Real-time product content ranking—an essential capability for creating fresh, relevant experiences
Traditionally, efforts to make recommendations have focused on products related to a specific customer action. If a customer clicks on an item, recommendations are based on that action and the customer sees products related to what they clicked (this often becomes repetitive). Similarly, if a customer searches for an item, that item, or very similar items will follow them around the Internet and in their email. The idea is that an action shows intent on the part of the customer. While some intent can be revealed by customer clicks and searches, this intent is short-lived.
This technique pre-computes product associations based on crowd-sourced traffic information. When the customer takes a specific action related to a product, the recommendation system offers pre-computed associated products. Unfortunately, this 'pre-computed' approach yields recommendations which quickly get stale and repetitive, and fail to address new or 'long-tail' items. And unless and customer comes to your site and makes a new action, their profile gets stale quickly. This leads to recommendations with little relevance to a customer’s quickly changing interests and behaviors or overall product performance.
Nearly every online, mobile and email customer touch point presents an opportunity to offer a curated experience that is highly adapted to a customer’s current circumstances. But rather than offering a robotic response to each individual action, users are looking for more of a dialogue that dynamically adapts to the current context while incorporating insight from previous interactions. In this way, relevance is determined by all of a customer's interactions, rather than the last action.
Traditional solutions fall short here. They simply react to the last action: if you look at a dress, you'll see more dresses; if you look at a jacket, you'll see a bunch of jackets. If you don't come back, you will see jackets in your email for the next 2 months! A system designed for adaptation and exploration would offer diverse items from the same brand, dresses of similar color in different styles, a variety of jackets when they go on sale, and more dresses when new collections arrive.
In order to deliver this kind of adaptive offering, a retailer needs to rank their catalog and decide which products to offer and display — in real-time (as opposed to pre-calculated associations like traditional systems use). The power of real-time ranking is that it constantly re-evaluates your entire catalog looking for product content that may be suitable for the user based on the current context, like location or device, real-time product performance and inventory information, and real-time changes in the user behavior.
Real-Time Product Ranking versus Traditional Alternatives
Re-ranking your catalog for each customer interaction provides superior predictions because it:
- Uses the full spectrum of customer's behavior (as opposed to a specific action), including purchases, browsing activity, search queries, and passive behavior, quickly adjusting to the customer's actions while incorporating prior history. Alternatives simply use individual actions like the last shopping cart item or the last click to deliver a pre-computed carousel of associated products.
- Incorporates recent product performance (trends, popularity) as well as stock-outs for ranking. Pre-computed product carousels are unresponsive to product performance and inventory information.
- Incorporates passive behavior and context in addition to the customer's active behavior. This ensures that the user will not see the same items over and over again or will not see items that cannot be shipped to the user's current location - this solves recommendation fatigue. Pre-computed recommendation carousels do not change with passive behavior or context.
- Surfaces relevant new products without bootstrapping time. Alternatives do not have this capability and need a bootstrapping time to attain a minimum level of volume of transactions to incorporate new products (basically, crowd-sourcing 'wisdom' takes time and won't work for new arrivals).
- Allows you to experiment with product offerings to survey and discover potential customer interests. Solutions using pre-computed product associations are not architected to iterate and explore.
How Does Real-Time Ranking Work?
Real-time ranking incorporates richer and up-to-the-second data to re-rank your entire catalog for every customer interaction.
Real-time Product Ranking for Each Customer Interaction:
Where Can Real-time Product Ranking Be Used?
Real-time product ranking can be used to enhance nearly every customer interaction whether it’s on your website, in emails or in a mobile experience (web or app), including:
- Promotional and lifecycle emails
- Home pages
- Category and section pages
- Mobile app environments
- Trigger emails
Q & A Related to Real-time Ranking and Delivery of Adaptively Curated Product Collections
If we have a lot of products in our catalog, is it feasible to do real-time ranking?
Yes, Jetlore has the proven ability to real-time rank catalogs with millions of items within milliseconds.
Will all my products be eligible for being ranked and displayed, or just popular/high traffic items?
All products, long-tail low traffic and new items will be surfaced as Jetlore explores customer interests. The Jetlore natural language processing engine extracts attributes at the product level (within your catalog) and uses this information to make predictions.
How does it work for email?
Jetlore produces HTML that you insert into your campaigns and triggered emails managed within your email service provider (ESP). Jetlore works with any ESP.
How does it work onsite and the mobile web?
The Jetlore Onsite API makes it easy to embed real-time ranked product collections unique to each customer on any page of your website. (And it’s responsive for the mobile web).
How does it work in an app?
Jetlore provides a mobile SDK that makes it straightforward to integrate and display product collections personalized to each customer in your mobile app.
What’s the work on my end to start using real-time product ranking?
It takes 5-6 weeks to perform minor integration, data model training and pilot go live. Click here to learn more about getting started.