Guide to Klaviyo's Predictive Analytics

read

Overview

Klaviyo applies a combination of data science and machine learning techniques to all of the data in your account to bring you useful and actionable insights. This guide covers the different types of predictive analytics data displayed in your account, the various ways Klaviyo computes this data, and guidelines for how you can utilize this data.

Please note that you will only see the Predictive Analytics section on profiles if you meet the following conditions:

  • At least 500 customers have placed an order.
    This does not refer to Active Profiles, but rather the number of people who have actually made a order with your business. If this section is on a profile but is blank, this means we don't have enough data on that individual to make a prediction.
  • You have an ecommerce integration (e.g. Shopify, BigCommerce, Magento) or use our API to send placed orders.
  • You have at least 180 days of order history and have orders within the last 30 days.
  • You have at least some customers who have placed 3 or more orders.

The Predictive Analytics Section of a Profile

Below is an example of the Predictive Analytics section of a contact's profile, and the information displayed:

unnamed.jpg

The table below defines the predictive analytics fields from the screenshot above. "CLV" stands for "Customer Lifetime Value."

Field Definition Example Value from Screenshot
Historic CLV

The total value of all previous orders an individual has made. The total number of orders is displayed below this value.

$1,178
(based on 17 orders)

Predicted CLV

A prediction of how much money a particular customer will spend in the next year. The total number of predicted orders is displayed below this value.

$363
(based on 5.24 predicted orders)
Total CLV The sum of Historic CLV and Predicted CLV. $1,541
Churn Risk Prediction The probability of a customer churning is based on their number and frequency of orders. Each time the customer makes an order, their churn probability goes down (green), but as time elapses between orders, the churn probability increases (red), with a medium churn risk represented in yellow. 11%
Average Time Between Orders The average of the number of days between each of a customer's orders. 62 days 
Predicted Gender Predicted gender is also a part of Klaviyo's predictive analytics features, however, this will not show in a customer profile   N/A

Along with the fields listed in the table above, the Predictive Analytics section also displays the order timeline.

Each tick on the timeline represents an individual order. You can hover over each tick to see the amount and date of the order.

2019-05-31_15-52-19.png

Ticks that are colored in represent an order that has been returned.

2019-05-31_15-52-34.png

The tick with a diamond represents the next expected order date for a customer. In the example screenshot below, this customer is expected to make an order on November 26th, 2019.

2019-05-31_15-52-47.png

How CLV Data is Calculated

Klaviyo automatically builds a CLV model using your company’s data and retrains the model at least once a week. For more details on how we calculate this model, check out the section on learning more below.

While no one can predict the future with absolute certainty, predictive analytics are a powerful tool for optimizing marketing spending and personalizing customer communication. However, predictions work best when averaged over many customers and are not expected to be exact for any single individual. While some individuals will spend more than their predicted CLV and some will spend less, as a whole they will average each other out.

For example, the Predicted CLV value displayed in the Predictive Analytics box is not an exact prediction. In some cases, you may see an impossible number of predicted orders. For example, you may see 1.43 as the number of predicted orders for a particular customer. 

When you see this, it means we expect the customer to make one or two orders, but there’s also a chance that they'll make more or fewer. These expectations start to make sense when you group multiple customers together because you can predict the total number of orders or spend for the group. If we have five customers with a predicted number of orders of 1.43, 0.25, 3.12, 0.78, and 2.97, we can expect approximately 9 orders across this group.

How Expected Order Date is Calculated

This prediction takes into account the specific customer’s order behavior and the order behavior of all your customers. If the customer’s orders exhibit a pattern, Klaviyo recognizes this pattern and will make a prediction based on it. If the customer’s orders don’t exhibit a pattern or if we don’t have enough data on the customer, Klaviyo will make a reasonable prediction based on how your other customers behave. Below are a few examples.

In this first case, the customer's orders exhibit a specific pattern. You can see that the distance between each order is roughly equal. Therefore, the next expected order is placed using the same distance.

Also, note that the expected next order takes place in the past. This, along with the predicted churn rate of 93%, is an indicator that this customer is not likely to place an order again in the future.

In the next example, the customer's orders do not display a specific pattern. In this case, the Klaviyo model makes a prediction based on the ordering habits of your other customers.

FAQ about Repeat Purchase Nurture Series Flow

Do I need to back-populate existing profiles into this flow? Do I need to tell the flow to populate with all possible profiles moving forward?

You don’t need to back-populate existing profiles or tell the flow which profiles to include, it will figure this out for you. Every customer who places an order with you has an “expected date of next order."

I saw the flow has a conditional split. How do we know the expected date for one-time purchasers?

For one-time purchasers, since we don’t know much about their purchase behavior, we calculate their expected date of next order using data across all of your customers.

Our brand has 3 classes of product frequency. For some products, customers come back randomly. Others are replenished between 60 and 90 days. The last group is replenished between 100 and 120 days. Can we teach the app to know what the customer purchased and send reminders based on the product?

The app doesn’t consider what products the customer ordered, so if you have products with distinct replenishment cycles, we would recommend instead creating multiple Placed Order triggered flows for each replenishment cycle — using Trigger Filters to restrict each flow to products that share the same cycle — and then using a Time Delay that reflects the known cycle so you know you’re sending replenishment emails at just the right time. Since the predicted date won’t consider what the customer last ordered and also the likely replenishment cycle for that product, if a customer has known replenishment cycles for most of their product categories they should stick with a standard Repeat Purchase Nurture Series Flow and not use this feature.

Things to Look Out for in the Repeat Purchase Nurture Series Flow

  • We don’t recommend counting down to the expected date of next order as repeat customers will simply get the same sequence of emails leading up to every order which may result in unsubscribes.
  • This flow shouldn’t replace the use of replenishment flows if customers know the general cycles for most of their product categories.
  • If you do have a high percentage of repeat buyers, you may want to consider only using this feature for customers that have purchased one time to nurture them for their second purchase.

How Predicted Gender is Calculated 

Klaviyo’s gender prediction algorithm uses a customer’s first name along with census data to make a gender prediction of either likely male, likely female, or uncertain.

Because predicted gender is still an approximation, you’ll want to make sure that, when using targeted communication, you include some information for both genders.

Gender_Emails.png

Learn more

Interested in doing more with your predictive analytics, or learning more about how they are calculated? Check out the resources listed below.

For information on how to segment by CLV and export this data, check out our help article.

For more information on how predictive analytics are calculated, check out our blog post.

For an interview with our VP of Data Science, check out this video.

x
Was this article helpful?
13 out of 16 found this helpful