Understand the different technical means for recommending products and their results for your online store

An online store is like any other website or mobile application in the sense that it’s success is measured by the amount of visitors it has and the level the engagement they have with that website or app. Your marketing efforts will determine the growth of your visitors and the combination of your user experience and content quality will influence heavily the level of engagement.

Product Recommendations directly influence the user experience in that they help all types of users, new and returning visitors, as well as existing customers, who are in “exploration mode”, “comparison mode” or “buying mode”. Let’s begin by understanding the technical possibility to “recommend” products.

“Recommending” means that either a human or an algorithm selects a chosen number products and displays them to the visitor in a chosen container, often called “widget”, at a page of your choice.

The process “selecting” of selecting is where product recommendations differ severely from one another. There are many other aspects such as configurability, ability to measure the ROI and styling of the recommendations that characterise different solutions for recommending products. We will go into the details of each of these aspects in future posts.

In our next post in the “Product Recommendations” category we will take a look at how recommendations work specifically for WooCommerce.

For now let’s only focus on the technical means for recommending products that you as a owner of an online store have. They are applicable to almost any eCommerce platform out there.

1. Manual

In this case the store owner manually “links” products by specifying the two ends of a “1 to many” relation between one individual product and other products in his store. One selected product may have zero, one or many other linked products in a “Up-Sell”, “Cross-Sell”, “Similar / Relative” or other pre-defined relation. The store owner then goes to each and every product and manually chooses a few “Cross-Sell” products, a few “Up-Sell” products and a few “Similar / Relative” products. By creating that relation he defines what products will appear in the “Up-Sell”, “Cross-Sell” or “Related items” widgets of his online store.

Example 1: “A laptop seller offering a customer a mouse, pen-drive, and/or accessories.” In this case a person has manually selected a few complementary products that should appear on the product page and cross-sell additional products such as accessories.

WooCommerce Manual Recommendations

WooCommerce’s Manual Recommendations screen

Pro’s:

  • The selection can be extremely precise. For example you can select an Apple mouse to cross-sell with a MacBook Pro.

Con’s:

  • The selection process is very time consuming. You have to do this for when cross-selling, up-selling, when offering alternative products or for every other different recommendation you want to make.
  • Your selection is static and you assume it will be suitable for every customer. Meaning that if I have already bought the Apple Mouse and come later to your store to buy an iMac you will most probably recommend me the same mouse again.
  • Your selection might be empty for that particular product. In this case you will have an empty widget, which is a negative experience for the user and lowers the perceived quality of the store.

2. Automatic

In this case the store owner defines a widget which should automatically select products according to a, often unclear for the store owner, logic that decides which products fall into the “Up-Sells”, “Cross-Sells”, “Similar / Relative”, “Best Performing” or other item categorisation. This is a fully automatic selection of products that relies on simple programmatic queries in the product database. Beware. Many such automatic recommendations often result in massive inefficiencies and lost benefits for the store owner. Because they are easy to implement and don’t require deep knowledge or technical resources often times people or platforms select items by simple “random”, “first item in a category”, “newest items” or other generic queries and use them for cross-selling or for displaying featured items such as “best performing products”.

Example 2: “A laptop seller cross-selling complementary products by picking a random items from the accessories category.” In this case a Macbook Pro might be cross-selled with earphones and laptop bags.

PcStore's Automattic Recommendations

PcStore’s Automatic Recommendations from the accessories category

Pro’s:

  • Selection is quick and requires no maintenance.
  • It makes it possible to feature products that rely on user actions such as purchases, for example “most bought”.
  • It allows for a more dynamic storefront displaying large collections of products such as “newest from category X, newest from category Y” and others.

Con’s:

  • Selection criteria is often limited to default programmatic queries such as last, newest and random items from categories and tag groups. No possibility for discovering true dependencies between products.
  • Hard to customize. The “most bought” widget might end up showing always the same couple of top selling products. If you have an item that is bought in high quantities you might end up having that product featured all the time.
  • Product recommendations often show no or unclear selection logic. This can decrease their efficiency and the client’s belief in the quality of your online store.

3. Automatic and Personalized

It means that you are able to automatically display product recommendations tailored to every individual user’s behaviour. It also means you have to turn to an external provider for a specific solution, to a plugin provider for a recommendation widget or to a company specialising in personalized recommendations for eCommerce. This technology is not anymore available only to big online retailers, but through advances in technology it is now available to stores of all types and sizes.

This is a fully automatic selection of products that relies on framing questions in the form of mathematical equations and searching for the right answer by means of statistical significance.

Through these mathematical equations you are able to answer very detailed questions on behalf of the user such as “What products have other users viewed after viewing this particular item?” and “What products have other users, similar to me, viewed after viewing this particular item?”.

Example 3: “A fashion clothing store is cross-selling complementary products by displaying what similar visitors are viewing after they view the current product.” In this case statistically a leather jacket might be viewed most often together with leather shoes. Where as answering the same question by first comparing what users are similar to me would yield recommendations that are more relevant to me personally. In this case say I am a sneaker fan and have never viewed leather shoes on your store. The same widget that displays recommendations by answering the second question would find what sneaker lovers have viewed together with this jacket and return much more relevant products that are tailored to me personally.

Amazon's Personalized Recommendations

Amazon’s Personalized Recommendations

Notice the difference between the two questions in the condition “similar to me”. This is where personalization kicks in. The first question can be answered by statistically determining what is the most viewed product right after the current product from visitors of your online store. The second however, will return products that users with similar browsing or purchasing history have viewed right after viewing the current one. The first set of recommendations over time will remain similar if not identical for each product. The second, will display recommendations for the same viewed product that are different for every individual person. They go by the name of “personalised recommendations”.

Pro’s:

  • Selection is quick and requires no maintenance.
  • It makes it possible to feature products that rely on more versatile user actions such as views, ratings, add to carts and so on. Imagine “most viewed”, “best rated”, “similar to me”, etc.
  • It allows for a more dynamic, personalized storefront on any given page of the store.
  • When used wisely can increase engagement and revenue significantly

Con’s:

  • Building it Involves a dedicated team of data scientists and expensive infrastructure.
  • Using an external service provider is often costly and requires upfront cost
  • Setup takes time and thus the true effects are visible after weeks, sometimes months
  • Hard to prove that the service’s ROI for the store owner

 

Two more notions about recommendations should be explained.

Cross-selling vs. Up-selling

There is a general overlap between cross-selling and up-selling. Cross-selling and up-selling are similar in that they each offer customers additional value than what they would have otherwise received had they bought what they were initially looking for. Cross-selling generally occurs when the store has more than one type of product to offer visitors that might be beneficial to them. In up-selling the online store is selling a higher-end version of the product the customer originally came to buy.

Expected results of recommendations

Their main result is that they expand and improve the browsing or buying choice the user has at any given moment in the customer journey. The immediate effect is most often seen in the direct increase in the average number of viewed items, in the average session time, in the product conversion rate (how often a product view results in a product sale) and therefore subsequently in increased revenue. For example, Perpetto tracks all these separate indicators to give you an easy to understand and detailed view of the quantifiable improvement our recommendations generate for your store.

Perpetto Dashboard

Perpetto’s Recommendation Metrics

We hope this was helpful to you. Perpetto’s team work’s hard to eliminate all the of the con’s mentioned above and to give you all the benefits of personalized recommendations. If you need additional information or consultation, don’t hesitate to contact us. You don’t have to be our customer to get our advice. You can find us on Twitter or just send us a message here.

Happy recommending!

About the author
Alex Kitov is the CEO of Perpetto and passionate about data, UX, tennis and building data-driven products that help eCommerce professionals grow their online business.

Ready to boost your online sales and retention?