Personalized Product Recommendations:
Five pitfalls to avoid when optimizing conversions
Product recommendations are a standard component in personalizing the online customer journey. They can both improve the customer experience and increase conversion rates and average order size – when implemented correctly.
The bad news is that numerous pitfalls can significantly degrade the value. The good news though is that it is easy to avoid these pitfalls if you are aware of them.
Read on and learn about the five biggest pitfalls and how to avoid creating bad user experiences and leaving additional sales unrealized.
Different recommendation types serve different purposes. Therefore, it is also common to have more than one recommendation module on a page, i.e., a module showing “Others have purchased” for cross-selling, as well as “Others have seen” for upselling.
However, when you have more than one module on the page make sure that the same product appears only in one of them – and of course in the module, where it is most relevant.
If the same product recommendation is displayed in multiple modules on the same page, you are wasting screen estate — an important screen estate that could have been used to display another relevant recommendation.
According to The Danish Chamber of Commerce (Dansk Erhverv) 2022 will be the year when more online purchases are made with mobile devices than PCs. Because of the limited mobile screen sizes product recommendation modules often only display one or two products on these devices making it even more important not to waste space by displaying the same products.
Example of wasting screen estate from a hobby store. Two recommendation modules on a product details page, both recommending the same products.
Example of wasting screen estate by displaying the same products twice in two recommendation modules on a clothing store.
It sounds like a simple and trivial pitfall to avoid – but nonetheless you see shops that have fallen into it all the time: of course, if the user has already placed a product in the basket, this product should NOT take up space in the recommendation modules during the rest of the customer purchase process.
The recommendation engine should always know the current content of the basket, and thus make sure that the recommendations do not contain products that are already in the basket.
A Danish wine store recommending the same product that has aldready been added to the basket.
Products that cannot be purchased, e.g. if the product is sold out, should normally not appear in recommendation modules. This one can be a little more tricky if the merchant offers in-store pickup and if the product instead can be reserved or ordered for pick up in a physical store, an exception can be made in some cases.
Similarly, in B2B shops, where customers often have customer-specific assortments, it is of course also important that no recommendations are displayed, that the current customer is not allowed to buy.
The recommendation engine should allow you to set up data-based rules for the items and content recommended — often called Merchandising. Merchandising means the ability to influence the results of the recommendation engine, based on rules and data about the user and the goods.
This way, i.e., you can create a merchandising rule removing (or burying as the merchandising term is called) all products from recommendation modules that are not currently in stock.
Half of the visible product recommendations are not available in this example from a Danish grocery store.
If you don’t have variants in your product catalog, you can safely skip this pitfall, but if you sell clothes, shoes, furniture, or some other type of product that is configurable or has variants, then it can be crucial for the user experience.
If you sell shoes, several data attributes will be defined at the style level, e.g. the shoe name, brand, and sales price, and some data attributes will be defined at the variant level, e.g. shoe size, stock level, and SKU. If the product is a sofa, the product configuration is often even more complex with dimensions such as color, fabric, legs, filling material, and others.
When the recommendation engine “understands the configuration structure”, you gain the following benefits:
- You will never see recommendations of the same style, just in a different size – it is not very relevant to recommend e.g. size 40, of the same shoe that you are looking at in size 38.
- In all recommendations, the recommendation engine will be able to indicate which variant is most relevant. If you are looking at a kid’s shirt in size 116, for example, you will be recommended matching pants also available in size 116, or when looking at a 2-seater sofa you will be recommended the armchair with matching filling material, matching grey fabric, and matching oak legs.
- You can avoid showing alternatives (and hereby wasting valuable screen estate) that are not available in the desired variant. Why show other shoes that are not available in size 38, when that is what the customer is looking for? It just provides a poor user experience.
The recommendation engine should support the data model for your product catalog, whether you’re selling shoes, furniture, or something else entirely. If your range consists of a mix of products that have variants and other products that do not have variants, it should also be supported.
The Relewise personalization engine supports the data model, and can select the most relevant variant when recommending products.
This last pitfall is about how to avoid dead ends in the customer journey by addressing the recommendation types displayed and when they are displayed.
On the product detail page, the recommendations should be used as a relevant guide and as inspiration for the user in their buying experience. Often the product details page is also the landing page, and even if the product is not exactly what the user fancied you want to avoid the user clicking on the browser’s “Back” button, as it will take them right back to the Google search results page leaving your shop.
There are a lot of different types of recommendations targeting different scenarios. So far, we have touched on two types of product recommendations: “Others have bought” and “Others have seen”.
Often it will be good to show modules with both these recommendation types on the product detail page. “Others have bought” recommends other products that go well with the product the customer is looking at, and “Others have seen” recommends products that can replace the product you are looking at.
In addition, there are also recommendation types that can look at the current basket and recommend products based on that, recommendation types for popular or trending products, recommendation types based on the customer’s previous purchases, etc.
Here are a few examples including which recommendation bands to choose:
- If the product the customer is looking at cannot be purchased online, e.g., if the item is sold out, one should look at prioritizing the recommendation module that shows alternative items. Here, of course, remember pitfall #3: “Avoid recommending sold-out items”.
- Once the customer has put an item in a basket, the power step should choose the recommendation type “Others have purchased”, and again remember pitfall #2: “Avoid recommending items that have already been basket”, pitfall #3: “Avoid recommending sold-out items” and finally pitfall #4: “Avoid recommending the same item, just in a different size”.
The recommendation engine should allow you to gain insight into how each recommendation module performs on each page. For example, you should be able to see how many products and how much revenue is added to the basket from “Others have seen” and “Others have bought” on the product detail page, and you should be able to see how much of the revenue comes from products added to the basket from the recommendations on the power step.
Baymard Institutes has a great article where you can find more inspiration and suggestions about how to improve the relevance of cross-sells in the cart.
Obviously, the algorithms used to recommend products are also very important for the value product recommendations bring – both to the merchant and the customer experience.
Product recommendations can contribute to a significant increase in both conversion rates and average order value – if implemented correctly.
By avoiding the pitfalls above you will be sidestepping some of the biggest conversion killers when it comes to product recommendations and in doing so be well underway to both increase revenue and customer satisfaction.
