Product recommendations rely on algorithms and business rules that are set by you. You must create algorithms that map to the common behaviors of visitors on your website. These algorithms are added to models in your recommendation strategies. You can fine-tune your recommendations by adding business rules to the strategy. In Personalization, you can create models that define the product recommendation strategy.
Create product recommendation models
A product recommendation model is a combination of an algorithm, its parameters, the lookback time frame, and the model inputs. For example,Most popular is an algorithm, whereas Most popular-All-Viewcount-Past 30days is a model.
Specify model inputs
The recommendations model needs the correct information to generate relevant recommendations. You can provide such information through the model inputs.
Note: To specify model inputs, you must have the Personalization library version 2.1.0 configured.
There are several approaches to specify the model input data dynamically. You can specify the inputs based on the context or the visitor behavior on the web page. The following table provides information on how these options work.
Option |
Description |
None |
Select this option if you do not want to use dynamic inputs based on web page or user behavior. When you select Type of Input as None, the Model Input field is not available. In this case, the strategy uses the standard inputs as per the algorithm you had selected. For example:
|
Web page |
Select this option to specify inputs based on the context of the web page on which the model is configured. Next, select any one of the following inputs to the model in the context of the web page:
|
User behavior |
Select this option to specify the inputs based on the visitor behavior on the web page. Next, select any one of the following inputs to the model in the context of the user behavior on the web page:
|
Specify the parameters
The parameters added to the recommendation model define the source of the model inputs. They can be product ID, category ID, product attributes, or user behavior events. Based on the type of model input you select, the required parameters can differ.
The View to buy and Abandon to buy models generate recommendations with multiple product IDs. These product IDs may belong to multiple categories. Thus, you cannot specify category ID and product attributes for these two algorithms.
Choose algorithms
The following algorithms are available for you can choose from in your models. For each of these algorithms, you can specify the number of products to be returned. By default, the algorithm returns ten items. You can also specify the number of products to be displayed on the zone. The channel developer determines how the products returned by the model should render on the zone.
View to view
This algorithm recommends other products viewed by visitors who viewed the current product. It uses the ProductView events to generate product recommendations. For example, visitor1 is viewing modelA running shoes. A list of products viewed by visitors who also viewed the same modelA running shoes is shown to visitor1. The View-to-view algorithm returns the product recommendations based on visitors' views.
Buy to buy
This algorithm recommends other products purchased by visitors who bought the current product. It uses CartPurchase events to generate product recommendations. For example, visitor1 buys the modelA running shoes. A list of products purchased by visitors who also purchased the modelA shoe is recommended. The Buy-to-buy algorithm returns the product recommendations based on CartPurchase data.
View to buy
This algorithm recommends other products bought by visitors who viewed the current product. For example, visitor1 views products A, B, C, and D. Then they purchase the product D.
This behavior of visitor1 is captured by the View to buy algorithm. If another visitor views the products A, B, C, or D, then the product D is shown in the visitor's recommendations.
The algorithm finds the pattern between the ProductView and CartPurchase events. The product recommendations are returned based on the pattern.
Most popular
This algorithm recommends the most popular products based on the parameters you set. You can specify the parameters in the product recommendation strategy. The following table provides information on the parameters that you can set.
Option |
Description |
View count |
Select the most popular products with the most views based on the ProductView events. |
Buy count |
Select the most popular products which were purchased most often. |
Margin |
Select the most popular products with the highest cumulative margin. The margin is based on the difference between the product sale price and the product cost. |
Revenue |
Select the most popular products with the highest cumulative revenue. The revenue is based on the overall sales. |
Note: Make sure you include the revenue or margin data when you create the product catalog. Only if that data is included in Personalization you can use it in the Most popular algorithm.
When you select this algorithm, you must select the measure of popularity parameter. Based on this measurement, recommendations are generated. Recommendations are for all categories in the given time frame of the behavioral data.
Most recent
This algorithm tracks products recently viewed by the visitors on the channel. The tracked products are saved in the local storage of the web browser.
Recommendations are generated based on the latest activity of visitors on the channel. The algorithm does not have a model associated with it, and it is not listed on the Model status page.
A channel has a recently viewed zone configured on the homepage. When a visitor first arrives at the channel, product recommendations will not display.
The visitor clicks on product A and navigates away. When they return to the homepage, then product A will be displayed on the recently viewed zone. The product details, such as the image, price are fetched from the product catalog.
For the Most recent algorithm, you need to specify only the Lookback time frame as the parameter. The algorithm tracks the recently viewed products within the specific Lookback time frame. The algorithm does not consider the user sessions on the channel.
You cannot create a fallback strategy using this algorithm.
Abandon to buy
This algorithm tracks the visitors with the following behavior:
- Visitor abandons (removes) a product from their cart.
- Then purchases a recommended product either in the same session or in the next sessions.
The algorithm uses attributes of the product that was abandoned to recommend items. It matches the purchase patterns of other users who had abandoned similar products. The product recommendations are based on what the other users then purchased.
For example, user A adds a product P to the cart and then removes it later. User A then selects product R and buys it. Now, when user B adds and abandons product P, the algorithm suggests product R as a recommendation.
You can specify the product abandonment event attributes as an input to the model. You can recommend items that are likely to be bought based on users' pattern of abandoning the same item.
You can apply this model to any page, such as product pages, cart pages, or order confirmation.
Model placement
Place the model where product recommendations will provide the best user experience. When deciding the model placement on the channel pages, consider the following criteria:
- Check the technical feasibility of placing the model on certain pages of the website.
- Verify the input data for the model. Depending on the algorithm, the input data may or may not be dependent on the page where the model is placed.
- Consider the significance/effectiveness of placing models on certain pages.
The following table provides a few suggestions for model placement and model inputs.
Note:The model placements shown in the table are for guidance purposes. You may change the placement of the models as per your requirements.
To display… |
On this site page…. |
Select this algorithm |
Type of input |
Model input |
Input data |
Most popular products across all categories in the catalog |
Home page |
Most popular |
None |
NA |
NA |
Most popular products within a specific category |
Product details/ |
Most popular |
Web page |
Category ID |
Category of the page under consideration |
Most popular products with the specified attribute combination |
Product details page |
Most popular |
Web page |
Attributes |
Specified attributes of the product under consideration |
Most popular products within a specific category |
Home page/ |
Most popular |
User behavior |
Most recently/popular viewed/purchased/carted category |
Category of the product based on model Input selected |
Alternate products of the product under consideration |
Product details page |
View to view/ |
None |
NA |
Product ID of the product page under consideration |
Alternate products of the product under consideration |
Product details page |
View to view/ |
Web page |
Product ID |
Product ID of the product page under consideration |
Alternate products with the same category of the product under consideration |
Product details page |
View to view/ |
Web page |
Category ID |
Product ID and category of the product page under consideration |
Alternate products with the same attribute combination of the product under consideration |
Product details page |
View to view/ |
Web page |
Attributes |
Product ID and specified attributes of the product page under consideration |
Alternate products with the same category of the product specified in Model Input |
Home page/ |
View to view/ |
User behavior |
Most recently/popular viewed/purchased/carted category |
Product ID and category of the product specified in model Input |
Alternate products of the product specified in Model Input |
Home page/ |
View to view/ |
User behavior |
Most recently/popular viewed/purchased/carted product or Largest purchase |
Product ID of the product specified in model Input |
Alternate and frequently viewed and then bought products, based on entire current session. |
Cart checkout page |
View to buy |
None |
NA |
Products viewed in the current session |
Alternate and frequently viewed and then bought products, based on entire current session. |
Cart checkout page |
View to buy |
Web page |
Product ID |
Products viewed in the current session |
Alternate and frequently viewed and then bought products, based on entire current session. |
Home page/ |
View to buy |
User behavior |
Most recently/popular viewed/purchased/carted product or Largest purchase |
Product ID of the product as specified in the model Input |
Products recently viewed by the visitor |
Home page/ |
Most recent |
NA |
NA |
All the product IDs viewed by the visitor within the Lookback time frame, across the sessions |
Alternate and frequently abandoned and bought products, based on entire current session. |
Cart checkout page |
Abandon to buy |
None |
NA |
Products abandoned in the current session |
Alternate and frequently abandoned and bought products, based on entire current session. |
Cart checkout page |
Abandon to buy |
Web page |
Product ID |
Products abandoned in the current session |
Alternate and frequently abandoned and bought products, based on entire current session. |
Home page/ |
Abandon to buy |
User behavior |
Most recently/popular viewed/purchased/carted product or Largest purchase |
Product ID of the product as specified in the Model Input |
|
Create business rules
Business rules define conditions for a product to be part of the recommendations. You can use business rules in your product recommendations strategy. Business rules fine-tune your product recommendations to pin, bury, or exclude products. The rules can filter and change the recommendation results presented to the visitors. The business rule name is case-insensitive and must be unique within the strategy.
You can use business rules in a variety of ways, for example:
- Narrow down the recommendations to the most optimal suggestions.
- Highlight newly launched products, or products with the best margin or revenue.
- Prevent recommending out-of-stock items.
- Facilitate up-selling and cross-selling.
The following business rules are currently supported, in the order of their precedence:
Pin
Use the Pin rule to display products that meet the criteria to always appear at the top of the list. The Pin business rule has the highest priority than the other rules.
When you create a business rule to pin products, you can create three types of Pin rules. When you select the action as pin products, choose the type of Pin rule from the drop-down:
- Specific products – You can use this type to pin specific products based on their IDs.
- Products based on catalog attribute value – Use this type to pin products based on their attribute values. You can use any attribute in the catalog to create conditions. You are not limited to use only the ID attribute. You must type in the expected value for the attributes.
- Products listed in the catalog field - Use this type to pin products listed in the catalog field. You can use any of the attributes from the catalog list. For example, say you have a special attribute Stylist suggests added to your catalog. You can use your special attribute Stylist suggests in this type to create conditions. The values you provide for this option must be PIPE separated list of values.
Note: After you create the Pin rule, you must save the business rules and publish the strategy to apply the rules.
The pinned product is always shown as the first recommended product. The algorithmic model processing results do not affect the pinned products.
For example, if you have a new product X launched in brand A, you can pin that product. Then, all visitors browsing brand A will see product X even if there are no other recommendations.
When the Pin rule is applied to several products, they are all listed at the top. You cannot combine Pin action with any other action (that is, you cannot use the AND operator when using a pin).
Bury
The Bury rule helps you to rank the product as the lowest in the recommendation list.
For example, you want products with lower margin to appear at the bottom. You can create a Bury rule that moves that product to the bottom of the product recommendation list.
Exclude
This rule eliminates or removes the products that meet the defined criteria. The excluded products will not appear in recommendations. For example, create the following rule to exclude products with less inventory. "Exclude products where the inventory of the product is less than 10".
If you have a Pin rule along with the Exclude rule, then the Pin rule takes precedence. For example, consider the following business rules:
- Rule A: If the inventory is less than 10, exclude the product.
- Rule B: If the margin is more than 30%, pin the product ADIDAS005.
When the criteria are met, the Pin rule prevails. Here the pinned product ADIDAS005 is shown in the recommendations, even if its inventory is less than 10.
Important: Usually, the Pin rule is the highest priority. But the Exclude rule can supersede the Pin rule in certain scenarios. For example, when the exclude rule is applied to all the recommended products. When the criteria are met, since all the products are excluded, no product will be shown. The pinned products are also excluded and will not be shown.
For example, consider a business rule as follows:
- Rule A: If the brand is not CHNL, exclude all recommended products.
- Rule B: If the inventory is more than 100, pin product PERFUME007.
Here all the products that do not belong to the brand CHNL are excluded. Including the pinned product PERFUME007 is excluded if it does not belong to the brand CHNL.
Filter
The Filter rule helps you to filter products in the recommendation list. Only products that match the attributes of the product that the visitor is viewing are shown. You can use the Filter rule only with the ID attribute.
Note: When the filter rule for the Most popular algorithm is applied on a home page zone, the zone is not be populated.
When Pin and Filter rules are applied together, the pin takes higher precedence. Hence, even if the pinned product is not in the filtered data, the pinned product is displayed. If a product has multiple categories, by default, the filter is applied to the first category.
Boost
The Boost rule helps promote the product to a higher position in the recommendation list. For example, "Boost a particular product with a category ID to be displayed at the top of the page."
Examples
Here are some examples to illustrate the usage of business rules.
Title |
Example usage |
Table 2. Business rule examples |
|
Exclude based on the value of an attribute |
You do not want to recommend products with an inventory of fewer than three units. You do not want to recommend products with a price less than $100. |
Pin based on the value of an attribute |
You want to pin the specific brand (Brand Nike) to appear as the first in the recommendation list. You want to pin the product with the name (X) to appear as the first in the recommendation list. |
Bury based on the value of an attribute |
You want to bury the product with a specific category (Gaming). You want to move that product to the end of the recommendation list. |
Filter based on the attribute |
You want to filter products to show only those that match the product viewed by the visitor. |
Boost based on the value of an attribute |
You want to boost (promote) the product with an availability of more than 100 to show at the top. |
Best practices
Here are some tips and guidelines to help you get the best out of your product recommendations.
Most recent algorithm
When a visitor has not viewed any products, the recently viewed zone can be empty. The Most recent algorithm does not have any data to show product recommendations. To avoid an empty zone, your channel developer can make the following code changes to the channel:
- Hide the Recently Viewed zone until recently viewed products data is received.
- Display an information message, for example: "There are no recently viewed products."
Using business rules
When you specify an attribute as model input, do not use the same attribute to create an Exclude rule. As the Exclude rule will filter out all the primary recommendations.
For example, you select Category ID as model input. Then create a business rule as Exclude-CategoryID-electronics. When the model input Category ID equal to electronics, relevant recommendations are generated. But then all the primary recommendations would be filtered out by the Exclude rule. Hence relevant recommendations may not be displayed on the channel.
Tips to show a more relevant and higher number of results in recommendation zones:
- Apply business rules on a well-defined collection of recommendations. Especially for a primary strategy using the View to view, Buy to buy, and Most popular algorithms. For example, generate recommendations for a specific category or an attribute such as brand. Here's a sample: Model Input = Category ID, Attributes= brand.
Examples of business rule you can create:
- Exclude category which is not the same category of the product under consideration.
- Exclude brand which is not the same as brand of the product under consideration.