Loomi recommendation templates

Loomi templates use machine learning models to generate recommendations. They require training time—algorithms are usually run overnight so the model is ready the next day.

Choose a Loomi template for recommendations based on similarity suggestions, co-occurrence patterns, and more—and when your catalog has enough historical data to train the model. If your catalog is new or sparse, start with a ready to use template instead.

📘
Personalization
Most Loomi templates offer personalization based on selected customer events. The following templates are not personalized:

Customers who bought this item also bought

Similar descriptions

Loomi templates quick overview

The following templates are supported by Loomi:

Template	Use it for
Customers who bought this item also bought	"Frequently bought together" blocks used for cross-sell on product detail and cart pages.
Customers who viewed this item also viewed	"You might also like" blocks on product detail pages, based on co-viewed patterns.
Personalized for you (Matrix-factorization model) (previously Personalized recommendations for you)	Personalized recommendations blocks across any placement—homepage, email, cart, or post-checkout.
Personalized for you (Two-tower neural network)	Personalized recommendations blocks that also use the sequence of a customer’s interactions to generate recommendations. A neural-network variant of the standard Personalized for you template available via the Recommendations+ add-on.
Similar descriptions (previously Textual similarity)	Recommendations blocks for content-heavy catalogs—media, books, apparel—where product descriptions, titles, and tags are the strongest signal for what's similar.
Personalized homepage (previously Homepage)	Personalized homepage block optimized for users arriving without a browsing history in the current session, for example, returning customers.
Personalized product page (previously Product detail)	Personalized "recommended for you" block next to the currently viewed product generated by blending item context with the customer's personal history for more relevant suggestions.

All Loomi templates support these optional merchandising operations:

Catalog filters (support varies per template): narrow the item set to products matching attribute conditions, such as a specific category or price range.
Block list: exclude items by event or item ID
Pin items: reserve fixed positions for specific items or attribute matches
Dynamic filters (support varies per template): narrow the item set based on the customer or the reference item
Customer preferences: sorts results by per-customer attribute preferences

📘
Note
Each template uses a single engine. To combine multiple engines to cover your business needs, use the Combined ready-to-use template.

Customers who bought this item also bought

Customers who bought this item also bought is a non-personalized template used for up-sell and cross-sell on the product detail page or cart page, and in reactivation emails. It shows items frequently bought together with the item the customer is currently viewing.

Personalized	Item-based
Recommended placement	Product detail page, cart, checkout, customer re-engagement email

Configuration

Required configuration	Catalog, catalog filters, learning window
Required events	Purchase item event
Dynamic filters (optional configuration)	Only item context dynamic filter supported
Request parameters	`items` (recommended); `last added item` (for cart use cases)
Data prerequisites	Recommended at least 6 months of `purchase_item` data
Fallback logic	Reference items have low purchase volume or no reference item → falls back to Popular right now. No results → falls back to random.

📘
Note
For shopping cart use cases, include the ID of the item added to the cart so the engine renders recommendations based on the customer's last actions.

Example use cases

Product detail page: show "frequently bought together" pairings on a coffee machine page—surface filters, descaler, and spare parts that other buyers added to their order
Cart: suggest complementary items in the cart drawer when a shopper adds a DSLR camera—recommend a memory card, camera bag, or lens
Checkout: prompt a last-minute add-on at checkout when a customer is buying running shoes—recommend socks, insoles, or a shoe bag
Reactivation email: remind a lapsed customer what others bought alongside a product they purchased months ago to prompt a repeat visit

Customers who viewed this item also viewed

Customers who viewed this item also viewed is a personalized template used for alternative product placements like "You might also like these" on product detail pages. It shows co-viewing patterns of the item currently viewed.

Personalized	Yes, based on items with customer history fallback
Recommended placement	Product detail page (“You might also like”)

Configuration

Required configuration	Catalog, catalog filters, learning window
Required events	Detail view event
Dynamic filters (optional configuration)	Supported
Request parameters	`items` (optional, when none provided, last 3 viewed are used)
Data prerequisites	Recommended at least 2 months of `view_item` data

Fallback logic

Reference item is included, but history is insufficient → falls back to most viewed products.
No reference item is included → uses the customer's last 3 viewed items as the basis.
No reference item and no view history is provided → falls back to most viewed products, then random.

Example use cases

Product detail page: surface alternative jackets other shoppers looked at after viewing the same parka a customer is currently browsing
Post-purchase email: show what other customers explored after buying the same item—useful for accessories or complementary categories the customer hasn't discovered yet

Personalized for you (Matrix-factorization model)

Personalized for you is a personalized template used for cross-sell, "we think you would like these", "based on your recently viewed products", and homepage placements. It shows items that match customer preferences based on purchasing and browsing history, including real-time events in the current session.

📘
Note
Switch to the neural variant when working with catalogs with complex, sequential purchase patterns such as fashion or electronics.

Personalized	Yes, based on customer history
Recommended placement	Any

Configuration

Required configuration	Catalog (without catalog filters), learning window, model type, required events.
Required events	Required events: Detail view event (item detail page visits), Add to cart event (cart additions), and Purchase event (completed purchases)
Optional events	Optional events: additional events the engine can incorporate for more signals, are also supported
Dynamic filters (optional configuration)	Supported
Request parameters	`customer_id` (mandatory), `items` (optional)
Data prerequisites	Recommended at least 2 months of `purchase_item` and `view_item` data

Fallback logic

Purchase history is thin → falls back to the customer's last 3 viewed items or Popular right now.
No purchase history → falls back to Popular right now.
The model produces no results → falls back to top interacted items, then random.

Example use cases

Homepage: show a "picked for you" carousel to a returning visitor based on their full browsing and purchase history
Email: populate a fully personalized product block in a weekly newsletter using each recipient's individual interaction history
Cart: add a "you might also like" row in the cart based on the customer's broader purchase and view history, not just the current session
Post-checkout: surface a personalized "what to explore next" block on the order confirmation page to keep the customer engaged

Personalized for you (Two-tower neural network)

Personalized for you (neural) is a variant of Personalized for you, designed for the same use case but using a different underlying model. This template is available via the Recommendations+ add-on.

This variant uses a Two-Tower Neural Network model instead of the standard ALS-based model. It predicts the next best product by considering the order of user interactions rather than treating them as unordered data. This makes it a better fit for datasets where interaction sequences and deep learning representations outperform matrix factorization.

Personalized	Yes, based on customer history (sensitive to the sequence of interactions)
Recommended placement	Any

Configuration

Required configuration	Catalog (without catalog filters), model type
Required events	N/A
Dynamic filters (optional configuration)	Supported
Request parameters	`customer_id` (mandatory), `items` (optional)
Data prerequisites	Recommended at least 2 months of `purchase_item` and `view_item` data

Fallback logic

Purchase history is thin → falls back to the customer's last 3 viewed items or Popular right now.
No purchase history → falls back to Popular right now.
The model produces no results → falls back to top interacted items, then random.

Example use cases

High-traffic storefronts: improve recommendation accuracy for customers with long, sequential browsing sessions—for example, a shopper who viewed 15 products across 3 categories in one visit

Similar descriptions

Similar descriptions is a non-personalized template used to show items on the product detail page that are similar to the item currently viewed, based on shared textual content. You can also choose attributes that must be shared between the recommended item and the reference item.

Personalized	Item-based
Recommended placement	Product detail pages with content-heavy catalogs (books, media)

Configuration

🚧
Catalog requirements
Catalog must contain textual fields. A block of text works better than a list of properties and values.

Required configuration	Catalog, catalog filters, exact match attributes, optional attributes, textual attributes
Required events	N/A
Dynamic filters (optional configuration)	Only item context dynamic filter supported
Request parameters	`items` (mandatory)
Data prerequisites	Catalog with textual attributes, blocks of text are recommended over lists of properties and values
Fallback logic	Textual attributes are too short → falls back to random.

Example use cases

Books product detail page: recommend novels with similar themes and writing style to the one a customer is viewing, based on description text rather than purchase patterns
Apparel product detail page: recommend similar dresses based on description text when a product is too new or niche to have built up co-view patterns yet

Personalized homepage

Personalized homepage is a personalized template used to provide personalized recommendations on the homepage based on the customer's historical activity. It reuses the logic of Personalized for you (Matrix-factorization model), optimized for the homepage with an automatically selected learning window and real-time support for recently tracked events.

Personalized	Yes, based on customer history
Recommended placement	Homepage

Configuration

🚧
Important
Data mapping is mandatory to ensure proper event tracking.

Required configuration	Catalog, catalog filters
Required events	Target event
Dynamic filters (optional configuration)	Supported
Request parameters	`customer_id` (mandatory); no `items`
Data prerequisites	Recommended at least 2 months of `purchase_item`, `view_item`, and `cart_update` data

Fallback logic

Purchase history is thin → falls back to the customer's last 3 viewed items or Popular right now.
No purchase history → falls back to Popular right now.
The model produces no results → falls back to top interacted items, then random.

Example use cases

Homepage carousel: show a returning visitor a personalized "for you" strip optimized for purchases—sorting items the model predicts they are most likely to buy
Homepage re-engagement: surface a tailored selection for a visitor who hasn't converted in 30 days, using their historical interactions to rebuild relevance

Personalized product page

Personalized product page is a personalized template used to show alternative products on the product detail page that were visited together by all customers. It reuses the logic of Personalized for you (Matrix-factorization model), optimized for product detail pages with an automatically selected learning window and real-time support for recently tracked events.

Personalized	Yes, based on customer history and the reference item
Recommended placement	PDP alternative products

Configuration

🚧
Important
Data mapping is mandatory to ensure proper event tracking.

Required configuration	Catalog, catalog filters
Required events	Target event and view/purchase history
Dynamic filters (optional configuration)	Supported
Request parameters	`customer_id` (mandatory), `items` (mandatory)
Data prerequisites	Recommended at least 2 months of `purchase_item` and `view_item` data

Fallback logic

No reference item is provided → falls back to customer's interaction history only, without item-similarity signals.
Purchase history is thin → falls back to the customer's last 3 viewed items or Popular right now.
No purchase history → falls back to Popular right now.
The model produces no results → falls back to top interacted items, then random.

Example use cases

Product page detail alternatives: blend item-similarity and customer history to show a shopper alternatives to a sold-out product that match both the item's attributes and their personal preferences
Product page detail cross-sell: sorts complementary products on a laptop page using session-aware signals—prioritizing accessories the customer has already shown interest in during the current visit