Presets (default algorithms)

What you will find here ?

How dynamic contextual rules and algorithms work within AB Tasty’s Recommendations & Merchandising platform. It covers the six main “starting points” available when building a strategy, along with a full reference of available algorithms.

Dynamic contextual rules use real-time contextual variables (page type, category, viewed/bought items, recency, frequency, etc.) to return relevant, up-to-date product sets for each user or page.

Items sorted by

Definition Returns products based on a chosen property or performance metric (e.g., sales, recency, margin, page views, purchased quantities). These are typically “Top X Products” strategies ordered by KPI and timeframe.

Logic The rule aggregates catalog and event data (page views, purchases, revenue, quantities) over a defined period, ranks the products, and returns the highest-scoring ones.

Parameter	Description	Example
Input variable	Global or category scope (categoryId, page_type)	categoryId = “Shoes”
Algorithm	Aggregates metrics over X days	pageviews_last_7_days, revenues_last_30_days, trend, etc.
Output	Ranked list of best-performing products	Top 50 bestsellers last 30 days
Dynamic	Auto-updates as data evolves (daily/hourly refresh)	—

Use cases

• Top sellers of the last 30 days
• Newest arrivals (sorted by creation date)
• Trending products (based on sales velocity ratio)
• Most relevant products (weighted by conversion × recency × stock)

---

Items associated to

Definition Returns products most often purchased or viewed together with the current product (“complementary logic”). Built on co-occurrence and TF-IDF weighting to detect statistically meaningful associations.

Logic

Parameter	Description	Example
Input variable	Current product ID (productId) or basket IDs (cart_items)	"SKU12345"
Algorithm	TF-IDF based co-occurrence analysis	“Bought together”, “Viewed together”
Output	Complementary product set	Printer → Ink cartridges
Dynamic	Real-time refresh with new behavioral data	—

Use cases

• Frequently bought together
• Cross-sell bundles (e.g., accessories, consumables)
• Cart-based add-ons (threshold fillers or impulse buys)

---

Items similar to

Definition Returns products that share strong affinities with the current one — acting as alternatives within the same universe. Built using co-occurrence + TF-IDF weighting, attribute-based matching, or embedding similarity (semantic / visual).

Logic

Parameter	Description	Example
Input variable	Current product attributes (productId, categoryId)	"SKU7890"
Algorithm	Similarity scoring model	“Similar products”, “Semantic”, “Visual”
Output	List of substitute or style-related products	“Another Oxford shirt in blue”
Dynamic	Updates as catalog and user signals change	—

Use cases

• Alternative products on PDP (prevent exit)
• Style discovery (same color / material / collection)
• Niche promotion (long-tail visibility for new SKUs)

---

Items from recommendation

Definition Reuses the output of an existing recommendation rule. Used to compose multi-layered or conditional strategies.

Logic

Parameter	Description	Example
Input variable	Output of a previous strategy	“Top sellers in category”
Algorithm	Sequential chaining / composition	Apply filter stock > 0 on previous output
Output	Filtered or refined set of recommended products	Top sellers with stock > 0 & margin > 20%

Use cases

• Composed strategies: combine multiple rules in sequence
• Fallbacks: use another recommendation if the first returns too few products

---

Manual selection

Definition A fixed curated list of products chosen by a merchandiser. Ideal for campaign or editorial content that should not change dynamically.

Parameter	Description	Example
Input variable	Product IDs manually selected	[“SKU111”, “SKU222”, “SKU333”]
Algorithm	None (static list)	—
Output	Always the same set until edited	“Christmas collection”
Dynamic	Manual only	—

Use cases

• Seasonal campaigns (e.g., Christmas, Black Friday)
• Editorial picks (lookbooks / content curation)
• Hero products for brand priorities or high margin

---

Products from variable

Definition Returns products based on a dynamic contextual variable from the page or user session. Typical examples: current category, basket content, recently viewed items, or user profile segments.

Parameter	Description	Example
Input variable	Dynamic contextual variable	categoryId, user_recent_basket, last_viewed_products
Algorithm	Variable-driven lookup rule	“Best sellers in current category”
Output	Contextual set of products linked to that variable	Personalized continuity
Dynamic	Automatically adjusts to user context	—

Use cases

• Category best sellers (using categoryId)
• Recently viewed products (using last_viewed_products)
• Basket-based rules (using user_recent_basket)
• “Because you viewed…” personalization

---

Let’s deep dive into the three core algorithms powering AB Tasty Recommendations & Merchandising

Each of them plays a unique role in how products are ranked and displayed - from behavioral patterns to semantic understanding.

Algorithm	Signal Type	Data Source	Main Objective	Typical Use Cases	Link to specific documentation
Co-occurrence	Behavioral	Orders, sessions, views	Identify products often bought/viewed together	“Frequently bought together”	Co-occurence documentation link
TF-IDF	Statistical weighting	Co-occurrence + global frequency	Reduce bias from generic products	“Next book in a series” instead of “Gift card”	TF-IFD documentation link
Semantic	Contextual (NLP)	Product text, attributes, metadata	Find products similar in meaning or style	“Other running shoes”	Semantic Cosine Distance documentation link

How dynamic rules and algorithms work within AB Tasty’s Recommendations & Merchandising platform?

It covers the six main “starting points” available when building a strategy, along with a full reference of available algorithms.

Dynamic contextual rules use real-time contextual variables (page type, category, viewed/bought items, recency, frequency, etc.) to return relevant, up-to-date product sets for each user or page.

---

🧩 Appendix — All Available Algorithms

Algorithm Name	Translation	Definition
pageviews_last_7_days	7-day page views	Products with the most views over the last 7 days.
pageviews_last_14_days	14-day page views	Products with the most views over the last 14 days.
pageviews_last_30_days	30-day page views	Products with the most views over the last 30 days.
trend	Trend ratio	Ratio of page views between the last and previous period (velocity indicator).
revenues_last_30_days	30-day revenue	Products generating the highest revenue over the last 30 days.
revenues_last_14_days	14-day revenue	Products generating the highest revenue over the last 14 days.
revenues_last_7_days	7-day revenue	Products generating the highest revenue over the last 7 days.
quantities_purchased_last_7_days	7-day purchased quantities	Products with the highest number of units sold in the last 7 days.
quantities_purchased_last_14_days	14-day purchased quantities	Products with the highest number of units sold in the last 14 days.
quantities_purchased_last_30_days	30-day purchased quantities	Products with the highest number of units sold in the last 30 days.

(If your CSV contains additional specialized variants, e.g., “margin_last_30_days”, they can be added under the same pattern.)

---

🔄 Data Flow & Dynamic Variable Usage

Dynamic contextual variables (see dedicated glossary) drive these recommendation rules:

• page_type determines which rule is triggered (homepage vs PDP vs cart).
• viewed_items, bought_items, cart_items feed “associated” and “similar” rules.
• recency, frequency, and monetary score freshness and priority.
• categoryId, productId, and user_recent_basket define the contextual scope for “products from variable”.

Flow overview: User Event (Tag) → Contextual Variables updated → Rule Evaluation (Algo) → Filtered Product List → Display on Site / Feed Analytics.