As we head into the holiday season, etailers everywhere are looking at ways engage customers and boost sales. Personalization tactics can be an etailer’s best friend -- helping you in the primary goal of getting an initial product in the cart and purchased -- but to take it to the next level, drilling down into advanced tips for optimized personalization can make a big difference.

From your product metadata to the way you group, filter and display your products, there are a variety of things that will ensure that your personalized recommendations are optimized for success. Below are four advanced personalization tactics that will help you pinpoint the sweet spot between automated and manual merchandizing techniques.

1. Begin with the End in Mind -- Catalog Attributes

The critical first step for creating relevant and personalized product recommendations starts with your catalog. A consistently and thoughtfully organized catalog is easier to understand by both machines and merchandisers.

The product attributes are the key. The attributes in your catalog should be consistent with the way you represent products on your site and expressed in a way that humans can understand.

Catalog Attribute Tips

  • Catalog item attributes should match the attributes on your website
  • Letter case and punctuation should also be the same. It’s about having exact attributes and unique case-sensitive values. For example, a sub-category should not have the values of thermals, men’s thermals, Mens Thermals and Men’s thermals. While these are the same thing, systems will view them as four different values.
  • Don’t assign multiple attributes to the same value. For example, you may not want to allocate the value “women” to Department, Parent Category or Category if you only have one over-arching “Women’s” top-level hierarchy.
  • Group individual catalog items into categories such as: price, title, image, category, category path, keywords, availability, sale price, brand, size and color.
  • Assign attributes based on star ratings or reviews. If there is a shirt with one rating of 5 stars and another shirt with 50 ratings of 4 stars, which would you use to rank one over the other? People are skeptical of one rating whether it is great or horrible, but it’s harder to refute 50 people with the same opinion. With ratings, you need to consider both quantity and quality.
  • Include non-product pages such as category pages as attributes in your catalog feed where category attributes are matched to a product attribute. For instance, if the category page tells us ParentCat=Furniture and all the product pages of sofas and chairs also have ParentCat=Furniture, this information ensures that only products that belong to furniture show in this category.

2. Filtering

A filter is a type of rule that tailors the recommendation set shown to consumers based on known attributes. For example, an automated system may think that the best recommendation to show on a page with dresses is a pair of pants or jewelry, but the merchandizer only wants other dresses to be displayed.

A category filter states that only products from within the same category can be displayed. This filter can be used across multiple categories and maintains the merchandiser’s desire to show in-category products on product pages. Rather than writing rules for every product or each category, a general category filter is a smart way to tune the recommendations to make everyone happy.

Filters can also be used to display accessories or cross-sells. Keep in mind that the personalization system is returning results that it thinks the consumer will be most interested in. So, the merchandiser can set up a cross-sell by creating a filter to show products that are not in the current category or products from a different category. It’s important with filtering not to get carried away. If you apply too many filters, you will be essentially hand-merchandising and defeating the purpose of an automated recommendation system that utilizes machine learning.

3. Content Groups

Content groups are groups of products or categories that share a set of common attributes which are not currently associated together in your catalog or site, for example, luxury brands. Essentially, you are creating a new attribute that multiple categories can roll up to that is useful for merchandising. Creating content groups helps to simplify rules and filtering.

When creating content groups, think about the easiest way to use your attributes and combine them into groups so that your rules become dramatically simpler. For example, it’s much easier to create category filter such as “Category does not equal Group,” then to manually input “Category does not equal Bikes,” “Category does not equal Shoes,” etc. across your site.

Using a content group allows you to cluster results to particular areas of your website. For example, if you wanted to treat luxury or high-end brands differently than other brands, you could create a “luxury group” with those brands. You can then use this group to ensure luxury items are only recommended along with other luxury items. Thinking ahead and creating content groups keeps you from recreating the same action for each brand individually, saving time and improving manageability.

4. Diversity Filtering

Diversity filtering forces your personalization system to surface a diverse set of recommendations, typically across categories or brands. For example, if you sell products that are generally bought with complementary products, you would apply a diversity filter based on category or subcategory which would result in a broader set of recommendations.

For example, for laptops the most popular accessory is a laptop case, but in addition to displaying the best-selling cases, you want to display other laptop accessories. A diversity filter ensures that the best-selling laptop cases as well as other laptop accessories will be displayed.

As with filtering in general, it’s important to filter judiciously and test quality rigorously so that you don’t unintentionally hamper the ability of your automated system to offer dynamically generated recommendations. That said, filtering for diversity will save you a lot of time over what you might have attempted with manual merchandising.

Editorial Note: Dan has shared other thoughts on this topic in 4 Personalization Techniques for Improving Average Order Values and Conversions