November 09, 2017
Predictive Intelligence in Ecommerce, Part I
November 09, 2017
Predictive Intelligence in Ecommerce, Part I
At a recent retail conference, Earvin “Magic” Johnson shared his experiences in the retail sector. He communicated that after spending some time and money in retail, he realized that people desire to buy what they liked rather than what “Magic” Johnson preferred himself to sell in his retail establishments.
The problem is likely to sound familiar – merchandisers and other experts within the industry attempt to predict consumer choices; however, the tools they use usually aren’t robust predictive measures and are largely based on manual data manipulation. Despite the combination of sophisticated studies and analytics, the recommendations we see in the online storefronts often are neither consistently relevant nor are they interesting. Technological and algorithmic advances in predictive techniques are available, and one would think that with these innovations successful personalization would be the result.
Recommendation engines are a specific focus within the online merchandising sector. In most cases, the “up-sells,” “cross-sells,” and customer-specific segment recommendations must be manually set through the platform’s back-end. While the term “Predictive Intelligence” (PI) is frequently the goal, it’s still perceived as a black box which is only available to a select few in the retail industry. As such, only a small fraction of the retailers we talk to and work with use integrated analytics and personalization engines.
Therefore, this brings us to two fundamental questions:
Fortunately, there has been much progress regarding PI, and you can immediately take advantage of its benefits.
At Valtech, we think that the most exciting part of retail is the creativity one has in connecting the digital and the physical worlds. Long gone are days when the focus of top e-commerce professionals focused on building a better shopping cart. Indeed, the best e-commerce platforms provide a very similar set of essential functionalities that create a fully responsive user interface (UI) and user experience (UX). The linear conversion funnel approach for digital marketing has fallen by the wayside. Subsequently, the path to purchase is increasingly multi-dimensional, which reinforces the need for prediction based on several different consumer features. Prediction accuracy is a prime differentiator between the top innovators in retail and those who are struggling with customer engagement and retention. As such, to differentiate their brands, top innovators in retail focus on true personalization of customer’s experiences and making unique product and service recommendations in unprecedented ways. Some retailers are taking into account many more factors than just demographics, geolocation, prior browsing & purchasing history of their shoppers when making their product recommendations. For example, B&HPhotoVideo.com is frequently mining social media and targets their shoppers with personalized ads that showcase complimentary products to those that a person has already purchased. Such recommendations take into account the social context and the intent of the shopper, such as travel or project plans published, their social sentiment etc.
Interestingly, some ecommerce platforms are better suited for product recommendations accuracy than others. The SaaS model-based Salesforce Commerce Cloud, for instance, collects shopper clicks across the entire number of retailer instances. While protecting personal data and preserving confidentiality of individual retailer instance shoppers, the entire stream of web clicks is fed into the Salesforce Einstein Cloud. There, this data then can be transformed and analyzed to define appropriate shopper cohorts, to identify their shopping patterns and preferences not across a single retailer brand, but across the entire industry vertical. Upon making sense of such data, the predictive intelligence recommendations make by Einstein Cloud can then be fed into individual retailer storefronts.
Very advanced predictive recommendations & analytics in ecommerce are now made possible due to two factors:
Given the amount of data flowing into your databases and languishing in the data lakes, additional questions arise, such as “What can be done with this data,” and “What is a Predictive Intelligence as it relates to the retailer sector?”
Personalization techniques for online retail fall into one of two broad categories:
Rule-based personalization has been widely used and is relatively simple to incorporate into your e-commerce platform. For example, all shoppers visiting a store from a specific geographical location receive a targeted offer or are presented with certain products. Or, if a consumer has previously purchased items from the store, they’re offered a product promotion or a free service that somehow correlates with their prior purchase behaviors. In many e-commerce platforms, the rule based function is further disaggregated into catalog-wide rules and shopping cart-specific rules. We call such personalization “hygiene” as it really represents the basics of online commerce and mirrors the core principles of merchandising already used by offline retailers for many years.
Beyond the rule-based personalization lies the Predictive Intelligence (PI) that is enabled, as discussed above, by digitization of vast volume of consumer data that is available for mining. At the core of such personalization are Data Mining, Natural Language Processing and Machine Learning. Collectively, these data science practices and methods represent are what we call Predictive Intelligence in commerce. In its very basic form, PI enables marketers to observe customer behavior, and with every action taken, to build a profile of individual customer preferences, likes, and dislikes. There is often confusion between PI and Artificial Intelligence (AI). AI is a more specific application of machines deriving patterns on their own by generating a set of heuristics that humans have not initialized within the system. In other words, an AI machine must be able to carry out tasks in a way that we, humans, would consider “smart”. In that sense, PI is a current application of AI based around the idea that machines learn from and find patterns in the vast amount of data available to them regarding shoppers’ preference and behavior.
Therefore, PI is not AI. PI requires various types machine learning algorithms to be applied to consumer data for pattern detection and subsequent classification for decision making. Though not wholly separate from Machine Learning, true AI has not yet been fully realized.
 Busath, M. (2017). UserVoice increases MQLs by 37% using predictive marketing. Infer.com. Available at: https://www.infer.com/category/use-cases/
 Wettemann, R. (2015). Salesforce marketing cloud: Room & Board. Available at:
 IBM Marketing Cloud. (2016). Outdoor retailer uses segmentation and targeting to increase open rates by 80%. Available at: https://www-01.ibm.com/common/ssi/cgi-bin/ssialias?subtype=AB&infotype=PM&htmlfid=ZZC03314USEN&attachment=ZZC03314USEN.PDF
 Tastehit.com. (2016). How personalized recommendations help increase conversion. Available at: https://secure2.sfdcstatic.com/assets/pdf/datasheets/mc-nucleus-salesforce-marketing-cloud-roi-case-study-room-and-board.pdfhttps://www.tastehit.com/blog/how-personalized-recommendations-help-increase-conversion/
 Tamturk, V. (2017). The ROI of recommendation engines. Available at: http://www.cms-connected.com/News-Archive/January-2017/The-ROI-of-Recommendation-Engines
 Kashmir Hill (2012). How Target Figured Out A Teen Girl Was Pregnant Before Her Father Did. Available at: https://www.forbes.com/sites/kashmirhill/2012/02/16/how-target-figured-out-a-teen-girl-was-pregnant-before-her-father-did/
 Spenser Soper (2016). More than 50% of Shoppers Turn First to Amazon in Product Search. Available at: https://www.bloomberg.com/news/articles/2016-09-27/more-than-50-of-shoppers-turn-first-to-amazon-in-product-search