With online becoming the only retail channel, customer expectations for the digital experience you deliver have increased. This means the mechanisms that could traditionally be used to compensate for digital shortcomings, such as in-store commerce, have disappeared. Embracing and applying Artificial Intelligence and Machine Learning to create intelligent buying experiences that are unified, relevant and valuable has become the only way to serve your customers and make them stay. 

Power of Intelligent Product Recommendations

Intelligent product recommendations can help you achieve that goal, as they empower your customers to find exactly what they need. In fact, more than 90 percent of shoppers reported that they would return to a site with a personalized digital experience that tailors product recommendations based on their previous shopping habits. McKinsey actually estimated that 35 percent of what consumers purchase on Amazon and 75 percent of what they watch on Netflix come from product recommendations. 

Essentially, intelligent product recommendations have the potential to serve as an effective stand-in for your friendly shop assistant at a time when in-person interaction is no longer possible. Yet, Gartner also reports that 80% of marketers are actually planning to abandon their personalization efforts due to lack of ROI, where product recommendations serve as a key component. So why isn’t the true personalization potential of product recommendations being realized? 

Part of the reason is that most digital business pros still haven’t borrowed all the recommendation engine techniques and concepts that giants like Netflix and Amazon have already mastered and leveraged to fit their business goals. In particular, research suggests that most Ecommerce websites leverage only a very narrow number of use cases for product recommendations, rendering many of today’s digital shopping experiences clumsy, fractured and inauthentic.

Don’t let yourself fall into that category. As long as you develop a clear understanding of the where, what, why, and how of intelligent product recommendations, you can make them work for you. Here is a guide on how to do just that.

EbookYour Guide to Delivering Intelligent Shopping Experiences

Where to Put Recommendations

Ecommerce customer journeys have become increasingly complex. You can’t just focus on individual touchpoints anymore, especially when it comes to providing product recommendations. 

Most ecommerce players don’t have the same percentage of registered users that Amazon and Netflix do. And requesting that a user create an account is actually a key driver of cart abandonment. This means that while you are now probably familiar with product or movie recommendations coming by email from Amazon and Netflix, the top priority for you should be to focus on delivering relevant recommendations on your website for both return users and those who are new/anonymous guests

As it turns out, there are actually quite a number of different places where recommendation widgets can prove effective. 

To begin, let’s look at where product recommendations can be surfaced along the digital customer journey. As they have many different applications, it’s important to determine where you actually need them:

  • Homepage recommendations: The homepage is the first thing that shoppers coming from direct traffic see when they visit a site. Homepage recommendations serve the purpose of informing customers about the latest deals and discounts, showcasing the product portfolio and personalized offerings. This might be what you have in mind when you think about Amazon or Netflix recommendations. 
  • Product page recommendations: A product page (or product information page) is where shoppers can find detailed descriptions of a product and its features and can choose to add it to their cart or order it right away. The main aim of recommendations on these pages is to display the most relevant items in order to compel shoppers to continue browsing. 
  • Cart recommendations: These can enable you to recommend accessories or products frequently bought together. 
  • Category page recommendations: The number of categories on such a page can be daunting. Facets and filters aren’t much help to a shopper that doesn’t quite know what it is that they need. Product recommendations here can be used to guide them towards what they might want to buy based on what you know about them up to that point. 

However, simply providing recommendations at different touchpoints is not enough to effectively assist customers on their journeys. In order to do so, you must acknowledge that different moments and touchpoints carry different expectations and then provide the most appropriate recommendations at each point. For instance, cross-selling on the product details page can at times be distracting to customers. Similarly, showing similar items on the cart page might hinder the purchase underway. 

Moreover, at some points during the journey, customers might benefit from content recommendations as well. A traditional view in ecommerce has been that content can act as a distraction and dilute the conversion funnel. However, this is not always true. In fact, content can enhance decision-making and conversion, depending on where the customer is in their buying journey. 

You can benefit from recommendations that blend content and products to enhance the likelihood of conversion, as Machine Learning identifies correlations in content consumed over the course of  previous successful buyer journeys. If a piece of content compelled one person to make a purchase, it will likely compel many others to do the same. 

Personalized or Trending Recommendations

Just as there are various places recommendations can be surfaced, there are also various different types of recommendations and functions associated with each. Try to think about which of the following types would work best for you. 

Consider the homepage recommendations use case described in the previous section. These recommendations can actually be broken down into a variety of different types. Let’s explore this breakdown using Netflix as a reference once again: 

  • Personalized recommendations: The Netflix recommender system predicts what viewers would like to watch in the moment.
  • Trending recommendations: These are defined by popularity in a relevant geographical area. 

Moreover, Netflix also wants you to notice the Netflix Originals – the series produced by Netflix. They bring them to your attention with:  

  • Business-focused recommendations: These are important for Netflix for two reasons: (1) they have spent a lot of money producing them, and are in most cases only found on Netflix, and (2) Netflix has to pay content owners when users watch their content, but if that owner is Netflix then they save money. 

After identifying where you need to use recommendations and what types you intend to use, it is important to determine why you need them. In other words, what are the business outcomes you’re trying to achieve as the result of their implementation? 

Recommendations Married to User Intent

It is commonly said that recommendations are valuable for commerce players because they help support up-selling and cross-selling initiatives which boost conversions. This is true, but also gives rise to two problematic assumptions that must be addressed.  

First, cross-selling and up-selling are not the only paths to conversion. Depending on the situation, down-selling might actually be the most optimal approach. 

The key to determining which approach is best comes down to understanding individual customer intent. What do they need from you? Only by uncovering that intent and factoring it in alongside your business needs can you ensure that both of your goals align in the product recommendations ultimately presented to them. 

For example, if a shopper is about to abandon the cart because the item is too expensive, offering a 15% discount or even recommending a less expensive product might help drive conversions. However, if shoppers value personal relationships over low prices, free membership to an exclusive buyers’ group may be even more compelling. 

The rise of predictive analytics and propensity modeling has made this balancing act far easier for you to achieve. They enable you to leverage signals collected during the journey as well as customer sentiment in order to unlock customer intent. With this understanding in place, you can then personalize every interaction (like recommendations) in a way that meets customer needs and those of your business.  

[If you want to see this in action, our researchers have actually built a model to predict purchase intent from clickstream data, understanding shoppers’ true intent from real-time signals. It’s no wonder 80% of business leaders consider predictive analytics very important to their companies’ success and 42% of them are already experimenting with predictive technologies.]

Second, recommendations can be utilized for far more than boosting conversions. Take zero-results recommendations as an example, which can be found after an unsuccessful product search. Getting to “zero search results” often means the end of a customer’s session on a site. These recommendations are designed to improve the customer experience and ensure that the shopper won’t leave the website at that point. While this obviously plays into conversion, the immediate goal here is actually retention.

So, you should start by identifying what you will (or intend to) gain by providing intelligent recommendations. Larger basket sizes? Maybe. But you can also aim to boost conversions and loyalty in more ways than one, move inventory faster, and reduce operational costs. 

Now that you know where you need recommendations, what type of recommendations work best, and the purpose that they’ll serve, the question remains: how on earth do you move from strategy to action? This requires taking a good look at the data you have available to deliver the recommendations you desire. 

Machine Learning and Product Recommendations

There are different ways of enabling product recommendations, but the best-known approach is:

Collaborative filtering

This is a type of recommendation system that predicts what might interest a person based on the taste of many other users. It assumes that if person X likes Burberry sweaters, and person Y likes Burberry sweaters and Jeckerson trousers, then person X might like Jeckerson trousers as well. 

However, this approach requires an abundance of data. If you don’t have registered or even recurring customers, it may prove to be difficult to collect enough information to create the rich customer profile necessary for collaborative filtering. 

The choice to use collaborative filtering without sufficient data isn’t without consequence. Doing so can actually lead to cold start problems where shoppers are provided with mediocre recommendations that may miss the mark completely. For example, if the user is unknown, or too little is known, many collaborative filtering-based recommender systems opt for favoring popular items to a very large extent. 

This results in a specific kind of cold start problem, which is actually far less magical than it sounds: 

  • Harry Potter Problems: i.e. recommending Harry Potter to everyone just because most people have bought the Harry Potter book. 

Moreover, especially in fashion and retail, which have large product turnover or introduce whole new catalogs seasonally, further instances of cold start problems emerge: what can you do if a large block of your catalogue is new or niche and just doesn’t have any data to go along with it? 

Other approaches to product recommendations are available and might be better positioned to address cold start problems. 

Content-based filtering

This is a type of recommendation system that focuses on the products themselves and recommends other products that have similar attributes. Content-based filtering relies on the characteristics of the products themselves, so it doesn’t need shoppers to interact with products before making recommendations.

For instance, it is possible to extract insights about a product and others related to it from textual data (product description, style and fit notes, ratings and reviews, etc.), and then use that deeper understanding to immediately provide recommendations and offers complementary to the new product. This allows for the development of  associations between items that, when coupled with on-session clickstream data, can also serve as a powerful way to address cold start problems for new users.

Best Practices

Intelligent product recommendations can be very valuable. However, simply putting them in place is not enough. This must be done strategically, which requires that you determine where you need them as well as what type would work best, figure out what it is that you want them to help you achieve, and identify a method for implementation that is feasible for you given your specific data constraints. 

With this strategy in place and the right technology at your side, you’ll be amazed by just how valuable product recommendations can be in creating the personalized experiences that your customers have come to expect in a world that has fundamentally changed. 

And as this new normal takes hold, elements such as personalization can actually mean success or failure for your business depending on how they’re implemented. 

Want to start creating more personalized recommendations? Learn more about how Coveo delivers relevance in commerce.

Dig Deeper

Wonder how recommendations power impulse buying? Sarah Beckham breaks it down for you.
Want to dive into some examples?  Check out Alexandra Rioux 4 Product Recommendation Strategies.
Ready to choose a recommendation engine? Peruse our 6 Most Popular Recommenders to Entice Shoppers.
Predict what people need, before they even know they need it with Coveo’s AI-powered recommender system.
Share this story:

About Andrea Polonioli

Andrea is a Product Marketer for our Commerce line of business. Prior to joining Coveo, he was at Tooso, the acquired AI search ecommerce startup. He has a passion for innovation-driven companies and a research background in cognitive science. When he is not working, he is likely to be experimenting new dishes, travelling or hiking.

Read more from this author

About Emily Hunt

Emily Hunt is a political scientist turned Senior Content Specialist at Coveo. As a data-driven content marketer, she combines her analytical skills with a passion for storytelling to produce compelling content across all areas. If she’s not writing, she’s exploring the great outdoors, jamming out to rock ‘n’ roll, or reading Harry Potter...again.

Read more from this author