How are recommendations generated?
There are two fundamental methods of creating product recommendations: manual or automated. Manual recommendations are developed based on the expertise of the retailer’s product merchandising staff. Each product entry must be hard-coded with recommendations for related products. This approach is incredibly resource-intensive especially for a store with a large or variable product catalog. This method is also unable to make recommendations based on the specific purchase history of individual shoppers.
Automated systems provide much more robust recommendations by leveraging a combination of data inputs:
- Historical sales data for all customers
- Individual shopping history (for customers identified via log-in or cookies)
- Product category and other attributes
- Pre-defined business rules
These data inputs are then processed in real-time with sophisticated algorithms that can produce a variety of different shopper recommendations. The resulting product recommendations provide one-to-one personalization, meaning they are unique for every customer and relevant to every stage of the shopping process.
What are the different types of recommendations?
Product recommendations can be categorized into five main types:
Cross-Sell: These recommendations highlight relevant complementary products. They are based on the aggregate sales history of the item currently under consideration by the shopper, whether a known or anonymous visitor. For example, recommendations for a shirt might include pants and jackets.
Up-Sell: These recommendations show comparable (i.e. alternative) products based on common attributes and allow you to present more of your product catalog to the shopper. This is helpful for browsers who have landed on a product page, but find it’s not really what they were looking for. To go back to the shirt example, similar recommendations would include different shirts – as opposed to cross-sell which might include pants and jackets.
Personalized: These recommendations show items that the shopper will likely buy. They are driven by the shopper’s individual buying habits and not by any specific product. For this reason, they are best used when the consumer first arrives at your site (i.e., a customized landing page) or in non-product specific email (i.e., monthly newsletter). They are one-to-one, meaning they are unique for each shopper.
Personalized Cross-Sell: These recommendations are related to the shopping experience and influenced by the shopper. Technically, personalized and cross-sell are combined to present a known shopper with recommendations that are related to the item(s) being viewed or purchased, and personalized to the shopper based upon prior buying habits. Because these recommendations are related to the product(s) and personalized for the shopper, they are very powerful!
Blended: These recommendations are used when a shopper is done with a shopping experience. They are available when there are one or more products and a known shopper. The recommendations are cross-sell OR personalized. As such, the recommendations may or may not be related to the viewed or purchased product(s). The system will display the most relevant recommendations, taking all known information into account. They are good in shipping pages, or user admin pages (if there are items in the cart).
Top Sellers: These are overall top sellers, or top sellers in a specific category. They are usually provided by the ecommerce platform, but can also be generated by the recommendation solution.
What is the best way to use these different types of recommendations?
There are numerous shopping scenarios for which you want to select the most appropriate product recommendations:
Product Page: When a shopper is looking at the product page, the shopper is always interested in other products that can be bought with that product (i.e. cross-sell, and personalized cross-sell if the shopper is known), and potentially interested in similar (i.e. alternative) products. Your recommendation vendor should enable you to show similar and/or cross-sell products. We suggest starting with cross-sell since items that will be bought with the product will increase the number of items in the cart and the average order value.
Shopping Cart: Each time the shopper adds a product to the shopping cart, they are interested in products bought with the added product and/or all products in the cart. Your vendor should enable you to recommend products that are bought with the added product or all products in the cart, at your choice (i.e. personalized cross-sell).
Checkout: When the customer is done with their shopping and checking out, they are ready to receive recommendations of products they are likely to enjoy (and buy), whether or not they are related to the items in the cart. Your vendor should enable you to show such recommendations (i.e. blended). If you decide you want to keep the recommendations related to products in the cart, you should have the choice to use personalized cross-sell type. It should be easy to change, and, test results with analytics – but our initial suggestion is to use blended recommendations.
User Admin: When a repeat shopper returns to your site, the shopper can logon and update settings in the user admin pages. When a shopper logs on, they should be shown recommendations that they are most likely to buy. It should be based upon their shopping history and/or items in the cart. Here, your vendor should provide blended recommendations of products that the customer is likely to buy based upon the shopper’s purchasing history, and, if there are any, the product(s) in the cart. If there are no products in the cart, personalized recommendations should automatically be used.
Search: For the search results page, your vendor should be able to serve recommendations based upon each search result. Specifically, for each result one or more similar and/or one or more cross-sell items can be shown with the result. This provides a better search experience that is broader than the search text, and converts browsers to buyers.
Email: For shipping, billing, or transactional email after a purchase, the recommended products are derived from one or more items that were just purchased, and should be personalized cross-sell recommendations, thus being related to the purchase and using the customer’s buying habits. For a periodic branding email, the recommended products are optimally personalized for the shopper, and not only related to recent purchases.
Where can I offer recommendations to my customers?
Product recommendations can be successfully integrated into every communication vehicle you use with your customers, including:
- Ecommerce web site
- Mobile commerce
- Social networks
- Email communications
- Online chat
- Phone orders
Additionally, recommendations can be effective at every stage of the customer shopping experience:
- Product detail page
- Shopping cart
- Checkout (billing and shipping pages)
- User admin pages
- Order and shipping confirmation emails
- Branding email communications
How is the recommendation technology integrated with my existing systems?
It is beyond the scope of this document to address specifically the multitude of different Ecommerce platforms. However in general terms, the integration process will include the following steps:
- Data Download – Provide sales data and product catalog feed on a regular basis to the recommendation vendor.
- Data Analysis – The recommendation system will use this data to generate recommendations offline. You may need to fine-tune the system to accommodate your specific needs and business rules.
- Recommendation Request – Add code to web page and email templates that will request product recommendations. This request code will send the relevant product and customer information (i.e., which customer is looking at what item) to the recommendation vendor and return product recommendations in real-time. The request and response should be fast — under 100 milliseconds.
- Recommendation Display – Display the recommendation(s), such as on a web page, mobile content or in an email, as needed. How this step of the process is accomplished will depend on whether you have a custom-built or off-the-shelf ecommerce system. It should accommodate the various display environments and neither require you to learn a complex template language nor interfere with the design of your website. Ultimately, it should enhance the user’s experience of your site. Ideally, the product recommendations should be displayed so that search engines can see the related products or content.
How should I evaluate different recommendation technologies?
Each vendor uses proprietary algorithms to generate product recommendations, but there are a variety of other ways in which you can assess a recommendation system:
- Ease and speed of integration with your ecommerce platform
- Minimal start-up time, meaning that the system does not require a lengthy burn-in/learning period prior to being able to start making recommendations
- Ability to generate recommendations for each customer, known as one-to-one personalization. Some vendors are selling market segmentation solutions as personalization. These segmentation solutions categorize shoppers into only a few segments, and, thus, are not as effective as a true personalized recommendation system.
- Ability to generate effective recommendations even for products with limited sales volume or history
- Recommendation preview capability to ensure that the system is functioning as desired
- Robust analytics for monitoring the effectiveness of the system and for informing future merchandising decisions
- Option to fine-tune the recommendations based on business rules
How many recommendations can I put on a page or email?
Our standard configuration allows up to 20 recommendations at a time. We suggest three to five visible at one time, and then adding a slider bar to show up to 20. An exception might be on a page with a number of products displayed (such as a search results page) where you want to display one or two product recommendations for each primary item.
Is this a hosted solution or does it run locally on my server?
Both. 4-Tell Boost for Web is running now on our redundant cloud servers for anyone needing a hosted solution. If, however, you would rather install the service on your own dedicated web server, we support that option as well. Just request our Boost installer and have your web administrator install it on your server.