> ## Documentation Index
> Fetch the complete documentation index at: https://help.maestra.io/llms.txt
> Use this file to discover all available pages before exploring further.
# Recommendation algorithm types
Before you build an algorithm, take a look at the visual map — it'll help you pick the right recommendation type for the channel and customer segment you're working with.
### Site recommendations
### Checkout (POS) recommendations
### Call center recommendations
### Campaign recommendations
To create an algorithm, head to **Content → Product recommendations**.
A quick note on timing: real-time algorithms recalculate on top of the model trained during the algorithm's last full refresh. That model is trained on whatever products were available when the refresh kicked off.
Here's a closer look at each algorithm and what makes it tick.
## Bestsellers
Ranks top-performing products over a time window you choose. Popularity is scored first by order volume per product within a category, then by view count as a tiebreaker.
**Good to know:**
* Best for: homepage, "Bestsellers" blocks, cold-start visitors with no personal history.
* Signals used: order count, view count.
* Refresh: once a day.
* Limit: up to 40 instances per project.
* Segment: can be limited to a segment.
* Auto-filters: checks the product's zone and brand (multi-brand projects); excludes items the customer has already bought.
## Category bestsellers
The Bestsellers algorithm, scoped to a single category.
**Good to know:**
* Best for: category pages.
* Limit: up to 20 instances per project.
## Bestsellers from previously viewed categories
This algorithm only generates recommendations for customers with activity in the last 90 days. Everyone else will see recommendations calculated earlier (if the algorithm is more than 90 days old) — or no recommendations at all.
The Bestsellers algorithm applied across every category a customer viewed during their session, over a time window you choose.
**Good to know:**
* Best for: "Category browse abandonment" flows.
* Refresh: real time.
* Limit: up to 1 instance per project.
## Similar products
Surfaces similar products for a specific item. Higher-priced items get more recommendations; lower-priced items get fewer. You can prioritize product attributes (like price or manufacturer) and even require exact matches on specific attributes — for example, size.
**Good to know:**
* Best for: product detail pages.
* Refresh: once a day.
* Limit: up to 5 instances per project.
* Segment: can be limited to a segment.
* Auto-filters: checks the product's zone and brand (multi-brand projects); excludes items the customer has already bought.
## Products similar to recently browsed products
This algorithm only generates recommendations for customers with activity in the last 90 days. Everyone else will see recommendations calculated earlier (if the algorithm is more than 90 days old) — or no recommendations at all.
The Similar products algorithm applied to every item the customer browsed on-site during their last session.
**Good to know:**
* Best for: "Product browse abandonment" flows.
* Refresh: real time.
* Limit: up to 2 instances per project.
## Similar products to product list recommendations
This algorithm only generates recommendations for customers with activity in the last 90 days. Everyone else will see recommendations calculated earlier (if the algorithm is more than 90 days old) — or no recommendations at all.
The Similar products algorithm applied to each item on a customer's product list.
**Good to know:**
* Best for: "Back in stock" and Favorites flows.
* Refresh: real time.
* Limit: up to 3 instances per project.
## Related products
Surfaces complementary products and builds recommendations for a specific item. Higher-priced items get more recommendations; lower-priced items get fewer.
**Good to know:**
* Best for: "Frequently bought together" and "Customers also bought" modules.
* Signals used: same-order co-occurrence, plus category and attribute co-purchase patterns. Works even for products with no order history yet.
* Refresh: once a day.
* Limit: up to 5 instances per project.
* Segment: can be limited to a segment.
* Auto-filters: checks the product's zone and brand (multi-brand projects); excludes items the customer has already bought. Product external systems match by default.
## Post-purchase cross-sell
This algorithm only generates recommendations for customers with activity in the last 90 days. Everyone else will see recommendations calculated earlier (if the algorithm is more than 90 days old) — or no recommendations at all.
The Related products algorithm applied to each item in the customer's most recently updated order.
**Good to know:**
* Best for: "Thanks for your order" and "Next order suggestion" flows.
* Refresh: real time.
* Limit: up to 1 instance per project.
## Often Purchased With Items in Product List
This algorithm only generates recommendations for customers with activity in the last 90 days. Everyone else will see recommendations calculated earlier (if the algorithm is more than 90 days old) — or no recommendations at all.
The Related products algorithm applied to each item on a selected customer product list.
**Good to know:**
* Best for: abandoned Cart, Favorites recommendations, on-site Cart modules.
* Refresh: real time.
* Limit: up to 3 instances per project.
## Custom recommendations
Lets you set complementary products based on category relationships, for a specific item. Once you specify that Category A should recommend items from Category B, each product in Category A gets paired with the item from Category B most often purchased alongside it. You can also layer on similarity parameters — price, manufacturer, or any other product attribute.
Higher-priced items get more recommendations; lower-priced items get fewer.
**Good to know:**
* Best for: product detail pages.
* Refresh: once a day.
* Limit: up to 5 instances per project.
* Segment: can be limited to a segment.
* Auto-filters: checks the product's zone and brand (multi-brand projects); excludes items the customer has already bought. Product external systems match by default.
## Custom post-purchase recommendations
This algorithm only generates recommendations for customers with activity in the last 90 days. Everyone else will see recommendations calculated earlier (if the algorithm is more than 90 days old) — or no recommendations at all.
The Custom recommendations algorithm applied to each product in the customer's most recently updated order, with recommendation volume proportional to price. Once you specify that Category A should recommend items from Category B, each product in Category A gets paired with the item from Category B most often purchased alongside it. You can also layer on similarity parameters — price, manufacturer, or any other product attribute.
**Good to know:**
* Best for: "Thanks for your order" and "Next purchase suggestion" flows.
* Refresh: real time, based on the customer's orders.
* Limit: up to 1 instance per project.
## Personal Recommendations
This algorithm only generates recommendations for customers with activity in the last 90 days. Everyone else will see either recommendations calculated earlier (if the algorithm is more than 90 days old) or a fallback to Bestsellers.
Predicts which product a customer is most likely to buy next.
**Good to know:**
* Best for: "Next order suggestion" flows and win-back mechanics.
* Signals used: the customer's views, orders, and list activity, plus behavior from similar customers.
* Refresh: once a day.
* Limit: up to 5 instances per project.
* Segment: can be limited to a segment.
* Auto-filters: checks the product's zone and brand (multi-brand projects); excludes items the customer has already bought.
## Event-based personal recommendations
This algorithm only generates recommendations for customers with activity in the last 90 days. Everyone else will see either recommendations calculated earlier (if the algorithm is more than 90 days old) or a fallback to Bestsellers.
Works just like standard Personal Recommendations, but calculates in real time and also supports anonymous customers.
**Good to know:**
* Best for: homepage, account page, 404 page, search results.
* Refresh: real time.
* Limit: up to 1 instance per project.
## Recently browsed products
This algorithm only generates recommendations for customers with activity in the last 90 days. Everyone else will see recommendations calculated earlier (if the algorithm is more than 90 days old) — or no recommendations at all.
Recommends products the customer viewed in-session over the last 14 days. Max 1 instance per project.
For setup details on any of these, check the algorithm configuration screen in the platform.