Want to implement a product recommendation strategy in your online shop? Then this is your guide. Learn why they matter, ways to use them, where to feature them, and how the pros are doing it.
“A lot of times, people don’t know what they want until you show it to them.” This famous quote by Steve Jobs clearly illustrates why product recommendations are an essential element of any online shop. They are one of the most powerful tools for increasing not only the value of the shopping basket but also customer satisfaction and repurchase rate.
Recommendations can be used in a variety of ways, making it all the more important to find the right product recommendation strategy for your online shop. What kinds of recommendations are there? What are the use cases and best practices? Does using AI really pay off? This article will provide answers to these questions and beyond.
A product recommendation strategy can increase revenue by 10%
The reason why product recommendations are so well established can be explained in two words: impulse buying. Just like in a physical store, the strategy is to put various products within the customer’s field of vision at checkout. The advantage of online shops is that you have a whole range of possible ways to spur customers into making a spontaneous purchase – from the homepage all the way to the checkout.
A much-cited report by McKinsey reports that Amazon draws approximately 35% of its revenue from recommendations. Even though this figure may seem high, hardly anyone would deny that product recommendations have made a significant contribution to Amazon’s meteoric rise.
How do product recommendations work in an online shop?
Recommendations are generated via a product recommendation engine – a software system that predicts which products are most likely to be purchased in a specific context. These predictions are often based on very different types of data such as the affinities of an individual user, the purchase history in the shop as a whole, abandoned shopping carts, suitable alternatives, and additional items or accessories for the current product.
The most important types of product recommendations can be divided into the following categories:
User-based recommendation (self-learning)
Whether you label this type of recommendation “Other customers also buy,” “Often bought together,” or a different heading, the foundation is still the same: learning from customer behavior and combined sales from the shop with the goal of automatically generating suitable recommendations. This type of recommendation is normally displayed on the product description page or even better, in the shopping cart once customers have made their purchase decision and are ready to make further decisions.
Until a few years ago, retailers often had the problem of needing the largest possible database in order to generate good recommendations. Since then, it has become possible to make appropriate recommendations with smaller amounts of data thanks to AI processes. Even from individual combined purchases, it is possible to deduce which categories usually match. This way, changes in the product catalog have an immediate effect on the recommendations shown, and recently listed products sell effectively right from the very beginning.
Apart from the combined sales of all visitors to the shop, the affinities of the individual customers also play an important role. For example, someone’s click and purchase behavior in the shop can be used to deduce which brands, colors, price ranges, etc., a particular user prefers. This kind of personalization can provide the finishing touches to your recommendations and give your customers the feeling that you really understand them.
Best practice: Walbusch
Product-based recommendation (rule-based)
This type of recommendation is perfect for product description pages. Your shop does not need to learn anything in order to implement it. Your recommendation engine only considers those products within your range that are similar to the product currently being viewed in terms of category and attributes.
The most exciting part about this method is that you can define the rules according to which similar products are to be displayed. For example, to encourage customers to buy a higher-priced product (up-selling), you can specify that only similar products at higher prices should be shown.
Best practice: Kaiser+Kraft
Expert-based recommendation (manual)
If you have the resources and don’t want to leave everything to AI, you can also ‘hardwire’ some of your products with the recommendations of your choosing, meaning that you determine precisely which recommendations should be displayed for which product. This is the most flexible approach and it can be implemented both statically (exact allocation) and dynamically (rule-based).
Two different modes are possible:
- Do not recommend – blacklisting which products should be excluded from the recommendations. For example, in a sports shop, no FC Bayern jerseys for Manchester United pants, or in an online pharmacy, excluding products that interact with the medication that is being viewed.
- Always recommend – whitelisting which products should definitely be shown. For example, product bundles, shop-the-look, etc.
Best practice: MyTheresa
Bonus: best sellers as a multi-use tool
One category of recommendations that you can use almost everywhere is best-selling products. As the name suggests, your recommendation engine only needs to know which products you have sold the most so far. Best sellers work best on generic pages, such as the homepage or zero results pages. Unless there is active personalization that has been learning from previous transactions, these generic pages reveal little about the customer’s interests. So, it makes sense to recommend items that are most popular among your customer base. Novelty items are also easy to sell here.
Best Practice: OBI
Next-level product recommendation engine
Today’s online shop systems are designed to assist with purchase decisions. However, in many industries, including online groceries, drugstores, pharmaceuticals, and B2B, on-demand purchases are customary, meaning that customers already know what they want to (re)order.
We developed the Predictive Basket for this exact purpose. It is a neuronal network that, like a recommendation engine, can make proactive purchase suggestions in order to simplify need-based purchases and reordering. In order to achieve this, AI takes many different factors into consideration. For example, it tracks the individual preferences of the customer, and the seasonal buying patterns of all customers (i.e., in online groceries, milk is reordered more often than jelly).
The AI also recognizes seasonal sales trends, such as charcoal for the grill. If a customer buys charcoal every two weeks during the summer, the Predictive Basket will show them charcoal products at the same intervals. Then, as soon as barbeque season is over, and few customers are buying these products, the Predictive Basket adapts to these changes in shopping behavior. The recommendation engine will no longer suggest related products to the customer who was previously buying barbecue charcoal regularly.
Where can product recommendations be used?
Here is a helpful matrix that shows which types of recommendations are suitable during the different phases of the customer journey:
How do I integrate a product recommendation tool into my shop?
The process and timeframe of integration differ depending on technology, shop system, and requirements. In the case of FACT-Finder’s recommendation engine, full implementation, including AI training, usually takes two to three weeks.
Checklist for your recommendation engine
To wrap up, here is a handy checklist that includes the aspects you should consider when selecting and using your recommendation engine. Your recommendation engine should:
- take multiple data sources into consideration. This should include combination sales as well as attribute and feature-related user affinities. For example, it generally pays to have various types of recommendations on the product description page.
- draw general conclusions from individual combination sales. This ensures that you don’t have to collect data for months before logical recommendations are generated. This way, your recommendation engine will achieve a much faster ROI.
- learn from customer behavior, in a GDPR-conformant way. This happens on the basis of session ID, without third party cookies. This way, your recommendation engine can work independently of cookie topics.
- enable manual optimization. In practice, it has been observed that the best results are obtained from a mixture of artificial intelligence and human knowledge.
Do you need support in defining and implementing your recommendation strategy? Our FACT-Finder team would be happy to help you. We have been supporting shop owners with the optimization of their search and recommendation strategies for over 20 years. Depending on the structure and volume of your product range, customer behavior, and product categories, we can give you a demonstration of our recommendation engine.