Q&A – Teradata Aster’s Product Recommender Solution: What Is It, and Why Does It Matter?

Learn Data Science
Teradata Employee

Product recommendations drive significant revenue for retailers, both in-store and online. But these recommendations are often based on one-size-fits-all promotions, or merchants and category managers manually create them.

However, big data analytics provides the opportunity for an alternative approach.

Teradata Aster recently launched a solution to help companies make data-driven, customer-centric recommendations. I sat down with David Gebala for a short video Q&A to discuss several questions we’ve been hearing about the Product Recommender solution. Below is an abbreviated transcript.

What is the Teradata Aster Product Recommender solution?

The Product Recommender solution makes it simple for retailers and merchandisers to take a data-driven, customer-centric approach to product recommendations. It lets you combine both online and offline purchase histories to make extremely relevant, personalized recommendations to your customers. The recommendations help retailers increase average order value by upselling and cross-selling products for which their customers show a statistical likelihood to purchase.

One thing that’s really important, while we are using both tried-and-true and cutting-edge analytic techniques, we provide this power via an application, so digital marketers, analysts, product managers and others can take advantage of this power without writing a single line of code. We’ve completely masked the complexity, from the businessperson’s perspective. All the business user has to do is select their desired parameters and click “Run.” The application is part of the Teradata Aster AppCenter, and if you want to expose results via BI tools or integrate with CRMs or marketing automation systems, that’s extremely simple to set up.

Product Recommender is for business users rather than technical users or data scientists?

That’s exactly correct. All we need is access to your historical market baskets. Many companies already store this in Aster, especially the data from their websites. If you can point us in that direction, everything else is point and click from an application within AppCenter.

You have the capabilities of some cutting-edge analytic functions from the Aster library, but this is not a solution targeted to data scientists. This is very much a solution for business people to be able to provide incredibly targeted, personalized product recommendations.

You mentioned that the Product Recommender leverages both tried-and-true and cutting-edge analytic techniques. Can you talk about those?

The two primary techniques to which I’m referring are collaborative filtering (cFilter) and personalized stochastic approach for link-structure analysis (pSALSA). Depending on the use case, one may be more appropriate than the other, but we often find a combination provides the greatest value. Our team is happy to work through this on a case-by-case basis.

To go a little deeper, cFilter is a popular analytic technique to explore pairwise affinity, or how frequently items are purchased together. In this case, we are looking at both an individual customer’s historical market basket, as well as the baskets of all customers. We look at all items to determine how frequently they are purchased together, and calculate a statistical confidence for the probability that they are purchased together. Then we examine a specific customer’s historical basket – online and offline – and provide recommendations based on what they are likely to purchase.

pSALSA is a newer approach to recommendations that is most popularly known as the technique behind Twitter’s “who to follow” feature. pSALSA, which leverage Aster’s graph engine, is great for large and evolving data sets, and it’s extremely scalable. pSALSA  considers each person a node and the products they’ve purchased connectors. It walks those connectors to find a set of customers similar to a specific customer, looks at what they’ve purchased, and makes recommendations based on that.

(Ed: Check out this blog for more details on cFilter and pSALSA as they are applied to product recommendations.)

How does the Product Recommender solution compare to competitive offerings?

A lot of the competitors have put most of their investment into the front-end. The analytics happening in the backend may receive some lip service, but they are relegated to second-class, or worse in some cases. With our solution, we have heavily invested in the backend and analytics working with multiple customers to make sure you get the best possible recommendations, at scale.

Also, this is built on Aster’s analytics accelerator, so there are a wealth of other capabilities that most marketing organizations will leverage. We have apps for things like: on-site search optimization; paths to churn; attribution analysis; customer satisfaction index; and many more. And data scientists who get their hands on the system will want to directly leverage many of the advanced analytic functions in our library.

Who would benefit from the Product Recommender solution?

Most of the customers using this are are retailers and ecommerce executives who want to increase share of wallet, average order value, customer lifetime value, and similar key metrics.

Well, it sounds very interesting. What next?


If you’re already a Teradata customer, contact your account executive. You may be able to get up and running in just a few weeks with our services offering. If you’re starting from scratch, send me an email at ryan.garrett@teradata.com and I’ll provide some additional resources and be happy to help you get started. Also, we have a lot of great resources in our user community at https://aster-community.teradata.com.