Recommendation Audiences

Recommendation Audiences lets you select a parameter and then build an audience of the people that are most likely to engage with it. Segment optimized the personalized recommendations built by Recommendation Audiences for user-based commerce, media, and content affinity use cases.

You can use Recommendation Audiences to power the following common marketing campaigns:

• Cross-selling: Identify an audience of users who recently purchased a laptop and send those customers an email with a discount on items in the “laptop accessories” category.
• Upselling: Identify an audience of users who regularly interact with your free service and send them a promotion for your premium service.
• Ranking: Identify an audience of users who frequently interact with one category of your website and send them a promotion that contains only items from this category.
• Moving excess inventory: Identify an audience of users who are in the top 5% of purchasers for a specific brand you sell and send them a coupon for the excess inventory you have of that brand.
• Next best action: Identify an audience of users who frequently read articles in your website’s “Sports” category and recommend those users your latest sports article.
• Increasing average order value (AOV): Identify an audience of users who frequently interact with the “For Kids” section of your website and send them a back to school promotion in August, with free shipping after a set price threshold.

Create a Recommendation Audience

Set up your Recommendation Catalog

A Recommendation Catalog identifies the product events you’d like to generate recommendations from and maps those events against your existing data set.

To create your Recommendation Catalog:

1. Open your Engage space and navigate to Engage > Engage Settings > Recommendation catalog.
2. On the Recommendation catalog page, click Create catalog.
3. Select up to 10 product-related events you’d like Segment to use as a basis for recommendations. Segment recommends selecting 3-7 different events that represent user interaction. For example: Product Added to Cart, Product Searched, or Product Viewed.
4. Select a product ID for each product-related event you previously selected.
5. Click Next.
6. Map event properties to the suggested model columns. Segment recommends mapping all properties of a product hierarchy to allow for increased granularity when building your Recommendation Audience.
   (Optional): To add an additional column to your model, click + Add column on the Map properties page.
7. When you’ve completed your mappings, click Save.

Create your Recommendation Audience

Once you’ve created your Recommendation Catalog, you can build a Recommendation Audience. A Recommendation Audience lets you select a parameter and then build an audience of the people that are most likely to engage with that parameter.

To create a Recommendation Audience:

1. Open your Engage space and click + New audience.
2. Select Recommendation Audience and click Next.
3. Select a property and value that you’d like to build your audience around (for example, if the property was “Company”, you could select a value of “Twilio”). For values that haven’t updated yet, enter an exact value into the Enter value field. If you’re missing a property, return to your Recommendation catalog and update your mapping to include the property.
4. Set a maximum audience size by selecting one of the pre-populated options, or move the slider to create a custom audience. Segment recommends audiences that contain less than the top 20% of your audience because as the size of your audience increases, the propensity to purchase typically decreases. See Best practices for more information.
5. When you’ve filled out all fields, click Next to continue.
6. On the Select Destinations page, select any destinations you’d like to sync your audience to and click Next.
7. Enter a name for your destination, update any optional fields, and click Create Audience to create your audience.

Best practices

• When mapping events to the model column during the setup process for your Recommendation catalog, select the event property that matches the model column. For example, if you are mapping to model column ‘Brand’, select the property that refers to ‘Brand’ for each of the selected interaction events.
• Because a number of factors (like system load, backfills, or user bases) determine the complexity of an Audience, some compute times take longer than others. As a result, Segment recommends waiting at least 24 hours for an Audience to finish computing before you resume working with the Audience.
• As the size of your audience increases, the propensity to purchase typically decreases. For example, an audience of a hundred thousand people that represents the top 5% of your customers might be more likely to purchase your product, but you might see a greater number of total sales if you expanded the audience to a million people that represent the top 50% of your customer base.