Using Predictive Analytics Webinar

Last updated at:


This is a recording of the webinar about how to use predictive analytics in your account. In this webinar, we covered:

  • How Predictive Analytics are Calculated
  • How to Use Predictive Analytics in Klaviyo
  • Smart Send Time
  • Live Q&A with One of Klaviyo's Data Scientists



Additional Resources


I don’t see Expected Date of Next Order in the predictive analytics. Where can I find it? Where can I find this value in a segment?

If you head into a particular customer's profile, in the predictive analytics box, the diamond represents the customer's expected date of next order. To segment based on expected date of next order, select that you'd like to segment based on properties about someone and choose expected date of next order from the dropdown.

Can you clarify the lifetime value number? This is what they are projected to spend in the next 12 months?

Yes, predicted customer lifetime value refers to what a customer is projected to spend with you in the next year. Historic customer lifetime value refers to what a customer has spent with you in the past. Total customer lifetime value is a combination of the two numbers.

Where can I find the Repeat Nurture Series flow?

Head to Flows > Browse Ideas and search for it within the flow library.

How many orders/subscribers/historical data is ideal for the algorithms to be accurate? I know you mentioned the minimum numbers but what are the ideal numbers?

As long as you have the values for predictive analytics to be enabled in your account, the model should work with accuracy; however, the more data in your account the better.

If you build a segment of CLV > 500, does that mean none of the customers currently have a $500 in purchases? If a customer then purchases $500 in an order, do they move out of the segment? 

If you build a segment that targets people with a total CLV of greater than 500, that means that all customers who have a combined historic and predictive CLV of greater than 500 will be included in the segment. If they spend more than 500, they will continue to be in the segment even if they do not make another purchase because their historic CLV will be at least 500.

Is there a way to incorporate secondary orders that come in through an API (ie google sheets via zapier) in predictive analytics?

Not currently, but it is on our radar.

How could we use this data for projecting impulse buyers?

You don't need predictive analytics to target impulse buyers. You can, however, create a segment of customers who have a short window between opening an email and purchasing from you and further personalize your messaging to that group.

If you have a pop up from another app will Klaviyo's come up instead or in addition? 

Klaviyo's popups will come in addition to another 3rd party app popups. Make sure to disable one if you don't want customers to receive double the messaging.

This feature has been around for a while. How accurate the predicted date of next order is? What is the percentage of orders placed exactly on the predicted date?

Does Klaviyo recognize historic purchase data from ecommerce platforms or just the data that Klaviyo has from the start of using it?

If you've integrated with your ecommerce store, Klaviyo will use all of the historic data that you have in the algorithm. If you're coming from another email service provider, make sure you integrate with that platform as well so you will have engagement information for your customers to use.

Any case studies you can share on customer success stories with using predictive analytics?

All of the suggestions in Best Practices for Using Customer Lifetime Value (CLV) Data and Best Practices for Expected Date of Next Order come from customer success using the predictive analytics features.

For predictive analytics, are customers calculated by people who buy? So if you have over 500 customers but 400 customers that placed an order. I would not see that?

No, you would need to have over 500 customers who have purchased. You can find more information about the requirements for predictive analytics to be included in your account in our article on Guide to Klaviyo's Predictive Analytics.

Does the machine learning component of predictive analytics take into account seasonal sales variances? Or is it projecting sales upon a more blanket purchase frequency?

It currently does not look at seasonal variations but is on our radar.

How frequently are the analytics numbers updated? Daily/Hourly/Real-time?

They are updated in real time. 

I have a couple thousand of customers that placed an order but barely any are repeat customers. Does it make sense to have predictive analytics?

Yes! If you meet the criteria predictive analytics will automatically be included in your account. You can try to turn on the Repeat Nurture Series flow to drive one time shoppers to repeat buyers.

Should we avoid taking the mean of predictive metrics (due to these metrics having a skewed distribution)?

It depends on the type of metric you are calculating. Mean values give good insight into business-wide trends.

How would you recommend collecting gender information for a gender-neutral product line (i.e. kitchenware, plants?)

If you want to feature more gendered lifestyle content for your gender-neutral products, use the predicted gender feature in your campaign and flow sends. You can always use a signup form asking your customers for their gender information for those who the model determined to be "uncertain."

Will there ever be a way to exclude a SKU from predictive analytics? For example replacement SKUs to not affect any predictive data?

We currently don't offer this feature but is on our radar.

Will predictive metrics be available within dashboards soon?

We currently don't offer this feature but is on our radar.

What size audience would you say is necessary to really get value from predictive analytics over crafted campaigns?

So long as you have enough information in your account, the model will work with optimal accuracy.

What would you consider to be essential behavioral segmentation to implement immediately?

Expected date of next order, high/low-value CLV, and gender.

You show predicted date and amount of next order is there any way you can show the actual date and amount of the actual next order when it happened?

We currently don't offer this feature but is on our radar.

I think the biggest thing is boiling it down to a pragmatic list of top 3 steps and things to do to implement to get the most out of this.

Turn your repeat nurture series live, use the template from the Best Practices for CLV to look at the future health of your business, create a lookalike audience on Facebook from your high CLV customer segment and target them with an ad.

For the Exploratory Send, are you saying the more you do that, the better the focused send time becomes?

Yes. When it comes to data, the more the better. 

Do you have a spot where case studies are shown?

The Best Practices for CLV and Expected Date of Next Order articles both draw from customer examples. Additionally, head to the blog for more ideas from our customers.

Is churn risk something that can be used for a flow yet?

It is not. However, we wouldn't recommend using churn risk for flows. Instead, use expected date of next order for a more personalized and accurate depiction if someone has lapsed in their purchasing cadence. 

Is that data for send time based on open rates or conversions?

It is based on open rates.

Can you change the time period for predictive models? (i.e. predicted CLV over the next year vs next 7 days?)

We currently don't offer this feature but is on our radar.

Once someone passes their expected order date without ordering, does that date still stay the same in the past?

It stays the same so that you can continue to target those customers who have lapsed.

Do you guys have a general idea of how accurate predictive next order is on average overall Klaviyo accounts?

The accuracy varies slightly by company so if you are interested in knowing your model's performance please email

Is there a way to use the Expected Next Order date with a mix of one-time orders and subscription-based orders?

If you add in a custom property for your subscribers versus your one time buyers, you can trigger different repeat nurture series depending on the customer's buying preferences. For more information about custom properties, check out our article on Add Custom Properties to a Contact Profile.

Can you use the smart send time in the flows as well?

We currently don't offer this feature but is on our radar.

I don’t see ‘Expected Date of Next Order’ under properties. Does this mean we have not hit the data threshold?

Expected Date of Next Order will be under the section Predictive Analytics About Someone in the segmentation dropdown.



Was this article helpful?
18 out of 32 found this helpful