6. The visitor’s search query often differs from the product description
in the catalog.
Product
Description
How shoppers may describe it
Crafted of soft 100% cotton
with a herringbone weave and
clean mitered seams, our
exclusive teal pillow is a classic
update for any seating
arrangement. Pick up multiple
colors to refresh your decor
instantly and affordably.
blue pillow
blue couch cushion
turquoise cushion
aqua throw pillow *
*not mentioned in the
description
Product Name:
Teal Herringbone Cotton
Throw Pillow
And this is just the tip of the ice-berg
7. Quick taxonomy
search query signals intent
user has segment & intent
product has purpose
store front search results page
8. problem #1 : Cart Abandonments
Initiate a search
View products
Add to cart
Convert
9. Diagnosis : Cart Abandonments
Well formed queries
alphanumeric queries
Queries with exact product_ids
metric for seperation frontier
branded queries non-branded queries
others
MECE
mutually exclusive collectively exhaustive
discoverability & engagement gap
..to find
10. Diagnosis : Cart Abandonments
Diagnosis: viewing & adding products to cart, but not converting
Cause : Most popular sizes OOS !
Inference : People use carts to bookmark
11. • Custom sizes
• No standardization across category
• Size map
Challenges & constraints
Goal : Blend the availability of SKU's/sizes to (re)rank the products
Pre-cursor: Need to know the real distribution of sizes, across
categories
score(rank) = f(availability factor,x2,x3,x4..)
13. Product design
re-rank the products where the availability is factored in by size-
popularity
availability factor =:
rate of fill of inventory [Supply]
rate of depletion of inventory [Demand]
Opportunity for the merchant is to align these
and fill inventory
15. Challenges & constraints
conflicting goals : prevent starvation vs. Business
performance
http://sayrohan.blogspot.in/2013/06/finding-trending-topics-and-trending.html
The Britney Spears Problem : Tracking who's hot and who's not presents an algorithmic challenge
http://www.americanscientist.org/issues/pub/the-britney-spears-problem/1
• huge demand generation, often not in line with intent
• short-lived events - too small a period to let the algorithm learn
• need to separate the trend from popular events
• fair bootstrapped impressions do not work
16. Solution is a mix of opportunities
new products , new intents & (reverse engineering) merchandizing plan
19. women's villanova wildcats navy blue classic arch full zip hooded sweatshirt
www.we-sell-stuff.com/COLLEGE_Villanova_Wildcats_Sweatshirts_And_Fleece
www.we-sell-stuff.com/prod-nm2614
women's sweatshirt
www.we-sell-stuff.com/NHL_Minnesota_Wild_Mens/pg/1/ps/72/so/newest_items
www.we-sell-stuff.com/NHL_Minnesota_Wild_Mens/pg/1/ps/72/so/top_sellers
www.we-sell-stuff.com/NHL_Minnesota_Wild_Mens/pg/1/ps/72/so/highest_price
www.we-sell-stuff.com/NHL_Minnesota_Wild_Mens/pg/1/ps/72/so/lowest_price
www.we-sell-stuff.com/NHL_Minnesota_Wild_Mens/pg/1/ps/72/so/top_rated
Exclusivity
Vanity
Quality & utility
spend thrift
Understanding user segments
redirect
product/category page
www.we-sell-stuff.com/shop/wd/womens-dresses?
quantity=144&evansignore=10051&filtered=true&catFilter=140325&cmp=SOC:ANF_BND_US_FBK_PRD_FMLdresses -
campaigns/promotions/repeat users
www.we-sell-stuff.com/webapp/wcs/stores/servlet/Search?search-field-submit=SEARCH&catalogId=10901&search-
field=tops&cmp=PDS:ANF_US_BNG_BRD_General-Tops&kid=6ccb15d4-7758-f1a9-bb46-0000732cf85f&langId=-1&storeId=11203
Queries that have campaigns
20. decoding user experience
• Pagination depth of users across queries - which queries are worse off?
• Price sensitive vs. Brand sensitive users - re-ranking & personalization
www.we-sell-stuff.com/search=hoodie+sweater&pn=2
www.we-sell-stuff.com/search=final+four&pn=4
Cluster brand facets vs. price signal facets to infer user-segments
22. consumers have different taxonomy
products have a purpose & positioning
What we know so far
match this well
23. Be metric driven
Common sense math beats intense data science :)
look beyond your cursory tool-kit
Match intent to purpose of the product
segment intimately till a separation frontier emerges
Reverse engineer quantitatively and then commoditize at scale
Isolate. Synthesize. Commoditize. Scale