The Expert’s Guide to Driving Revenue with Product Recommendations

Sep 1, 2016

Online recommendations are the product of complex algorithmic decisions, which mine users’ interactions onsite and offsite, and present custom-tailored opportunities for increased engagement in real time.

Anyone who has visited an online retailer has witnessed the power of product recommendation engines first hand. These systems use aggregated data, as well as predefined algorithms, to predict and present the products or services in which individual prospective consumers are most likely to have an interest.

Effective product recommendation shouldn’t be left to the market leaders; product recommendations are now an essential requirement for any eCommerce business that wants to increase engagement, conversions, and revenues at scale.

Maximize Revenue through Smarter Recommendations
Leverage a full suite of eCommerce segmentation, personalization, recommendations and optimization solutions to drive more revenue.

HOW PRODUCT RECOMMENDATION ENGINES WORK

Broadly speaking, there are two approaches to product recommendations: Global and Personalized.

Global Strategies

Global approaches recommend the same products to all site consumers or all consumers belonging to a predefined segment. There are two types of global strategies: Generic Ranking and Contextual.

Generic Ranking: Product recommendations based on product performance, ignoring individual consumer behavior and current context. Examples include “most popular on site” and “trending now.”

Contextual: Product recommendations based on current context of individual consumer, such a product category of current page, but not based on consumer behavior or activity. Examples include strategies such as “similar items,” “bought together,” “most popular in category,” or “trending” in a user’s geo-location.

Personalized Strategies

Personalized approaches seek to identify the right products to recommend based on the individuality of the consumer.

Collaborative Filtering (Product Focused): Product recommendations based on what similar consumers have engaged with previously. Collaborative filtering algorithms collect and analyze massive datasets of user behavior and activities, and mines that data to predict who will purchase what based on their similarity to other users.

These predictions are specific to the consumer, but are based on information gleaned from many consumers. A distinct advantage of collaborative filtering is its broad applicability; these algorithms don’t need to understand the essence of a particular item, and are capable of accurately recommending a wide range of products. Further, collaborative filtering allows brands to enable serendipity shopping for their consumers by presenting items they would not have necessarily sought to purchase.

Content-Based Filtering (Consumer Focused): Recommends products based on similar items the same consumer has engaged with previously. This method relies on two factors: product attributes and affinity profiles (also known as “scores”). Affinity-based algorithms recommend items to individual consumers that are similar to ones they’ve engaged with previously. The product attributes are the information about a specific product (e.g. women’s black waterproof sandals) while user profiles are built based on items searched, viewed, added to cart, purchased, shared socially, etc.

Content-based filtering systems are considered highly accurate, as they are tailored to the demonstrated interests of individual consumers. Another advantage is the ability to leverage data collected in real time, while other strategies require data aggregation logic calculated offline.

PRODUCT RECOMMENDATION STRATEGIES

High impact product recommendations that resonate with your visitors and drive them to convert with high order values rely on several critical factors: the optimal recommendation strategy, based on the current site context and user, on the right location of the page and presented at the right time.

Different types of users at different stages of the user journey call for different types of product recommendations, and identifying the right opportunities to recommend the most appropriate products for the user can either amplify the visitor experience, push the visitor to purchase and increase the total purchase value, or dissuade and distract the visitor.

Determining the correct recommendation strategy to be used in any context or location on the website is a highly calculative effort. Look for a machine-learning recommendation engine that provides all of the following strategies out of the box:

  • Automatic– The “black box” strategy, which automatically recommends the most appropriate products considering the context, user data and more. The Automatic strategy is adaptive, and will deploy the most appropriate strategy based on data availability at the current moment.
  • Similar Products – Recommends items similar to the product currently viewed.
  • Bought Together – Recommends complementary products that are typically purchased together with the product currently viewed.
  • Most Popular Products – Recommends the products that are currently most popular on the website.
  • Most Popular in Category – Recommends the products that are currently most popular within the product category currently viewed.
  • User Affinity – Recommends products that match each visitor’s preferences and affinities, based on browsing behavior (product views, cart additions, etc.)
  • Personalized – Recommends products based on product performance per lookalike audiences and visitors.
  • Recently Viewed – Recommends the products each visitor has recently viewed.
  • Mixed Strategies – Any combination of the above strategies, as illustrated below:

multiple recommendation strategies in a single unit

RECOMMENDATION STRATEGIES BY PAGE CONTEXT

Product recommendations are ordinarily implemented on the Homepage, Category or Grid page, Product pages, Cart pages, or between any of these stages by means of overlays and popups. Different recommendation strategies can be chosen and optimized for each of the specific page contexts. For optimal results, please refer to the following guidelines:

Homepage: The Homepage, which is ordinarily visitors entry point to your website, should include either “Most Popular Products” that showcase to new visitors what the store is about by displaying the best sellers, or “Personalized”/”User Affinity”/”Recently Viewed” strategies for visitors who have returned or have exhibited previous browsing.

Category & Grid Pages: “Most Popular in Category” and “Personalized”/”User Affinity” are the only options to consider in this case. We recommend always serving personalized strategies for returning visitors, and falling back to the global, “most popular” strategies for new visitors.

Product Pages: “Similar Products” and “Bought Together” are the best options for PDPs. If only one recommendation unit or strategy can be implemented, choose “Similar Products” as case studies have revealed that such recommendations play a more integral role in visitors’ decision making process of purchasing a product. “Bought Together” is a relevant strategy as well, but usually performs best when introduced after the current product has been added to cart. If a “just added” popup can be introduced upon cart addition, “Bought Together” is the ideal strategy guaranteed to inflate your AOV.

Cart Pages: “Bought Together” is the ideal choice for cart pages, preferably to exhibit products that are slightly cheaper than those in the cart, in order to make the additional purchase decision quick and easy, without requiring much thought and deliberation. Depending on the nature of the products, and whether your site analytics reveals that similar products are often purchased together, the “Similar Products” strategy can be mixed into the recommendations using the “Mixed Strategies” option.

Other: Marketers must be able to place recommendation units anywhere else on the site—in exit popups, null search results pages, theme-based landing pages, emails and more.

RECOMMENDATION STRATEGIES BY AUDIENCE

No single strategy will work optimally for all site visitors. Some consumers may visit your site several times per week, while others may be first-time visitors who arrived via a search query.

The best recommendation engines are able to assess what’s known about a particular visitor, and deploy the most appropriate strategy based on data availability, context, user behavior and so on to solicit engagements.

For example, visitors with low data availability (“new users”) will receive generic ranking recommendations such as “Most Popular Products.” Consumers with histories of purchasing or viewing items may be served with contextually-driven product recommendations. Loyal consumers, on the other hand, have high data availability, making them ideal candidates for personalized recommendations, with “recommended for you” sections throughout the site (e.g. homepage, product pages, checkout).

Personalized Recommendations Per User Segment

An example of how automated strategy selection may serve different recommendations to various types of users can be illustrated in the table below:

Product Recommendations

Any recommendation unit can have numerous merchandizing rules worked into its logic, which allow for micro-segmentation. For example, determining that visitors in the Men’s and Women’s Audiences should only see Men’s and Women’s products, respectively while making sure that kids’ products are recommended to visitors with children in the household are important merchandizing rules that should be prioritized within your recommendation units.

MANUAL MERCHANDIZING BY AUDIENCE

Although automated algorithms have a strong track record, your merchandizers have tremendous expertise and insight into your customers. For this reason, your recommendation engine must be flexible and support manual merchandizing rules, such as pinning a specific item, including or excluding a particular set of items, from the automated recommendation results. For example, merchandisers can set rules that inform the recommendations engine to:

  • Include popular items in a product category a consumer has viewed over a set period of time (e.g. display of a popular item in “mens’ shorts” to a visitor who has viewed at least two pairs of mens’ shorts in the past 30 days)
  • Recommend highly profitable items to consumers who have purchased or added to their carts items valued at $200 or more in the past 7 days
  • Refrain from displaying clearance items promoted in other areas of the brand’s site to high-value customers
    Create more granular business rules to accomplish specific business goals, such as moving remnant stock or slow-moving inventory.

manual-merchandizing-rules

Maximize Revenue through Smarter Recommendations
Leverage a full suite of eCommerce segmentation, personalization, recommendations and optimization solutions to drive more revenue.

9 Powerful Tips, Tricks and Best Practices

1. Email Product Recommendations – Remove Price

Email marketing and newsletters have one purpose – to bring traffic back to the website. Therefore hard-data and detailed specifications that cause the recipient to think and calculate prior to having arrived back at the site should be avoided. Price, especially if it’s high, will often scare off and deter the recipient. Shiny images, high ranking, and favorable reviews will attract the recipients, and lure them back to the online store.

2. Engage Side-Door Traffic

Referral and Search traffic that arrives directly to a specific Category or Product page is often a one-time hit or miss chance – and it’s usually a miss that drives an immediate bounce, as the single product that was chosen has a very low probability of meeting the visitor’s interests. Therefore, this type of traffic should immediately be exposed to a variety of additional captivating products. This can be achieved by:

  • Introducing product recommendations at the top of the page only for this traffic.
  • Introducing an extended product recommendations zone with a “Pinterest”-like design below the main featured product. The Pinterest design encourages scrolling and further exposure to many more products down the page, and de-facto functions as product and category page hybrid.

3. Merchandize to Segmented Audiences

Demographic, geographic, behavioral and ad-hoc data about each and every one of your visitors – gender, weather, LTV, household income – are all precious data points that should be leveraged in product recommendations by incorporating merchandizing rules. If the weather is cold right now in Boston, you can create rules that will automatically recommend winter apparel to Bostonians, corresponding to the visitor’s gender.

4. Grid Personalization

No one visitor is like the other, and every visitor makes his or her product selection differently. The category page’s function is to showcase the entire selection of products in the specified category, which often reaches hundreds of products in display.

Expecting visitors to easily make a selection under such circumstances is a lot to expect – on the contrary, the process could be so discouraging that visitors will simply give up, walking away more confused than they arrived. Your site therefore offers product discovery tools like filters, sorting options and search – but still, the visitor will often remain with a wide selection of products to choose from.

The layout and sorting of the products on the page is critical – case studies reveal that over 70% of products purchased by visitors who navigated through category pages were chosen from the first two product rows at the top of the category page! Visitors who did not find what they were looking for in the top several rows and scrolled down the page had a much lower probability of making a purchase and converting.

Personalizing your category and grid pages guarantees that each and every visitor is exposed to the products that interest him or her the most, within the first few product rows of the page!

Using an API, along with the personalized/user affinity strategy, is the optimal solution for personalized category pages. The API returns for each visitor the relevant list of SKUs sorted according the visitor’s preferences, and thus allows you to automatically sort the page according to the visitors’ affinity profiles.

grid page customization

5. Sale and Clearance Items

Everyone is interested in reduced-price items, including your most loyal heavy shoppers who are happy to inflate their cart with additional sale items, together with the full priced items they’ve selected. Since sale items are relevant to everyone, they should be recommended – anywhere on the site. Contrary to some schools of thought, there is no reason to restrict sale items only to the Sale and Outlet sections of the site – doing so can significantly limit your AOV and conversion rate.

6. Similar Products in Cart Page

As previously mentioned, the default recommendation strategy for Cart pages should be “Bought Together” – complementary products to those currently in the cart.

However, your site analytics may reveal that your shoppers often purchase two or three products from the same product family.

This highly depends on the nature of the goods and is often observed with apparel, horticulture products, groceries, and more. In such cases it therefore often pays off to designate some recommendation slots on the cart page to similar products. Visitors will often grab the second and third pair of pants during checkout, and you can encourage them to do so.

7. Recommendations Everywhere!

Don’t limit your recommendations to the bottom of your Product pages; include them in newsletters, digital receipts, designated landing pages, your native mobile apps (and push-notify your consumers with a relevant recommendations when they are in the physical brick and mortar store!), product ads, navigation menus, welcome messages and exit intent popups.

Exit popup widget

8. Recommendation Layout Optimization

Product recommendations aren’t always made in a rotating carousel laid out in one row – there is no one template, and optimizing the look and feel of your product recommendations can make a world of difference – especially in mobile web and native apps.

Marketers need the flexibility to alter, change and customize the layout and functionality of every widget. Look for a recommendation engine that allows you to incorporate tests and personalization tactics that position the recommendations in different areas of the page for different audiences, and alter the layout according to visitor type.

9. Recommendations Experiments & Testing

A/B testing different recommendation strategies or different settings/merchandizing rules will often come in handy. Aside from exploring the “Mixed Strategy” option, real A/B testing of recommendations is essential to understanding which recommendation strategy is working best.

CONCLUSION

Recommendation strategies vary greatly from one another, and selecting a recommendation engine should be driven by your goals. Make sure you look for a recommendation engine that allows for flexible strategy setting, testing, and advanced settings to leverage your merchandizers’ expertise so you can extract the highest value from your recommendations and realize your highest ROI.

Maximize Revenue through Smarter Recommendations
Leverage a full suite of eCommerce segmentation, personalization, recommendations and optimization solutions to drive more revenue.

Rate this post

  • Great guide and tips to drive the revenue on product recommendations.

Menu Title
Contact Us
×