Purpose, Speed & Visibility : Facilitating product discovery & engagement on a e-commerce website
Submitted by Ekta Grover (@ekta1007) on Friday, 29 April 2016
Each product on an ecommerce website has an opportunity to sell and market dynamics determines what’s selling and at what speed . This has Merchandising implications for stock re-fill, flash sales, promotions & special events - along with the actions a merchant’s platform team takes in anticipation for such events. By reverse engineering this quantitatively, and tuning the proprietary Search ranking signals, we can meet a Merchant’s business goal both from a discoverability and bottom-line incremental revenue perspective.
This session is about 2 main things –
1. Search & Personalization : Understanding a user’s footprint & quantifying the weblogs to meet two broad goals - facilitating discoverability and driving user engagement .
2. Hypothesis driven engineering - 3 specific problems we solved at Bloomreach - while pushing them upward as proprietary signals in Product design
I will start with defining a user’s intent, web journey and how products sell on an ecommerce website - defining the “discoverability” and “engagement” goals. I will then introduce relevance, recall & performance/engagement based discoverability from a search ranking perspective.
Though the two goals above(discoverability & engagement) look inclusive, I will showcase using the 3 problems, how conventional metrics such as Bounce rate, product view rate, ATC rate, Conversion rate etc. can mask serious challenges that end-users face - thus opening up opportunities about a Merchant can actually address for its users.
Problem # 1 : Understanding lost carts
Data from a merchant with abandoned carts - where we figured out it was a problem with stock-replenishment of popular sizes and people were using carts to “bookmark”. While we do not control/influence a merchandiser‘s inventory - how could we change “the search storefront” (aka the search results page) - to reflect the availability factor that blended the popularity factor - also introducing the supply & demand dynamics at this point.
In the process, we ended up building a clustering solution for mapping sizes across different categories, which then fed into the availability factor.
Problem # 2 : Handling Special events - Mother’s day, Back to school, Halloween & holiday sales , special sports events etc. , Marketplace products & New launches
Using data from the sudden redirect pages that a Merchant’s platform team sets up - to understand a Merchant’s Business goals, quantitatively and then reflecting this in “discoverability” score I introduced before. This means that the evolving intent for these special events now ties to the hot products that merchandizer knows best will sell.
Also, all products are created unequal & by revisiting to a product’s purpose - we can proxy a user’s intent. Since a product only exists in the realm of a “user intent” - bootstrapping some fair impressions is more challenging than it looks.
One goal could be ensuring the new products have some impressions before they starve & set a downward spiral in the ecosystem, but this when also coupled with non-sellable/zombie products that a merchant introduces ahead of launches (eg iphone 6s, Motorola next gen etc.) skews the performance data. eg, some products get a lot of impressions, but do not sell since they are not sellable -this problem will showcase how do we flex the two uber goals of discoverability & engagement for products.
Problem # 3 : Understanding users segments on a website
We can’t fix what we don’t understand is broken. Aside of search, users use very suggestive sort parameters, dynamic filters, paginatation - this problem will showcase data about “intent classification” that helped our search tuning efforts. Brand sensitive vs. price sensitive users are a different breed - and this has search storefront implications, not just for what to show (close substitutes,complementary goods, related but neither substitutes, not complementary goods)- but also on how we measure the intent’s differently. There is no such thing as an average user, afterall.
Ekta Grover is a Member Technical Staff & Data Scientist at Bloomreach Inc, a firm that helps B2B and B2C companies make their content more discoverable, relevant & personalized, while growing the bottom line for the Businesses. At Bloomreach she focuses on influencing proprietary Search ranking signals in core search, commoditizing analysis and shaping decisions that impact bottom line Business metrics for some of the largest retail merchants.
She has experience leading & building data tools in location based Mobile advertising, e-commerce, Search & Personalization. Across her professional footprint, she has worked with both Enterprise product companies (SAP, VMware) and midsize/small startups in consumer web - and has helped shape products that affects million of users and billions of dollars in incremental revenue for some of the largest e-commerce firms. She has a background in Quantitative Economics & Computer Science.
Her previous talks with HasGeek
Fifth Elephant Conference, 2014 : Experimentation to Productization : developing a Dynamic Bidding system for a location aware Mobile landscape
Pycon 2013 : Experiments in data mining, entity disambiguation and how to think data-structures for designing beautiful algorithms