Skip to content
Go to raptorservices.com
  • There are no suggestions because the search field is empty.

Personal Shopping Assistant

How to set up GetUserItemRecommendationsWeb

The Personal Shopping Assistant (technical name: GetUserItemRecommendationsWeb) recommends items to every step of the customer journey with a short-term focus, providing a personal shopping experience by giving the user highly relevant recommendations.

The module calculates your customer‘s next step from their previous and current behavior, and you can adjust it to both chase the conversion and/or the cross–sales when the first purchase is completed.

We recommend you place the algorithm as a sidebar on your website, to constantly calculate and show recommendations.

Before you start configuring the module, make sure the tracking is set up correctly and that you have set up a product catalog.
 

Module configuration

When setting the weights for parameters, remember that:

- negative values result in given parameter having less impact (or disappearing) from the recommendation results, with the impact lessening as the value gets lower

- zero disables the given parameter

- positive values result in given parameter having more impact, with the impact increasing as the value gets higher

The final impact of a given parameter is calculated based on other parameters' values, as they are all evaluated together and are relative to each other.

🔍 Note:  For every parameter of the form, hover over the information icon to learn more about it and see the recommended values and strategies you can choose.

Step 1 - User identification and interaction

Map your live tracking stream data to the parameters to provide recommendations:

  • Use CookieId or UserId to pass information about the user visiting the website.

  • Use "Search phrase" to pass the information about what the user is searching for.

  • Use "ProductId" to pass information about the currently visited product.

Step 2 - Candidate set strategy: adapts to the customer journey

Select how many of the user's most recent interactions are taken into account when providing recommendations. Then, select the behavior of the system depending on the provided events.


There are three recommended approaches you can consider:

Use case 1: Automatically adapt to the customer journey (default)

User profile for look alike: visit
User profile for cross sell basket, buy
The purpose is to recommend items corresponding to the user's actions and needs.
 
 

👀 Example; if the user clicks on a chainsaw, the look-alike profile will provide an output with similar chainsaws for the user to purchase. At this stage, the recommendations are provided to increase conversion, have the user make a purchase.

When the user adds the chainsaw to the basket, the user has shown an intent to purchase, and the model will start recommending complimentary products like motor oil, chains etc. The recommendations are provided to increase user's basket size.

 
The web version favors recent user activity. The algorithm adapts to the short-term goals of the user and changes output rapidly according to the users' interactions.

 

💡The mail version takes both the recent user activity and number of interactions. The algorithm adapts to the long-term goals of the user, providing recommendations that are most suitable for the user in the long run. For example, if a user has visited the same product several times, that product will have more influence on the recommendations than the items visited just one time.
 
Use case 2: Focus on the conversion (GetUserLookAlikeItemsWeb)
User profile for look alike visit, basket, buy
User profile for cross sell (empty)
This combination results in the model always recommending similar products to the ones the customer interacted with, resulting in a strategy like "Try this instead".
Note, the default for the mail version is just 'visit' since typically user segmentation has already been performed and it's only relevant for customers who have made a 'visit' event.

 

Use case 3: Focus on the additional sales (GetUserCrossSellingItemsWeb)
User profile for look alike (empty)
User profile for cross sell visit, basket, buy

 

This combination results in the model always recommending visitors to purchase products in addition to the original items they have interacted with by providing cross selling (complementary) recommendations in every step of the customer journeys, leading to an increase in the value of the sale.
 

💡Note: The default for the mail version are the 'basket' and 'buy' events since typically user segmentation has already been performed and it's only relevant for customers who have made a 'basket' or ´'buy' event.

 

Step 3 - Backfilling 1: top products in user's favorite brand and category

If the primary recommendation rules don't return enough recommendations, user-based backfilling is activated to supplement personal recommendations to the output.

 

You can choose how much influence user's favorite categories and brands have for this recommendation strategy.

 

Step 4 - Backfilling 2: overall popular or trending items

The secondary backfilling strategy is used when the model didn't provide enough recommendations by using the primary rules and the first backfilling strategy.

 

You can use it to include additional items in model's output using one of two strategies:
- popular items
- trending items

 

Customize the settings to specify the period in which the items are calculated and the event types.

 

Step 5 - Output enriching, filtering and selection

In this section, you can filter out unwanted products based on various factors. This is also where you can enrich your data output with information from your product catalog.

 

Step 6 - Split testing

After you enable split testing, you can set up two configurations (A and B) at the same time. To do this, build a list of parameters, together with their values, separated by the '&' symbol, for example:

- Split test setup A: `VisitHistoryWeight=10&BuyHistoryWeight=5`
- Split test setup B: `VisitHistoryWeight=11&BuyHistoryWeight=6`

 

🔍Note: To find the name of each parameter, hover over the information icon next to each one.

👀 Use case: If you would like to read more about how to predict your customer's next step with our most advanced module yet, please read our blog.