Upload
dynamic-yield
View
99
Download
2
Embed Size (px)
Citation preview
UNDERSTANDING RECOMMENDATIONS: VALUE, FUNCTIONALITY & BEST PRACTICES
THE VALUE OF RECOMMENDATIONS
AVERAGE ORDER VALUE
Encourage the user to explore products“I’m just browsing around”
CLICK THROUGH RATE
Optimize conversions by selecting and presenting the most relevant products based on user’s mindset and stage of purchase funnel.
BROWSING SITE
PRODUCT VIEW
ADD TO CART
PURCHASE
REPURCHASE
THE VALUE OF RECOMMENDATIONS
Surface related products & drive the user to convert
Do not distract yet increase cart value
CONVERSION RATE“I’m looking for something specific”
“I want to buy now”
RECOMMENDING THE RIGHT PRODUCTSHow do I select 3-5 products to recommend out of tens of thousands of options?
WHAT DATA SHOULD I USE?
Product attributes in the feed
The user’s behavior and activity
Common aggregate trends in behavior
HOW RECOMMENDATIONS WORK
Filter ProductsRule-based selection of
the set of eligible productsReorder the eligible
products by strategy score
Score ProductsBased on your
recommendation strategy
THE 3-STEP MACHINE
Sort & Present
SCORING BY STRATEGIES
INTRODUCING RECOMMENDATION STRATEGIES
CONTEXTUAL
PERSONALIZED
GLOBAL
By Products or Categories
MOST POPULAR
TRENDING NOW
NEWEST
SIMILAR PRODUCTS
BOUGHT TOGETHER
VIEWED TOGETHER
VIEWED AND THEN BOUGHT
COLLABORATIVE FILTERING
AFFINITY BASED
MOST POPULAR PRODUCTS Global
Now6 Months Ago
Purchase
Recent
Add to cart
Product view
● Weighted sum of all product interactions by all users
● Favors recent interactions
SIMILAR PRODUCTS Contextual (by products) Categories:
Men's Tops Short Sleeve Shirts Keywords: Stay Ready Stay Cool Loose Charged Cotton HeatGear New Arrivals Microthread
● Keywords and categories value comparison between the product in context and all other products in feed
● Factors in product popularity
BOUGHT TOGETHER Contextual (by products)
● Occurrences of product(s) in context purchased in the same transaction with other products
● Demotes products bought together with many items
AFFINITY BASED Personalized (by user)
● Derive user preference from interactions with products (real time + previous sessions)
● Reorder the most popular items by user preference of product attributes
COLLABORATIVE FILTERING Personalized (by user)
● Identify the products a user is most likely to purchase
● Based on what similar users have purchased
FILTERING PRODUCTS
• Dynamic Filters Using ‘Product Dimensions’ • Targeted Merchandising Rules
FILTER PRODUCTS BY PRODUCT ATTRIBUTES
● Match the viewed product in selected attributes (PDP)
● Differ from the viewed product in selected attributes (PDP)
● Category: Current / Parent / Any (PDP or Category)
INSERT DYNAMIC FILTERS USING PRODUCT DIMENSIONS
● Only Include (whitelist)
● Exclude (blacklist)
● Pin Product to Slot
DEPLOY TARGETED MERCHANDISING RULES
PRESENTING SORTED PRODUCTS
HOMEPAGE
Most Popular ProductsNEW VISITOR RETURNING VISITOR
PersonalizedAffinity Based
Filters: Match the user’s gender if known
PRODUCT PAGES
HIGH INTENT SIGNALS LOW / NO INTENT SIGNALSSimilar + Bought Together Similar
Viewed TogetherFilters: Match theme and category of product displayed
CART PAGES
ANY VISITORBought Together
Filters: Match items up to a certain price
THE DYNAMIC YIELD DIFFERENCE
TEST & TARGET DIFFERENT LAYOUTS & STRATEGIES
50% 50%
Users Condition
Yes No
MOST POPULAR SIMILAR
PRODUCTSMOST
POPULAR COLLABORATIVE FILTERING
FUSE MULTIPLE RECOMMENDATION STRATEGIES
“I’m just browsing around”
Surface related products & drive the user to convert
Do not distract yet increase cart value
“I’m looking for something specific”
“I want to buy now”
Encourage the user to explore products
MOST POPULAR
TRENDING NOW
NEWEST
SIMILAR PRODUCTS
BOUGHT TOGETHER
VIEWED TOGETHER
VIEWED AND THEN BOUGHT
AFFINITY BASED
COLLABORATIVE FILTERING
GlobalContextualPersonalized
INSERT RECOMMENDATIONS ANYWHERE
Personalize entire layout of your site and place recommendations anywhere on the page to drive most conversions
RENDER IN REAL-TIME, EVEN FOR EMAIL
Dynamic recommendations based on omni-channel data
USER 1
USER 2
USER 3
USER 4
ITEM 1 ITEM 2 ITEM 3
COLLABORATIVE FILTERING Personalized (by user)
ITEM 1 ITEM 2 ITEM 3
USER 1 0 1 1
USER 2 1 1
USER 3 1 1 1
USER 4 0 1
COLLABORATIVE FILTERING Personalized (by user)
DEPLOY SMARTER RECOMMENDATIONS AND DRIVE MORE REVENUE
REQUEST A FREE DEMO