How to boost upsell with AI Product-Based Recommendation Audiences

On this page

Upselling existing customers can be a huge revenue opportunity for businesses. But, how do you know who is ready to buy an upgraded item?

With AI Product-Based Recommendation Audiences, we make it easy for marketers to find the people most likely to purchase their advanced or upgraded products so you can send those targeted audiences to any of your key customer communications channels — whether that be ads, emails, SMS, etc. And, with our AI Recommendations, you can do this all without needing to rely on data science teams. Here’s how: 

Step 1: Create your Recommendation Catalog

In order to use Recommendations, you first need to build your Recommendations Catalog in Segment. We make this easy by using the existing events you have set up in your Segment workspace to help infer your product catalog. This means you don’t have to link a separate catalog to Segment. 

To get started, open your Engage space and navigate to Engage > Engage Settings > Recommendation catalog.

On the Recommendation catalog page, click Create catalog.

Anything can be a catalog: ecommerce or retail products, music, movies, content articles, food items, events, etc. For this recipe, we’ll use retail clothing items.

Select up to 10 product-related events you’d like Segment to use as a basis for recommendations. We recommend selecting 3-7 different events that represent a user interaction. For example, Product Added to Cart, Product Searched, or Product Viewed.

The interaction events you choose here will not only help generate the catalog, but will also inform the machine learning recommendation model.

Then, select a product ID for each product-related event you previously selected, then click Next.

If you’re not a retail brand, this is where you could select an alternative type of item to build your catalog around. For example, a B2B business that may want to make event recommendations could select an event_id; or a media company that wants to recommend content to people could select an article_id. 

Next, you’ll map the event properties to the suggested model columns. Although Category is the only required model column, we recommend mapping all properties of a product hierarchy to allow for increased granularity when building your Product-Based Recommendation Audience.

For a retail brand, “category” will likely be the best event property to select in the Category fields. For Name fields, you’ll likely select “name” or “product_name.” Or, if you operate on SKUs instead of product names, you can update both the model column name and property fields to be SKUs.

(Optional): To add an additional column to your model, click + Add column on the Map properties page.

When you’ve completed your mappings, click Save. Your catalog should then be built within an hour.

Step 2: Build your Product-Based Recommendation Audience

Once you’ve created your Recommendation Catalog, you can build a Product-Based Recommendation Audience.

Open your Engage space and click + New audience.

Select Product-Based Audience and click Next.

When getting started in this audience builder, you want to think product-first. Since this is an upsell campaign, you’ll want to select the specific upgraded or advanced product you want to find the best audience for. 

For this example, let's say you just released a new premium running sneaker called “SuperiorStride” that you’re looking to upsell to your customer base. You would select products where the “Name” is “SuperiorStride.”

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.

Then, scroll down to select the audience size  by choosing one of the pre-populated options. Or move the slider to create a custom audience. We recommend 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.

For this example, let’s choose the Top 5% most likely to purchase your new SuperiorStride sneakers and click Next.

On the Select Destinations page, select any destinations you’d like to sync your audience to. For this example, we’ll choose Twilio SendGrid Marketing Campaigns, and click Next.

Then, enter a name for your destination, update any optional fields, and click Create Audience to create your audience.

Step 3: Send a personalized email for people most likely to purchase your upgraded product

Once you’ve sent your AI Recommendation Audience of the top 5% of people most likely to purchase premium “SuperiorStride” sneakers to your email tool, you can craft and send a highly personalized email highlighting this new item — maybe even offering a special promotion for it.

Giving marketers the ability to build their own AI-powered Recommendation Audiences is already driving significant results for Segment customers. In our Recommendations beta, one big box global retailer ran an email test with customers who were most likely to purchase Apple products. As a result, they saw a 592% increase in sales per email.

Try Recommendations today to discover what kind of upsell results you can achieve.