Predict When Your Customers Will Reorder
Marketing platforms are built to enable you to form close relationships with your customers in the same way as brick and mortar stores.
Imagine that you’re the owner of a hair salon. Although customer behavior is hard to predict, after a while you would figure out that people tend to come in once every six weeks.
Of course, there will be a lot of things that you won’t be able to predict. People might move away, change their hair style, or decide they want to try out another salon.
But some portion of your customers will come back on a regular basis. If you know who these people are and when you should expect them to return, you can reach out to them at the best time.
As someone running an online business with thousands to tens of millions of customers, you have many more customers than a small brick and mortar store. But wouldn’t it be nice if you could understand them just as well?
If you know when you should expect a customer to return, you can time your marketing efforts for when they’ll be most effective.
Two Things to Keep in Mind
- For most of your customers, the expected date of next order is based on the general purchasing patterns for all of your customers.
The expected date of next order is only personalized for a customer if they have made three or more purchases at regular intervals (we’ll go into more detail on how we determine if purchases are regular in a later section).
For these customers, you predict the next order using their previous purchase behavior. For everyone else, you predict the next order using order data across all of your customers.
- The expected date of next order tells us when a customer is most likely to order next, assuming that they will make another order.
- It does not tell us how likely the customer is to place another order.
In other words, the expected date of next order can tell you that given a customer’s past behavior, it’s better for you to wait 30 days to email them with an enticing offer to come back rather than 10 days or 60 days. It does not indicate whether they will actually purchase again.
If you want to see how likely or unlikely it is for a customer to make another purchase, you can look at the Churn Risk Prediction feature.
Where will you find the feature
Most email marketing platforms like Klaviyo, can calculate the expected date of next order for all of your customers, and update it every time you get a new customer or an existing customer makes another purchase.
Using the example of the Klaviyo platform (most other platforms will have a somewhat similar section), you can see a customer’s expected date of next order if you hover your mouse over the blue diamond on the timeline in the Predictive Analytics box.
Note: If you don’t see the Predictive Analytics box on a customer’s profile page, it means there is not have enough data for your company.
If you see a box but everything is gray and marked as n/a, it means the customer has not purchased from you before.
The Repeat Purchase Nurture Flow
When is the best time to reach out to a customer with an email incentivizing them to come back and buy again?
Waiting a month to email a customer with an explicit come-back offer after they make an order seems reasonable, but the email would arrive too early for a customer who orders once every two months, and too late for a customer who orders once every two weeks.
The expected date of next order feature gives you a reasonable date to work with on a per-customer basis, so that you no longer have to guess.
For Klaviyo email flows, the Repeat Purchase Nurture Series, uses the expected date of next order to prompt your customers to make another purchase at an appropriate time for them, based on their previous purchase behavior and the general purchasing patterns of all your customers.
A sophisticated marketing tool will have a Flow Library, like Klaviyo’s, along with many other prebuilt flows. The Repeat Purchase Nurture Series looks like this:
As you can see, the Repeat Purchase Nurture Series includes a Conditional Split.
One sequence of emails nurtures a first-time buyer to make a second purchase. The other sequence has slightly different content and encourages a repeat buyer to make another purchase.
Prompting Purchases Before the Expected Order Date
Let’s say that you own an online business selling pet products, and you want to be top of mind for your customers around the time that they’re likely to make another purchase.
Create a flow that emails the customer in the week leading up to their expected date of next order to prompt them to make a purchase:
Keep in mind that when counting down the expected date of next order, these emails will be sent to the same customer multiple times if they make repeat purchases. After a customer makes a purchase, a new expected date of next order is set for them and they are rescheduled in this flow.
If you decide to create this type of flow, we recommend using a Dynamic Product Feed in the email to vary the content each time. You can set up the dynamic feed to always show popular and trending items, or personalized recommendations. Here’s an example:
How the Feature Works
Let’s go back to our hair salon example. Over time, you would figure out that some customers come in on a regular basis. For these customers, you would use their pattern to predict when they will next come in. For first-time customers and people who don’t come in on a regular basis, you would predict when they’ll return based on how your overall customer base behaves. That’s pretty much how our expected date of next order predictions work. Let’s walk through the different cases in more detail using the Klaviyo platform as the example:
Customers with three or more orders, whose orders exhibit a pattern
For these customers, we use the average number of days between their three most recent orders to predict their next order. In the case of the example customer below, the time between the three most recent orders were 68 and 58 days, so we predict that their next order will occur in 63 days.
Customers whose orders don’t exhibit a pattern
For these customers, we make a reasonable prediction that can be used to time an email or other marketing activity, and we constrain the prediction so that it’s neither too short (e.g. next order will take place in two days) nor too long (e.g. next order will take place in two years).
We use a combination of the customer’s historical orders and order data from the company’s other customers to predict the customer’s next order. Specifically, we use the average of their historical days between orders, and we restrict this number to be between 50% and 150% of the median days between orders for all of the company’s orders. Here are two examples.
The customer below placed two orders that occurred 306 days (!) apart. It’s not very useful to predict that the customer’s next order will be in another 306 days, and two orders is not enough evidence to indicate that this will be the start of a pattern. So, we restrict our prediction to be no greater than 150% of the median days between orders for the company. The median days between orders for the company is 28 days, so we predict the next order for this customer will be 42 days after their most recent order.
As another example, the customer below placed four orders that don’t exhibit a pattern. Their orders are 7, 42, and 318 days apart. In this case, we first calculate their average days between orders, which is 122 days. Next, we calculate the median days between orders for the company, which is 55 days. Finally, we restrict our prediction to no less than 50% and no more than 150% of the median days between orders for the company, so we predict the next order will be 82 days after the most recent order.
Customers with one order
On the other end of the spectrum are the customers who have only placed one order. We have very little data on these customers, so we predict their next order using data from the company’s other customers. Specifically, we use the median value of the days between orders for all of the company’s orders. You can think of this as representing the date by which half of the customers would have placed their next order. The median value is similar to the average but is affected less by customers who either order very frequently (e.g. once every few days) and those who order very infrequently (e.g. once every few years).
In the example below, the customer has only placed one order. The median days between all orders for the company is 28 days, which means half of the customers in this company who place another order do so within 28 days. As a result, we predict that this customer will order around January 24, which is 28 days after their most recent order.
Expected Order Dates in the Past
Our prediction for the expected date of next order is made relative to the date of the most recent order. In other words, if we predict that the next order will take place three months after the most recent order, and the most recent order was one month ago, then the expected date of next order will be in two months. However, if the most recent order was one year ago, then the expected date of next order was nine months ago.
As a result, you will often come across customers whose expected date of next order is in the past, such as the example customer below.
It’s completely normal to have expected order dates in the past. This means the customer did not make another order when it was most likely for them to do so. We don’t recalculate the expected date of next order if it’s in the past because it means the customer is unlikely to return (note that the churn risk prediction is 96% for the above customer), which is useful to know. If you want to reach out to these customers with marketing content, you can use this information to help craft your message.
You will have more customers with expected order dates in the past if you have an older company with a longer order history. Because we calculate the expected date of next order for all customers who have placed an order, a longer order history means that you will probably have more customers who ordered many years ago and did not return.
Increasing CLV through predictive analysis
The above process can be applied to any marketing platform that offers email marketing automation. If your platform does not explore Klaviyo to see how we can help you elevate your retention marketing strategies.Back to Blog Home