The Success of Your Cross-sell / Upsell Strategy Relies on Customer Data

The Success of Your Cross-sell / Upsell Strategy Relies on Customer Data

Increasing Share of Wallet Through Cross-Sell and Upsell

This is the third post in our blog series on the Fundamentals of Disruptive Marketing, which explores how technology and innovation are combining to increase the effectiveness of traditional marketing tactics.

Cross-selling and upselling are traditional merchandising strategies employed by retail and eCommerce marketers. These strategies are intended to increase ‘share of wallet,’ an industry term used to indicate the percentage of total spending that a shopper will spend on one specific brand. In today’s competitive business climate, retailers are actively competing for the consumer’s share of wallet, strategizing ways to woo shoppers away from their competitors.

The Success of Your Cross-sell / Upsell Strategy Relies on Customer Data

  • Cross-selling: suggesting an additional item to accompany one under consideration
  • Upselling: suggesting a superior model for a product under consideration

Until somewhat recently, cross-selling and upselling generally followed a one-size-fits-all execution, with product recommendations based on either past purchase and browsing behavior or on known characteristics of items. For example, if a customer bought The Merriam-Webster Dictionary from an eCommerce website, the website might also recommend The Merriam-Webster Thesaurus, or another writing reference book, in a prominent placement to be easily added to the customer’s shopping cart. In this basic example, the retailer relied on limited data, i.e. 50% of people who bought product X also bought product Y last year.

Nowadays, retailers can make use of a growing repository of data on prospective customers, such as which competitors’ sites they have visited, which keywords they have searched and whether they are a new or repeat customer. And since recent research indicates that 88% of consumers research products online before making a purchase decision, it is increasingly necessary for retailers to tap into this data — and combine it with other sources of data — to inform smarter product recommendations.

The Role of a Recommendations Engine in Increasing Share of Wallet

Trying to increase share of wallet without reliable data can prove to be a dangerous tactic. If a retailer fails to recommend the right products in their cross-sell or upsell placements, they might lose customers for good, not just not miss out on the added revenue. According to a recent survey by Bazaarvoice, today’s customers expect personalization, especially if the customer believes the company knows something about them: “More than 50% of shoppers say it is very useful when retailers show them products they’re looking for, and improves the shopping experience. But those numbers quickly change when retailers send too many emails and make bad recommendations. In fact, shoppers cited these as some of the two biggest reasons to shop somewhere else. In our survey, 38% of consumers said they won’t return to an online retailer that recommends things that don’t make sense for them.” (p.11)

Smart retailers know that customer demographics don’t tell the whole story — and certainly don’t help predict what the customer might be likely to purchase. In its personalization eBook (p.15), Bazaarvoice cites an example from the marketing world: Prince Charles and Mick Jagger, both British men of approximately the same age. Based on demographics alone, a retailer might show these two shoppers similar product recommendations. But considering the difference in their lifestyles, attitudes and personalities, it’s easy to see how this strategy could backfire.

Including what you know about a customer’s buying and browsing behaviors helps you offer them very distinct experiences. Behavioral data — especially when combined with other forms of data, such as demographics, location and weather — turns out to be one of the best ways to learn about a person’s shopping interests. Examples of behavioral data for retailers include:

Get Treasure Data blogs, news, use cases, and platform capabilities.

Thank you for subscribing to our blog!

  • Browsing product reviews
  • Submitting product reviews
  • Engaging with product pages
  • Keywords searched
  • Items added to cart
  • Transaction history

These kinds of data display what a person is doing, and what their current interests are. Consider, for example, the case of a 64-year-old woman who is browsing a furniture website. Based on her demographics, you may not think to show her baby cribs for sale. But if you were paying attention to what she has been browsing – and could see that she recently typed “best selling baby cribs” into Google — you would have a much better chance of displaying relevant content, including the crib she is intending to buy for her granddaughter.

Recommendation Engines Power Personalization

Recommendation engines are data filtering tools that use data to recommend the most relevant items to a particular user. Their algorithms determine what to display as ‘related products’ based on what they know about each individual. With the growing data on individual shoppers, it is becoming increasingly possible — and important — for companies to search, map and provide each prospective customer with a relevant assortment of products tailored specifically to their preferences and tastes.

More than 80 percent of the TV shows people watch on Netflix are discovered through the platform’s recommendation engine. Two of Netflix’s product executives, Carlos Gomez-Uribe and Neil Hunt, assert in a recent product paper that “the combined effect of personalization and recommendations save us more than $1B per year.” A recommendation engine typically creates recommendations in one of two ways — collaborative filtering or content-based filtering. Collaborative filtering looks at browsing behavior and compares it to other shoppers’ behavior to predict other items that the user may have an interest in. Content-based filtering uses a series of an item’s characteristics to recommend items with similar properties. Sometimes these approaches are combined in what is known as a hybrid recommendation engine. is a hugely popular shopping app that provides discounted and unbranded direct-to-consumer products at the lowest possible price. The company’s products are discoverable mainly via browsing instead of search, and rely on products that are recommended based on a massive amount of data from customers’ wishlists.’s giant recommendations engine makes it easy for a shopper to discover other relevant products to the ones they are already browsing.

Customer Data Platforms (CDPs) Power Recommendations Engines

A recommendation engine works by collecting data, storing it, analyzing it and filtering it. Because of the sheer complexity of these tasks and the enormous quantities of data they handle, a CDP is a must-have for a recommendations engine. Treasure Data’s ability to integrate with hundreds of systems is crucial for machine learning. The more data the better for machine learning to work efficiently. Moreover, the flexible open-source technology includes an extensive machine learning library that supports key functionality for recommendations engines such as regression, classification, recommendation, and more.

With its powerful machine learning capabilities — the backbone of recommendations engines — Treasure Data’s enterprise CDP powers some of the very largest recommendations engines in the world, such as

Wish needed to ingest 17 billion events per day from multiple sources and and process in the trillions. Their goal was to ensure 95% relevance of the products that crossed the consumer’s field of vision. In order to achieve this, the app and CDP would need to successfully predict customer needs and offer the right product recommendations. Treasure Data’s enterprise CDP helped Wish to:

  • Build and scale a personalized shopping recommendation engine
  • Unify customer data collected from multiple sources (website, Facebook, mobile app)
  • Constantly improve the customer experience through A/B testing in order to deliver the best customer experience
  • Grow into the #1 mobile shopping app in the US, experiencing 2x conversion growth YOY

To learn more about how Treasure Data helped Wish, see our case study.


If you’re a retailer competing with the likes of Amazon, you’ll need to work very hard to woo shoppers to your brand. Creating a personalized experience for each customer is becoming a necessity — and at multiple levels of the purchasing process. Customer data will continue to play a key role in informing your strategy: what to offer, when to offer, how to offer, and who to offer. And an enterprise CDP can be an enormous help in managing the entire process: building segments, making predictions, and managing, analyzing and optimizing campaigns. To learn more about to the advantages of an enterprise CDP, download our new white paper, “The ROI in a Customer Data Platform.”

Rob Glickman
Rob Glickman
Rob serves as Head of Marketing, Data Business at Arm. Previously, he was VP of audience marketing at SAP where he led a global team of marketers chartered with driving modern marketing demand generation programs. He brings nearly 20 years of marketing experience ranging from lean startups to large enterprises, including running product marketing for Symantec, eBay and PayPal, where he held various marketing leadership roles globally.