation, which means having a housing loan negatively impacts our model and the person is less likely to sign up for premium membership if they have a housing loan.

The other predictors such as prev_outcome – success, contact_times all have positive correlation and impacts over 0.50, which means they contribute to our outcome predictions positively and people that signed up last term and contacted multiple times are likely to sign up again.

Moving on, we want to get a list of these members so that we can reach out to them. In order to do so, we move to Salesforce’s front page and create a list based on the premium_membership object.

We proceed to filter the list by unknown outcomes and include contact emails with high impact variables. We also include the prediction results (score). This will be blank when first added, but will populate after about ten minutes or so.

After waiting for our results, we refresh our Salesforce page and filter by highest prediction_results. Prediction results show the likelihood of a successful sign up. We can see that the first user is most likely to sign up with a score of 91. We should reach out to him to see if he is interested in signing up for membership.

Monitoring your Prediction Model

So far, we have built and enabled our prediction model and created a list of users who are most likely to sign up for premium memberships. As new data comes in, our model will continue to make predictions. Based on this new data, the models performance could change. Thus, it is recommended to check periodically how well the model is doing by comparing the prediction results with the actual results.

Usually, this check is done by looking at the true/false positive predictions of the model over a certain period of time. An example of this is provided below:

Stay Tuned For More To See How fusionSpan Can Help

Overall, you’ve seen how you make a prediction with Einstein Prediction Builder and single out high value targets within your association for your next membership campaign. As mentioned before, this is a hassle free, no code required application that allows you to make predictions with Salesforce objects.

For those using the Fonteva platform, this will be particularly useful. Users will be able to quickly deploy models, comprehensively view how well the model performed and create dashboard visuals around the prediction outcomes.

Of course, that is not the only tool your association can leverage. Apart from Einstein Prediction Builder, there’s also Einstein Discovery; Discovery is another tool on the Einstein platform that can automatically analyze millions of data combinations in minutes. Einstein Discovery can help associations better understand who their customers are and why they do what they do.

If you would like to know more about Einstein Prediction Builder, stay tuned for our upcoming articles diving deeper into the tool. In the meantime, reach out to fusionSpan for help setting up your next membership campaign for success!