Customer predictions are a useful way to anticipate what your customers are likely to do next, and when; this can help you to prepare appropriate strategies to keep your customers engaged and continue purchasing your products or services.
You can create segments that contain predictive blocks, which lets you filter your contacts into segments based on their predictions, such as when they will next order, or their churn risk.
Before you start
Things you need to know:
Create a segment with predictive blocks
The following predictive blocks are available to add to your segments:
Predicts the percentage chance that the customer will churn.
Churn is when a customer has stopped using your company's product or service during the current time frame.
Predicted CLV (Customer lifetime value)
Predicts the monetary revenue of the customer.
Next order date
Predicts the next date the customer will order.
Number of orders
Predicts the total number of orders the customer will make.
View customer predictions
You can view the customer predictions for individual contacts using the Single customer view. There are two areas that display the customer predictions for your contact:
The left side section of the Single customer view displays your contact’s current customer predictions and is a useful way to quickly see their upcoming interactions.
The summary at the top of the Single customer view plots your contact’s past and present customer predictions on a timeline. The timeline includes your contact’s orders, engagements, predicted orders, and the churn probability over time. You can use this timeline to measure how accurate the contact’s predicted orders were to their actual orders, how this correlates with their engagements, and when they were at their highest risk of churn.
Customer engagements represent when a customer is sent an email campaign and enrolled on a program.
Customer predictions explained
Predicted next order date
The model uses RFM and other machine learning techniques to make a prediction.
When there is a smaller buying history, such as one or two orders, our machine learning finds patterns in recency and monetary value. When the buying history is greater than two orders frequency is also introduced.
The below shows how the predicted next order date is calculated along with RFM data. The model looks at previous purchase dates which give a date range for the next purchase. A more specific date is calculated looking at an average of next purchase dates in a date range.
We can determine a collective purchase pattern for all of your customers in your account. For example, All contacts make a purchase every 16 days. If a contact does not purchase anything within that timeframe, their churn probability will increase from 1%, and continues to increase until they make a purchase.
The churn probability returns to 0% if the customer makes a purchase. See three different buyer scenarios below:
Predicted number of upcoming purchases
To predict a contact’s number of upcoming purchases, we break down their buying patterns, looking at their purchase history and how frequently they make those purchases.
Predicted customer lifetime value (CLV)
Driven by RFM values, the model looks at purchase history to predict the number of upcoming purchases (see predicted number of upcoming purchases). It considers events such as seasonal changes or sales events (Black Friday, discounts etc). The model then uses churn probability to calculate how much revenue they’ll generate over their lifetime.