Recommendation Algorithms
- 15 Nov 2023
- 3 Minutes to read
Recommendation Algorithms
- Updated on 15 Nov 2023
- 3 Minutes to read
Article Summary
Share feedback
Thanks for sharing your feedback!
The Recommendation Algorithms page provides you a direct access to the complete list of recommendation algorithms.
On this page, you can:
- Monitor the real-time health status of each algorithm.
- Customize algorithm parameters, allowing you to tailor recommendations according to your preferences.
To access the Recommendation Algorithms page, navigate to Components > Recommendation Algorithms. 
Use Cases
With the Recommendation Algorithms page, Insider aims to increase the transparency between how your recommendation algorithms are performing and how you customize their parameters to match your business and strategy needs.
First, you need to keep in mind that in order to ensure continuous recommendation accuracy, you must carefully select the two calculation parameters: user event data sources for your algorithms and an appropriate look-back period. Customization in these aspects is vital for optimizing recommendation accuracy.
Your title goes here
Each algorithm's parameters are selected centrally. This means that an algorithm's parameters, such as data sources and look-back periods, cannot vary between campaigns. For instance, an algorithm cannot utilize offline-only data on one campaign and app-only data on another, nor can it have different look-back periods for different campaigns.
Below are some use cases of how you can benefit from customizing recommendation algorithms:
"Time-bounded" Recommendation Strategies
Flash sale deals (daily)
If you have discount strategies that are renewed every day,
- Select the "Highest Discounted Products" algorithm.
- Set the look-back period as 1 day.
Algorithm will calculate your discounted products on daily basis, showing products that have discounted price yesterday.
New products of the week
Adjust the look-back period of your "New Arrivals" algorithm based on your line or collection release frequency. This ensures the display of products added during that specific period while preventing 1-month-old products from appearing.
- Select "New Arrivals" algorithm
- Set the look-back period as 7 days.
Trending products of the season/collection
In each new line of products added to the catalog, user behavior undergoes changes as they interact with these additions. To spotlight the top picks among trending products and identify success within a 2-week timeframe, you can:
- Select "Trending Products" algorithm
- Set the look-back period as 14 days
Hottest products in your area
If you want to launch a "top sellers of the store" campaign specifically targeting users in the city where your largest store is located, you can:
- Select the "Location Based Top Sellers" algorithm.
- Select the data source as "offline" only.
Platforms you engage with your end users (web page, app, offline stores)
When selecting user event sources for algorithms, consider the following:
- Identify the platforms you operate on, such as websites, apps, or offline stores.
- Ensure you have the necessary integrations to utilize the chosen event source (Mobile App and Upsert API).
- Evaluate if the user event data collected from that source is relevant to include in recommendations
Platforms Operation
On three main platforms, you can engage with your end users:
- Web site
- Application
- Offline stores
Regarding your Insider usage on these platforms, your recommendation algorithms can ingest user events collected from all these platforms, and create multi-source aware recommendations.
Integration Capability
Integration with Insider is a prerequisite for gathering user events from those particular sources.
- Insider can ingest user events from app user interactions only if there is mobile app integration.
- Insider can ingest user events from offline store interactions only if there is Upsert API integration for offline stores.
Relevance of User Behaviors
In general, a greater diversity in user event sources tends to enhance the accuracy of recommendations. However, if you think you have differences between offline and website user behaviors and believe that offline data may not align well with online interactions, it's valid to not include the offline data on your recommendation algorithms.
Your title goes here
We strongly recommend enabling all user event sources for all algorithms whenever possible. This boosts data quantity and variety, leading to more accurate recommendations that consider user behaviors across all platforms.
Velocity and volume of end-user traffic
More user events generally lead to more accurate recommendations. If you are experiencing low Monthly Active Users (MAU), declining traffic, or concerns about recommendation accuracy, consider adjusting the look-back period and enabling additional user event sources.
- Increasing the look-back period captures more data, particularly during low-traffic periods.
- Enabling more user event sources allows for ingesting a greater volume of user events, enhancing recommendation accuracy.
For instance, in a real-life scenario, if you're an outdoor sports retailer with decreasing traffic in November, extending the look-back period of your user-based algorithm to 60 days can improve recommendation quality by considering a larger set of user events.
Reminders
- If you don't utilize Insider's Mobile App product or have Upsert API integration, you are unable to incorporate user events from your app or offline stores.
- Selecting the data source and setting the look-back period for an algorithm cannot be tailored to individual campaigns. Instead, these configurations apply to all campaigns.
- Each algorithm's data source and lookback period can be changed at any moment.
- Data source selection of an algorithm is not a frequent decision, a change in data source may occur due to seasonal cross-channel campaigns, or a general strategy change.
Was this article helpful?