How to set up Search Recommendation

When it comes to utilizing Raptor for Search Engines, Raptor tracks and understands the website behavior and integrates to the Search Engine in two ways:

ProductCatalogueRank

Firstly, we can utilize the data from the website on an aggregated level with which we can sort the prioritization of products based on:

Conversion rates
Click popularity
Trends
Number of purchases

We can add additional inputs such as stock level, margins etc. and we can combine the scores of the different inputs to come up with a sorting that meets your business needs (additional types of sorting such as Price high-low is also available).

Typically, this output is made available via a static URL to a Blob Storage from which the output can be fetched in batch. The output can be updated daily or hourly. The output will simply contain the ProductId and a Priority that needs to be fetched and sorted ASCENDING. The output is available in CSV, XML or JSON.

💡 Example:

[{"ProductId":"13243747","Priority":"1"},

{"ProductId":"13251010","Priority":"2"},

{"ProductId":"15791021","Priority":"3"},

{"ProductId":"15799363","Priority":"4"},

{"ProductId":"14711288","Priority":"5"},

{"ProductId":"13243063","Priority":"14998"},

{"ProductId":"15791025","Priority":"14999"},

{"ProductId":"15799194","Priority":"15000"}]

PersonalSearchBoost

Secondly, we can personalize the search engine output on a 1:1 level, which means that the results of the search engine accounts for and is personalized the users’ preferences. Depending on the configuration of the Search Advisor module, the output can be a combination of:

- User’s purchase history

- User’s clickstream history

- Twin Analysis

- Cross & Upselling

- Trends

Again, additional inputs such as stock level, margins, campaigns etc. can be accounted for. Whereas the first sorting is a batch file for all users, this, on the contrary, is a personalized real-time dynamic output that matches the preferences and expectation of the users. For our Search Advisor to work, we typically make the following endpoints available:

Brand Recommendation Endpoint

This module can be used to rerank a search result based on the individual customers preferences and affinity towards specific brands.

API call for Raptor : 
https://api.raptorsmartadvisor.com/v1/{CustomerId}/GetSearchPersonalBrandIds/10/{apikey}?CookieId={CookieId}&json=true

Output   

[{"BrandId":"A","Score":"2.153985"},{"BrandId":"B","Score":"1.441762"},{"BrandId":"C","Score":"0.999654"}]

Product Recommendation Endpoint

This module can be used to rerank a search result based on the individual customers preferences and affinity towards specific products.

API call for Raptor  

https://api.raptorsmartadvisor.com/v1/{CustomerId}/GetSearchPersonalProductIds/10/{apikey}?CookieId={CookieId}&json=true

Output   

[{"ProductId":"A","Score":"2.153985"},{"ProductId":"B","Score":"1.441762"},{"ProductId":"C","Score":"0.999654"}]

Category Recommendation Endpoint

This module can be used to rerank a search result based on the individual customers preferences and affinity towards specific product categories.

💡 Example:

API call for  Raptor 

https://api.raptorsmartadvisor.com/v1/{CustomerId}/GetSearchPersonalCategoryIds/10/{apikey}?CookieId={CookieId}&json=true

Output   

[{"CategoryId":"A","Score":"2.153985"},{"CategoryId":"B","Score":"1.441762"},{"CategoryId":"C","Score":"0.999654"}]  

The exact API calls and the specific documentation for the different endpoints can be found in the Controlpanel under Search Advisor. The output of the different endpoints can be used as inputs in your Search Engine Index. The endpoints can be called with whatever frequency you need.

Since the majority of Product Listing Pages (PLPs) are results of the Search Engine’s Index, implementing Raptor Search Advisor means personalizing the actual search results in the search bar and also the order with which products are presented across the different PLPs.

Understanding the score

Each recommendation returned by the Brand, Product, and Category Recommendation Endpoints includes a score showing how strongly a customer prefers that brand, product, or category, relative to the other results in the same response. A higher score means stronger affinity. The score is not on a fixed scale such as 0 to 10,000: it has no fixed upper bound, but it is always greater than 0.

🔍 Note: Because there's no fixed maximum, don't build logic that assumes scores fall within a specific range. Compare scores relative to each other within the same response instead.

The score is built from two customer behaviours over the last 30 days:

Visits: pages the customer viewed relating to a given brand, product, or category.
Purchases: items relating to a given brand, product, or category that the customer bought.

Both are weighted so recent activity counts more than older activity. Visit activity halves in weight every 14 days: something viewed today counts at full strength, something viewed 14 days ago counts at half strength, and something viewed 28 days ago counts at a quarter strength. Purchase activity halves in weight every 28 days.

💡 Example: A customer viewed pages for the brand "Hummel" 14 days ago and again today. Today's visit counts at full strength, and the visit from 14 days ago counts at half strength. Together they contribute more to the Hummel score than a single visit from today would on its own.

The visit score and purchase score are then combined using the visit weight and purchase weight configured on the module (currently 3 and 5), and the result is multiplied by the module's ScaleFactor. Because ScaleFactor and the weights are configurable per module, scores from two different modules aren't directly comparable, even when built from the same customer behaviour.

⚠️ Warning: Changing the ScaleFactor or the visit/purchase weights on a module changes the resulting score. If you've built business logic or thresholds on top of the score, revisit them if a module's configuration changes.

If you have any further questions, please reach out to our support at support@raptorsmartadvisor.com . 