foryou.png

Carousell: 'For you' Recommendations

Carousell 'For you':

Recommendations tailored to your preference

 

Context Carousell 2022 Q2

Role Lead Designer

Timeframe 3 months

Phases Discovery, ideation, research, delivering final assets

Collaborators Product Manager, Engineers, Data Scientists

 
 

Project Overview

The current buying process is cumbersome, requiring multiple steps and repetitive searches. The new 'For you' tab aims to streamline this by allowing users to seamlessly resume their shopping and track items of interest.

 
 
 
 

 

Problem

The dominant buying behavior on Carousell is search. However, due to the unpredictable nature of secondhand supply, it often requires buyers a few times of searching, waiting, and finally finding something that matches their expectation.

 
 

How might we help users easily follow-up on items they’re searching for and encourage them to browse items that they have a high-interest in?

 
 

It takes a lot of effort to find something users want

 
 
 

 
 

Goals

Business goals

  • Increase user engagement on homescreen

  • Increase views, impression, and quality chats

User goals

  • Easily pick up from where they left off

  • Find items they are looking for faster

 
 

 
 

Solution

 
 

Recognition over Recall

Allowing users to browse instead of manually inputting keywords simplifies the process so that they can follow up on items easily.

User control of preference

Users can remove keywords they no longer want to see.

Relevant keywords

Showing users the most recent and relevant searches in the front. The keywords refresh every few hours to show new listings.

 
 
 

 
 

Design Process

 
 

User interview goals

  • Understand users’ process of shopping/finding items/browsing

  • How users follow up on items they like/search for

  • How they re-engage with previous content

  • What touch points users utilize to find items they’ve shown interest in

 
 
 
 

Explorations and Concept Testing

I created several types of ‘retargeting’ content to test which ones resonate with users the most (see mockups below).

Along with our UX researcher, we interviewed 10 users - 5 of whom are light users (browsed less than 5 sessions in the past 3 months) and 5 of whom are heavy users (browsed more than 5 sessions in the past 3 months).

 
 

Learnings from concept testing

  • “Continue shopping for” section is able to make recurring and iterative searches less tedious

  • Incorporate # of searches, like, chat signal to identify user’s high intent searched keywords

  • It’s important to know why something is recommended - showing keywords or reasons, but relevance of suggestions is still key

  • Vertical scrolling is a more intuitive browsing experience

 
 
 

More design explorations

 
 
 

Defining the architecture of the ‘Explore’ tab and the ‘For you’ tab

Prior to the introduction of the ‘For you’ tab, there was the ‘Groups’ tab, which was not yielding good numbers and engagement from users and was not maintained by any team. I proposed to remove the ‘Groups’ tab from the bottom bar and replace it with this new tab.

 
 

Explore

When you’re feeling explorative to find deals, trending/cool items, free items, and items nearby on Carousell.

20% personalization, 80% exploration

For you

A feed customized by you and for you. All things you want to follow up on: favorite sellers and categories, saved keywords.

100% personalization

 
 
 
 
 
 

Fine-tuning the ‘For you’ algorithm with Data Science

 
 

I worked with the product manager and data scientist to come up with the algorithm most attractive and useful for users.

Content interest

  • Recommend similar items they liked/clicked/chatted


Listing attribute preference

  • price (similar price range x% from the listings they clicked)

  • size range, for fashion

  • brand, for fashion, electronics and luxury
    condition

A/B Testing

  • We tested several types of algorithms to see which one was the most engaging for users

 
 
 

Extension to other markets 🇸🇬 🇭🇰 🇹🇼 🇲🇾 🇮🇩

Since the launch of this feature was a success in Singapore, we wanted to also extend it to other markets including Hong Kong, Taiwan, Malaysia and Indonesia. However, we noticed that the search behavior is slightly different in the Mandarin-speaking markets.

With Mandarin characters, users often search several variations (including abbreviations) of the same thing in different order.

 
 

The suggested keywords end up having similar variations (repetitive), which were not helpful for the users. Therefore, we did some tweaks on the algorithm to make sure that users have a variety of different enough keywords to browse.


 
 

 
 

Results

The experiment led to significant impact on listing views, responses, and quality responses from homescreen (Explore + For you tabs). It was a feature that yielded the highest increase in engagement from homescreen compared to all features that have been released, which means that we increased viewing on the top funnel and that users were browsing more items than before. The replacement of the old ‘Groups’ tab to the new ‘For you’ tab resulted in higher usage and engagement overall.