Expert Interview Series #2 Jiwon Hong
Jiwon Hong, August 2020
Currently I am CEO & Co-founder of YesPlz, the next-gen visual search for eCommerce. Prior to YesPlz, I launched the search and recommendation engine for SmartTV at Samsung to the global market. I am committed to transforming online shopping search experience to be easy and delightful.
Generally speaking, a Fashion Recommendation Engine (FRE) extracts the key attributes from a product and finds items with matching attributes, such as silhouette, material, price and brand. Most of all, a recommendation engine should find clothing the customer likes.
The most well-known recommendation engine is collaborative filtering which you probably have seen at Amazon. At the end of the product detail page, Amazon shows a list of items under the title “Customers who viewed this item also viewed.” This recommendation engine assumes a shopper will like items that other shoppers viewed after checking out the original item.
A traditional fashion recommendation has been recommending clothing based on the same brand or similar product title. The richer the keywords, the better the recommendations work. Except creating product titles and descriptions is expensive.
With computer vision technology, a recommendation engine has evolved to detect the key attributes from an image and recommends styles accordingly. No text information is required. A photo of a long tank top will search another 100 similar images of long tank tops assuming a customer is looking for that particular style. Even better, sophisticated vision technology narrows a search to the fine details such as a slim fit waist and a flat neckline for the long tank top matching a customer’s specific preferences.
If a retailer’s website is based on eCommerce platforms such as Shopify or Magneto, a retailer can simply go to their app store and download a recommendation app. It only takes one or two clicks to install. The pricing is about $7 -$35 monthly which is much cheaper than building an in-house engineering team.
Enterprise clients can test a recommendation engine through API that can be integrated in as early as a few weeks. The cost for an enterprise varies by size, but A/B testing can be a great way to test a recommendation engine performance and set a price. Most solution providers offer either an introductory rate or a free trial for A/B testing.
It is easy to test different solutions and see which recommendation engine fits your business best. There is no reason to delay offering this virtual shopping assistant to your customers.
Well-curated recommendations transform the shopping experience, making a shopper happy and wanting to come back. Plus, it is impossible to manually curate the recommendations for each individual customer.
Over time, a good recommendation engine learns an individual’s shopping preferences, if a customer allows it. Imagine a virtual shopping buddy who deeply understands a customer’s preferred styles that can quickly find the right item! That capability can be on your website now.
Too many retailers are stuck with static, expensive product recommendation systems that aren’t delivering accurate, fashion-forward recommendations to customers. That’s why we decided to create the YesPlz Product Recommendation Engine.
By utilizing the Style Filter and YesPlz Product Recommendations, Lately and Kolon Mall were able to create the best-possible online shopping experience for customers.
It's hard to know what customers are really looking for, and the fashion taxonomy can complicate the search process even further. In this piece, we demystify the fashion taxonomy and explain how understanding customer search intentions can shape your eCommerce strategy.