Recommendation engines are everywhere – from Tinder to Netflix. And you may have one in your Ecommerce stack too. But is it annoying or helping your shoppers? We’ll go through some common techniques to help guide you to the best recommendation experience.

Virtual Personal Shopper

If you’ve been in Ecommerce for more than a few minutes, you’ve likely heard the stat that more than a third of Amazon’s purchases are a result of product recommendations. Conscious of it or not, recommendation engines have likely been your constant shopping companions every time you purchase online.

Recommendations can be made at any time: 

  • From the moment you log onto a site (ala Netflix) 
  • In the search bar (with drop down suggestions)
  • After you have selected something to purchase (you might like this too!

In fact a whopping 80 percent of entertainment consumed on Netflix is because of recommendations. That shouldn’t be surprising, says Louis Tetu, CEO of Coveo. “Netflix wants you watching – not searching.” 

Why Are Recommendation Engines So Powerful?

So what are these recommendation engines? And why are they so powerful?

Amazon’s recommendation engine predicts what users might be interested in and guides them to products, services, and even information. The beauty of this has been users are exposed to items they might not have known about – but may find the perfect match regardless.

With a goal of getting users to swipe right – whether selling romance or stick-on-tile – recommendation engines look at widely varying data sets, but in a narrow set of approaches. 

Predictions Based on Ratings

Think of the recommendation engine as a ratings engine. The higher the rating the more likely it will be “liked” by a consumer.  And while ratings can be explicit or implied often they are binary:

Purchase = 1     No purchase = 0

To generate meaningful recommendations, engines must have a fair amount of data to determine preferences. That data is then filtered accordingly, so that users will be exposed to products they are most likely to buy. 

So what kinds of data will predict behaviors? Previous buyers’ histories is one of the most common types of data. This is also true of Tinder, by the way; the more someone has been swiped right on, the more power they have as a data point for other people on the app. 

Collaborative Filtering Vs. Content-based Filtering

Many recommendation systems, such as Amazon’s and Netflix’s, work by aggregating the behavior histories of all of their shoppers

They analyze that data so that the system can predict behavior of what others similar to you have done in the past. 

Marketers call this persona-based personalization. Data scientists call it collaborative filtering.

Advantages to collaborative filtering 

  • Gives a great starting point for most visitors
  • Provides serendipity – a way to explore and find what you weren’t looking for

Disadvantages to Collaborative Filtering 

  • Cannot handle new items in the catalog (since no one else has looked at it)
  • Does not include variants or accessories for queried items
  • Requires a lot of behavioral data
  • It also may turn out to be just plain wrong

Collaborative filtering/persona-based merchandising work-ish. But it can also introduce a lot of noise. I bought a unicorn blanket for a friend’s new baby from an online retailer. Now I am constantly deluged with unicorn baby items – across all of my devices. 

Unicorn items

Reduce Annoyance of Unwanted Products

The best recommendation engines avoid annoying customers  in this way through content-based filtering. This real-time analysis of the shopper’s behavior offers the most precise and personalized way to handle recommendations. 

Prior user history is considered, but is weighted much lower than what a shopper is doing while in the current shopping session. 

In content-based filtering, everything the shopper is looking at counts, including product attributes – such as size, description, color, and price point. 

An effective recommendation engine is like an astute personal shopper. The customer is happy because it is easier to make smarter choices, faster. And the retailer is happy because the recommendation has driven more sales.

Advantages to Content-based Filtering

  • Can provide truly personalized recommendations
  • Does not rely on aggregate user-behavior data
  • Reduces cold start problems with new products

Disadvantages to Content-based Filtering

  • Requires a good amount of product data (brand, variants, colors, size, descriptions, accessories, etc)
  • Can limit recall if narrowly tuned to user search

Machine Learning Recommendations

Recommendation engines have been a highly manual process involving intensive analysis of shoppers’ behavior that then gets hard coded into rules. Think of it as “if-then” situations from school tests. If shoppers buy baby blankets, they will want more baby stuff.

The rules may work some of the time. However, it’s not unusual for legacy systems to have literally thousands of rules. And you likely won’t know when they conflict.

The best recommendation engines then are based on advanced machine learning, says Simon Langevin, director of product management for Coveo.   

Hybrid Approach: Collaborative and Content-based Filtering

Recommendation engines don’t just feed on data about user behavior, though. They also savor all of the rich information in product catalogs and descriptions. If you are looking for women’s golf slacks, size 4, the attributes of the slacks you have selected are identified.

In this case,

Womens, size 4, golf, slacks

If you then were to search for gloves, the prior attributes would dynamically order the autosuggest to:

Sporting gloves → golf

[Note: You can see dynamic autosuggests in detail by checking out Jacopo’s blog on how to grow a product tree.]

It’s only recently that machine learning has enabled recommendation engines that can suss out these more subtle relationships. With natural language processing and understanding, it’s possible to “vectorize” the catalog. 

Understanding the product catalog is ideal when launching a new product, line of business, website or app. If, for example, your retailing website has a new, cutting-edge TV that’s debuting, you want it to show up in recommendations before users have bought or even browsed it. This is a great solution to the cold-start problem.

Content-based filtering on product catalogs are so useful, in fact, that recommendation engines can work solely on catalog information when there’s an absence of user behavior. 

Of course we don’t advise that. The Hybrid approach is best, because all that rich buyer history from all touch points should never go to waste!

Dig Deeper

In this 5-minute read, learn the economic impact of recommendation systems in Ecommerce.

About Diane Burley

Diane Burley is an expert on what it takes to stand out in the digital age. She showcases her expertise when writing about the necessity of AI, NLP, and intelligent search in DX projects. She cut her teeth as a reporter, is a long-time technologist, and knows how to tell a story in a gripping way

Read more from this author