User Predictions

You can use each of these predictions in the Prediction Manager interface, in Zephyr code to produce email or SPM template logic, in Audience Builder to dynamically segment your audience, and on individual User Profiles.

Improvements

We have made significant improvements that make the Prediction Manager update more frequently. 
  • Predictions are up-to-date with the latest user behavior. Latest user behavior is collected in real-time, but it can take 24 hours for it to be reflected in predictions. 
  • Since predictions use the latest user behavior, they are now more accurate.

Below the list of predictions, you will find key use cases and examples for each, including implementation steps, to help you get started.

Name

Description

Code

Recommended Value Type

Type

[7] Order Value

The expected order value if a purchase occurs within 7 days

aov_7

Number

Revenue

[1] Purchase

The probability of making a purchase within 1 day

purchase_1

K-Tile

Revenue

[7] Purchase

The probability of making any purchase within 7 days

purchase_7

K-Tile

Revenue

[30] Purchase

The probability of making any purchase within 30 days

purchase_30

K-Tile

Revenue

[30] Revenue

The expected revenue from this user within 30 days

rev_30

K-Tile and Number

Revenue

[7] Per Item Value

The predicted value of an individual item in a purchase order if a purchase is made in the next 7 days

aiv_7

Number

Revenue

[7] # of Items Purchased

If a user makes a purchase in the next 7 days, the number of items in their purchase basket

item_7

Number

Revenue

[7] Open Rate

The probability of opening a message within 7 days

openrate_7

K-Tile

LTV

[7] Click

The probability of clicking a link in a message within 7 days

click_7

K-Tile

LTV

[7] Opt Out

The probability of opting out within 7 days

optout_7

K-Tile

LTV

[30] Pageviews

The expected pageviews within 30 days

pv_30

K-Tile and Number

LTV

In the Zephyr code of your email or SPM template, you can access the current user's value for any prediction using a variable name matching the following format: profile.vars.predictions.<code>.<num_or_ktile>.

For example:

  • profile.vars.predictions.aov_7.ktile
  • profile.vars.predictions.pv_30.num

User Prediction Use Cases

A common theme among the top use cases for predictions is that they are useful for segmenting your users based on their engagement and value, enabling you to treat those segments differently in order to increase customer lifetime value and make your marketing efforts as efficient as possible.

You'll note that there are a variety of ways to identify these characteristics among your user base. While each user prediction has recommended uses below, in many cases, you'll want to A/B test your strategy to determine what is most effective for you. For example, you could test multiple prediction value or k-tile thresholds, or compare tactics involving predictions of purchases over the next 7 days vs. 30 days, to determine what works best. In the A/B test section, learn how to confirm the tactics that are most effective for your unique user base before you choose which ones to implement.

In addition to these use cases and instructions, listed for each prediction is the Zephyr code you would use to access the prediction's value for the current user within an email template or SPM template. If you have any questions about these instructions or how to adapt them to your needs, please contact Support or your Customer Success Manager.

Purchase

(purchase_1, purchase_7, purchase_30)

Likelihood that a user makes a purchase in the next 24 hours, 7 days, or 30 days.

Ideal for optimizing messaging to maximize purchase conversions.

  • Key use case: Identify your users with the highest intent to purchase, and concentrate your efforts on them because their purchases account for the vast majority of your revenue. Meanwhile, offer larger incentives to all other users to make a purchase.
  • Example: For all users, except the 1% who are most likely to make a purchase in the next 30 days, include a free shipping coupon or banner in their email or while they're on-site. How-To
    In your template code (in email or SPM), include the following snippet:
    {if profile.vars.predictions.purchase_30.ktile >= 900}
    <img src="discount_banner.png">
    {else}
    <img src="regular_banner.png">
    {/if}
  • Zephyr Variables:
    • profile.vars.predictions.purchase_1.ktile (recommended)
    • profile.vars.predictions.purchase_1.num

Order Value

(aov_7)

Predicted value (average order value) of any purchase made in the next 7 days.

  • Key use case: Increase AOV by offering tiered discounts that push users higher up their price point range.
  • Example: For users with an AOV below $50, display a minimum order value of $75 for them to qualify for a discount. For users with an AOV of $50-$100, display a minimum order value of $125 for them to qualify for a discount. For users with an AOV of $100-$175, display a minimum order value of $200 for them to qualify for a discount. For users with higher AOVs, display no discount. How-To
    In your template code (in email or SPM), including the following Zephyr snippet:
    {if profile.vars.predictions.aov_7.num < 5000}
    Get $20 off when you spend $75 or more!
    {else if profile.vars.predictions.aov_7.num < 10000}
    Get $30 off when you spend $125 or more!
    {else if profile.vars.predictions.aov_7.num < 17500}
    Get $40 off when you spend $200 or more!
    {/if}
  • Zephyr Variables:
    • profile.vars.predictions.aov_7.ktile
    • profile.vars.predictions.aov_7.num (recommended)

Revenue

(rev_30)

Predicted revenue from a user in the next 30 or 365 days (can be used as a proxy for customer lifetime value). It is a function of the likelihood that the user will make purchases and the predicted order value of those purchases. It can be used to identify and target your highest-value users.

  • Key use case 1: After identifying your highest-value users based on rev_30 , find and acquire other potentially high-value users leveraging thes Facebook integration, to optimize your ad spend and acquire higher-value customers.
  • Example: Create a Smart List that contains the top 1% of revenue-generating users and push the list to Facebook to create a Lookalike Audience.
    How-To
    1. In Audience Builder, create a Smart List using the criteria that rev_30 is greater than a certain number (e.g. the top 1%: a k-tile exceeding 990).
    2. On your Lists page, click the cloud-upload button to send the list to Facebook. Note: This requires that you configure the Facebook Custom Audiences integration.
  • Key use case 2: In email or SPM, for only your highest-value users, display a call-to-action to join your loyalty program.
  • Example: Use a conditional statement in your template code, containing text that is only displayed to users who fall in the 990th rev_30 k-tile and above.How-To
    • In email or SPM, include the following Zephyr snippet:
      {if profile.vars.predictions.rev_30.ktile > 990}
      CALL TO ACTION!
      {/if}
  • Zephyr Variables:
    • profile.vars.predictions.rev_30.ktile
    • profile.vars.predictions.rev_30.num
  • Optout

    (optout_7)

    The likelihood that a user opts out in the next 7 days.

    • Key use case: Lower your email optout rate for certain campaigns, suppressing users who are likely to opt out from receiving emails.
    • Example: Create a Smart List containing the 1% of your user base that is the most likely to opt out. Then, prevent them from receiving certain campaigns and consider reaching them through other channels. How-To
      1. In Audience Builder, create Smart List using the criteria: optout_7 is greater than, for example, the 990 k-tile (identifying the top 1% who are most likely to opt out).
      2. When building a campaign, select this Smart List as your suppression list. If it is a recurring campaign, it will continually suppress the dynamic list of users who are most likely to opt out. If you're sending multiple campaigns, this is an ideal tactic for less-vital campaigns among them, as some users will not receive the campaign.
      3. Optional: You may choose to target these users through other channels, such as mobile or social media.
    • Zephyr Variables:
      • profile.vars.optout_7.ktile (recommended)
      • profile.vars.optout_7.num

    Openrate

    (openrate_7)

    The probability that a user opens a message in the next 7 days

    Ideal for optimizing messaging for to maximize awareness.

    • Key use case 1: Target users who are highly likely to engage with your email, in order to boost all key email KPIs (clickthroughs, PPM, etc.), and avoid causing message fatigue, promoting deliverability.  How-To
      1. In Audience Builder, create a Smart List using the criteria 'openrate_7 is greater than k-tile 900', or any other high k-tile threshold of your choice.
      2. When building your campaign, choose to send it to this list.
    • Key use case 2: 
    • In SPM, target users who are least likely to open a message, and offer a call-to-action for them to view another channel (e.g. app download). 
    • How-To
      • Include the following Zephyr code:
        {if profile.vars.predictions.openrate_7.ktile > 900}
        <App download call to action and link>
        {/if}
  • Zephyr Variables:
    • profile.vars.predictions.openrate_7.ktile (recommended)
    • profile.vars.predictions.openrate_7.num
  • Click

    (click_7)

    The probability a user clicks a message in the next 7 days.

    Ideal for optimizing messaging to drive site visits.

    • Key use case: Find users who are likely to click links within your emails and in doing so boost your email KPIs while maximizing for the number of sessions you drive. How-To
      1. In Audience Builder, create a Smart List using the criteria 'click_7 is greater than k-tile 900', or any other high k-tile threshold of your choice.
      2. When building your campaign, choose to send it to this list.
    • Zephyr Variables:
      • profile.vars.predictions.click_7.ktile (recommended)
      • profile.vars.predictions.click_7.num

    Pageviews

    (pv_30)

    The predicted number of pageviews a user will experience over the next 30 days

    Ideal for media clients, the pageview predictions tend to follow the power law to an extreme where 80% of pageviews typically come from the top 5% of your customer base.

    • Key use case 1: Identify your most engaged customers. Find and acquire other potentially high-value users leveraging the Facebook integration, to optimize your ad spend and acquire other highly engaged and higher-value customers. How-To
      1. In Audience Builder, create a Smart List with the criteria where the pv_30 k-tile is greater than a certain number (e.g. top 1% of users; a k-tile exceeding 990).
      2. On your Lists page, click the cloud-upload button to send the list to Facebook. Note: This requires that you configure the Facebook Custom Audiences integration.
    • Key use case 2: In SPM, identify readers with a low likelihood to stay on-site or otherwise engage and point them to sponsored links. How-To
      • In your template, use the following Zephyr code. (Customize the k-tile range and link code as needed.)
        {if profile.vars.predictions.pv_30.ktile < 200}
        <Links via Taboola, Outbrain, or other service>
        {/if}
    • Zephyr Variables:
      • profile.vars.predictions.pv_30.ktile
      • profile.vars.predictions.pv_30.num

    Per Item Value

    (aiv_7)

    The predicted value of an individual item in a purchase order if a purchase is made in the next 7 days.

    • Key use case: Further personalize your product offerings in email or on-site (SPM) by only recommending products that fit within the customer's predicted price range. How-To
      1. To access aiv_7 in the template, use the following Zephyr code
        {profile.vars.predictions.aiv_7.num}
      2. To offer products in a range of, for example, 90%-120% of aiv_7, you could use the following code to filter the content feed and then display them (in this example, the top 5 products in the price range):
        {foreach content as c} {filtered_products = filter(content, lambda c: (c.price >= (0.9 * profile.vars.predictions.aiv_7.num) && c.price <= (1.2*profile.vars.predictions.aiv_7.num)))} {/foreach}
        {foreach filtered_products as i,c} {if i <= 5} {if (i % 3) == 0 && i != 0} <br> {/if} product {i}: {c.title} - ${number((c.price/100),2)} - {c.description} {/if} {/foreach}
    • Zephyr Variables:
      • profile.vars.predictions.aiv_7.ktile
      • profile.vars.predictions.aiv_7.num (recommended)

    Number of Items Purchased

    (item_7)

    If a user makes a purchase in the next 7 days, the number of items in their purchase basket.

    Ideal for optimizing messaging to maximize purchase conversions.

    • Key use case: Understand if a user is likely to purchase many or few items and alter how you present your offerings.
    • Example: If a customer is likely to buy only one item, showcase individual high-priced items to push up AOV. If they're likely to purchase 3 items, show them how you would create an outfit with 3 different categories. How-To
      • {if profile.vars.predictions.aiv_7.num <= 1}
        Single item priced at user's AOV
        {else}
        Here are {profile.vars.predictions.aiv_7.num} products for you to purchase
        {/if}
  • Zephyr Variable:
    • profile.vars.predictions.item_7.ktile
    • profile.vars.predictions.item_7.num (recommended)