Skip to main content
A few practical answers to the most common questions about how recommendation algorithms behave in real campaigns — what to show inactive customers, how to keep products from repeating, and how to make sure an algorithm doesn’t quietly stop refreshing.

​
Which recommendations to send to inactive customers

Maestra Platform only calculates recommendations for customers who have been active in the last 90 days. For customers who fall outside that window, the behavior depends on the algorithm: Algorithms that fall back to popular products These algorithms keep working for inactive customers by surfacing generally popular items:
  • Personal recommendations
  • Popular products
  • Event-based personal recommendations
Algorithms that return previously generated or empty results These algorithms rely on recent customer behavior, so for inactive customers they either return the last result they generated or nothing at all:
  • Popular products in viewed categories from the last session
  • Related products to the last order
  • Similar products to those viewed in the last session
  • Manual category mapping
  • Recently viewed products
For win-back and reactivation campaigns, use personal recommendations or event-based personal recommendations. They guarantee that the email never lands in an inactive customer’s inbox with an empty or stale recommendation block.

​
How to avoid duplicate products

​
Within a single recommendation widget

If you use Maestra’s embedded recommendation widget, deduplication happens automatically — the widget filters duplicates for you. If you build the widget yourself through the API, you need to handle deduplication on the front end.

​
Across multiple widgets on the same page

Turn on Filter by other recommendations on the page in the widget’s main settings. Maestra Platform will make sure no product appears in more than one widget on that page.

​
Within a single campaign

Maestra Platform can’t automatically deduplicate across recommendation blocks inside the same email. To prevent overlap, configure the algorithms to draw from non-overlapping product segments — for example, one block from “New arrivals” and another from “Discounted items.”

​
An algorithm stopped after 20 days of inactivity

If an algorithm isn’t called for 20 days in a row, Maestra Platform stops recalculating it. The recommendations it returns become stale, and eventually the algorithm goes idle. To keep an algorithm warm even when you’re not actively using it:
1

Create a test flow

Build a flow that sends a test email to an internal service address.
2

Call the algorithm on a schedule

Use an operation inside that flow to call the algorithm periodically. As long as it’s called at least once every 20 days, Maestra Platform keeps it on the recalculation schedule.
You only need to make sure the algorithm is requested — the test email doesn’t have to be sent to real customers. A daily or weekly test send to your own team address is enough.