Create Segments Using Customer Lifetime Value

Editor’s Note: This post is co-authored by Ezra Freedman, Christina Jaworsky, and Eric Silberstein.

Our AI-powered Customer Lifetime Value feature, which we launched six weeks ago, makes predictions for individual customers.

Today we’re launching the ability to build segments using these predictions. Now, you can take advantage of our algorithms to target your customers.

Here’s what Customer Lifetime Value looks like on an individual customer profile:

You can now build segments with the following per-customer dimensions:

Historic Customer Lifetime Value

Total spend on all past orders

Predicted Customer Lifetime Value

Predicted spend over the next year

Total Customer Lifetime Value

Historic CLV + Predicted CLV

Predicted Number Of Orders

Predicted number of orders over the next year

Average Days Between Orders

Average days between past orders for customers with two or more orders

To use these new dimensions, select “Predictive analytics about someone” in the segment definition dropdown.

Example

We’re sad that summer is coming to an end (at least here in the northern hemisphere), but it’s a great time to run a summer clearance campaign. We’re going to use aggressive discounting and target customers who have a low spending prediction but are engaged with our emails.

Let’s say you’re running a cosmetic brand, and the average order value is $15. If you want to identify low spenders, you should look for customers who are predicted to spend no more than $5 in the next year using predicted CLV. This allows you to isolate customers who are unlikely to spend much with you in the next year.

Here’s why that works. Let’s say you have 100 customers with a predicted CLV of $1 each. No single customer is actually going to spend $1 because you don’t have products priced that low. However, you can expect the 100 customers to collectively spend $100 (100 x $1) which could consist of four $15 orders and two $20 orders. So a $5 cap on predicted CLV for the segment is actually comfortably high and should find many customers for your summer clearance campaign.

Add the first condition using the new “Predictive analytics about someone” dropdown and add a second condition that the customer has opened at least one email in the last ninety days.

Predictions work best for groups of customers

Predictions are most accurate for groups. Customers rarely spend exactly what we predict. However, groups of customers are generally well predicted and underspending by some customers is balanced out by overspending by others.

Think about timing

Predictions update in near real-time. After a customer places an order, their predicted CLV, total CLV, predicted number of orders and churn risk will be updated. This means customers may enter or leave segments shortly after a purchase, just as with other types of Klaviyo segments. Take this into account when triggering flows using segments.

For example, if you create a segment based on a low predicted number of orders and your business tends to have many one-and-done customers, then brand new customers will immediately enter that segment. You probably don’t want to trigger an email immediately after a customer’s first purchase based on a prediction that the customer won’t buy again. Use additional conditions in segments and/or time delays in flows to achieve your desired timing.

Where is Churn Risk Prediction?

We predict the probability of churn for customers but you can’t segment customers based on it. We’re working on an intuitive way to use churn risk in segments. In the meantime, you can export that information.

Feedback

If you have ideas or want to share your thoughts about our work, our data science team would love to hear from you. Please send feedback to datascience@klaviyo.com.

Also read about our new model improvements and ability to export customer lifetime value information.

Back to Blog Home
Get email marketing insights delivered straight to your inbox.