In the “old days,” When people needed to buy something (think washing machine or lawn mower), it was all about asking trusted friends and acquaintances for their advice and product recommendations. We would ask if they had ideas based on their own experiences, or whether they could point us to someone who could help us feel comfortable that we were making the best choice. 

Today, people are still more likely to buy things personally recommended by a friend, but now you can also just browse online and get a bunch of different recommendations from your new virtual friends, without ever having to bug your actual friends. And it’s a good bet that these helpers may “know” you even better and be able to give you more-accurate recommendations than your actual friends. It’s a win-win.

Recommended technology

It’s staggering how much the retail ecommerce industry has grown. It initially took off because people had to stay home and shop online, some for the first time, but the convenience and speed many consumers discovered online made for some monumental lasting changes in the shopping habits of millions.

As online shopping functionality has been getting more firmly entrenched, online product recommendations have been busy becoming the gold standard for customer service. People can’t get enough of personalized suggestions in ecommerce stores, and for good reason: artificial intelligence “knows” them and what they like incredibly well, and it doesn’t hesitate to speak up. 

With so many retailers competing for consumers’ business, and an abundance of product information to sift through on some companies’ sites, effective product recommendations have become one of the best ways, if not the best way, to create a great customer experience that keeps shoppers coming back. 

A recommender system is like a map that highlights unknown territory for shoppers to explore. It can also open people’s minds to new possibilities that they might love. It suggests the best products based on people’s preferences and gives them access to a wealth of “insider” information about what other customers like.

The concept of free choice is so important to us that it’s cited in the First Amendment, but studies show that more choice isn’t always better. Contrary to what might seem like conventional wisdom, sales are better when a limited number of product recommendations are made. In the now-famous Jam Experiment, researchers found that offering grocery-store shoppers only 6 flavors of jam to choose from as opposed to 24 led to more sales; people were found to be overwhelmed by having too many choices in front of them, and rather than rack their brains and attempt to choose from a huge selection, they simply lost interest.

One result of this research is that when it comes to selling products, marketers are getting smarter and closely tracking site metrics. With advancements being made in data capture and attribution, product embedding, and multi-device tracking, product recommenders can efficiently optimize and suggest what customers want with pinpoint accuracy. In short, the world of online shopping is becoming much more personalized. 

Recommendation engines can even learn about users’ preferences in real time as they’re browsing, and respond on the spot by, for instance, offering a discount based on their user behavior. 

Recommendations help conversions

Due to their track record for improving conversion rates and sales, product recommendation engines are one of the most popular applications of data science. And they can be very lucrative: recommendations account for up to 31 percent of ecommerce revenues (Barilliance, 2018). On average, 12 percent of customers’ overall purchases are attributed to them. 

Offering recommendations can also improve the average order value of what’s in a person’s shopping cart. If a retailer can use cross-sells, for instance, to generate customer interest and ship a bunch of items instead of just one, their profit margins improve.

“The key reason why many people seem to care about recommender systems is money,” points out data science expert Yanir Seroussi. “But this is the more cynical view of things. The reason these companies (and others) see increased revenue is because they deliver actual value to their customers — recommender systems provide a scalable way of personalizing content for users in scenarios with many items.”

Regardless of the motivation to use this technology, ecommerce sites that utilize recommendation engines can drive conversions by suggesting:

• Additional items that other shoppers have bought with the item the customer is looking at
• Items that align with the user’s search query content
• Relevant products that “go with” an item they’ve put in their cart

Here are some online retailers that have generated significant sales from using and refining recommendation engines to improve their merchandising.

Amazon

Amazon.com has been researching recommendations technology since 2003; it began applying its findings long ago and is still the pioneering industry leader. 

When you’re on an Amazon product page and see “Frequently bought together,” “Products related to this item,” and “Customers who viewed this item also viewed,” you’re being served relevant upselling recommendations. These groupings of complementary products may sound like substantively different types of information, but these recommended products are all aimed at the same goal: giving you a rewarding customer journey that ends up making you buy more stuff. 

Amazon’s sales apparatus works exceedingly well. According to McKinsey and Company, 35 percent of what people buy on Amazon comes from product recommendations based on sophisticated algorithms and predictive models. 

Netflix

People often comment that subscription-based Netflix “knows them well” in terms of their movie and show preferences. That’s because the hybrid recommendation engine generating its personalized product recommendations continually takes note of the behavioral data of its media-hungry customers in order to improve over time and keep them satisfied. According to Business Insider, the Netflix recommendation engine saves the company $1 billion a year.

Best Buy

Since 2015, this mega retailer has used an ecommerce recommendation system that collects and analyzes shoppers’ querying and clicking data. It then makes suggestions based on the information gleaned about people’s browsing and buying patterns. Best Buy’s executives credit its recommendation system with a 24% increase in online sales (CNBC). 

Spotify

According to Bloomberg, this music-streaming service’s algorithm-based recommendation system, which provides an updated “Discover Weekly” playlist and “Release Radar” (new releases) geared for its listeners, is responsible for helping the company increase its number of monthly users from 75 million to 100 million at a time.

YouTube

With YouTube, everything revolves around customer engagement, so relevant content recommendations are regularly updated to reflect users’ activity and call their attention to items they might like. Netflix management says recommendations are responsible for roughly 60% of the clicks people make from the home page to videos. 

What’s under the hood?

Relevant recommendations magically appear when we’re shopping online, but what’s happening in the back end to make all of that happen?

In a nutshell, recommendation-engine technology uses machine learning and artificial intelligence to generate product suggestions and offers. The data collected is analyzed and used to create customer profiles. When the profiles are available, they’re used to help generate particular content or highlight types of products that seem well suited to customers based on their interests. 

Recommendation engines work by utilizing sophisticated recommendation algorithms that are applied to customer data, including:

• Browsing history
• Buying patterns
• Feedback people have left
• Most-viewed products
• Preferences
• Earlier purchases
• Recently viewed items
• Search history
• What’s in the shopping cart
• What’s on wish lists
• Purchase history

Recommendation-creating data about shoppers is collected implicitly (by watching what they do, look at, and choose to buy) and explicitly (through information they willingly provide online, such as ratings and product reviews).

Then, algorithms are used to enable automatic generation of relevant recommendations based on the data, and the most relevant suggestions are displayed. Recommendations are updated when new customer data is collected and analyzed.

Recommendations data can be applied not just on product pages but in apps, in marketing content, in search, and even in ads on other web sites (a practice that people often find “creepy;” as they’re instantly made aware that they’re being “watched.”

Recommendations data associations are made and understanding is gained based on three types of relationships between items and users:

• A user-to-item relationship: which things and types of items do people like?
• An item-to-item relationship: which items are focused on the same type of subject matter, written by the same author, have a similar design, or are in the same genre? 
• A user-to-user relationship: which people have similar backgrounds (for instance, similar users are the same gender or have similar preferences)

In addition, for analysis purposes, retail ecommerce recommender systems rely on these key types of data:

• How users behave online: how do they rate products, what do they say in reviews, what are their habits when clicking and browsing product pages? What do they typically buy?
• People’s demographics: What’s their income and education level, gender and age, ethnicity; where do they live? 
• Attributes of products: details such as size, weight, color, feel, and product ID number

Filtering methods

When substantial amounts of data have been collected, the information goes through the recommendation engine’s product filtering system. The data can be filtered in three ways:

• Collaborative filtering: focused on user behavior, activities, and preferences in order to predict preferences based on how much they’re like other users
• Content-based filtering: focused on product features and item similarities
• Hybrid: utilizes techniques of both collaborative and content-based filtering

The data can then be used in a variety of ways to enhance the shopping experience, such as by:

• Personalizing search and category page results
• Recommending bestsellers on the home page
• Recommending similar products when a search has no results
• Recommending frequently bought together and related products (e.g., items from the same brand) on the product detail page
• Highlighting relevant promotions and offers
• Recommending appropriate items on the cart page

A dynamic situation

Many algorithms serve recommendations based on what a shopper has been buying recently. However, the best kind of recommendations are dynamic: the engine keeps track of people’s real-time activity on the ecommerce site and instantly adjusts what it shows them in response. 

Dynamic recommendations factor in which products shoppers are looking at, which ads they’re clicking, and which categories they’re browsing. In essence, the experience is like that of having a sales clerk listen to your feedback and then go get additional items for you to try on or consider. 

Who knows what recommender systems will be like in 5 or 10 years? 

It’s safe to say that they’re vital tools for creating engaging, conversion-producing customer experiences, and their popularity is only going to increase, leading to higher customer retention and loyalty as consumers become further spoiled by their smart online help. 

The only question is what will happen to the ancient practice of asking our real friends for their recommendations? Will we bother? And if we do, perhaps their suggestions will be along the lines of “Check Amazon; there are a bunch of good new products trending” or “I know an online store that has great recommendations for those.” 

Want an ecommerce site that enhances your customer satisfaction with the right personalized strategies? Contact us today.

For more information

Introducing Agolia Recommend

Algolia Recommend