Shortens product recommendation refreshing frequency from every few days to every two hours
Enhances user engagement by 2.5X with more relevant product recommendations
Enables swift development of the image search feature through Vertex AI
Automates platform policy violation screening and product categorization with Gemini
PopChill leverages Vertex AI in Google Cloud to enhance user engagement with more relevant product recommendations and search results, while ensuring safe trading on its ecommerce platform through automation.
PopChill leverages Vertex AI in Google Cloud to enhance user engagement with more relevant product recommendations and search results, while ensuring safe trading on its ecommerce platform through automation.
We needed to improve user engagement on our platform by providing more relevant product recommendations. The AI and data solutions of Google Cloud deliver the best results and require the least development work, because we can incorporate data from different sources more easily.
Roger Yang
CTO, PopChill
Pre-owned item trading platforms are loved by many people, as they offer bargains while promoting the circular economy. However, verifying the quality and authenticity of the items for sale can sometimes be challenging, especially for luxury products like designer bags or vintage clothing.
PopChill is dedicated to ensuring safe trading of pre-owned luxury items through its ecommerce platform launched in 2021. With PopChill's authentication service and trade-facilitating features, users are assured that all the luxury articles sold on its platform are original, and there are no concerns for scams during the trading process. Being one of the most popular pre-owned item ecommerce sites in Taiwan and Hong Kong, PopChill's platform provides more than 100,000 luxury items for sale.
"Pre-owned luxury item trading requires a high level of trust and security, and we've been dedicated to providing the best service in this domain," notes Roger Yang, CTO at PopChill. "On our platform, buyers can acquire authentic products at ease, while sellers can quickly sell their items at a good price."
To offer more personalized shopping experiences, PopChill in 2023 started developing a new product recommendation feature using the solution of the cloud platform where it deployed its ecommerce site.
However, the result fell short of expectations, because the solution could only track user behavior when users logged in. As it would have required too much development effort to improve the recommendation feature, the PopChill team decided to replace it with another solution that delivers higher performance. It chose Google Cloud for its seamless integration with Google Analytics, which the team had long been using for user data collection.
"We needed to improve user engagement on our platform by providing more relevant product recommendations," explains Yang. "The AI and data solutions of Google Cloud deliver the best results and require the least development work, because we can incorporate data from different sources more easily."
Realizing 2.5X higher user engagement with more relevant product recommendations
In early 2024, PopChill began to revamp its product recommendation feature leveraging Google Cloud. The team first built a data warehouse in BigQuery by transferring its business data through ETL and streaming the user and ad data from Google Analytics and Google Ads through export links to BigQuery. With only little development effort, PopChill has gathered all its data in one place ready for further use in Google Cloud.
PopChill's DevOps team then uses the recommendation models in Vertex AI Search for commerce to enable the product recommendation feature on PopChill's ecommerce site. By importing data of product catalogs and user events from BigQuery to Vertex AI Search for commerce, PopChill can now provide more relevant and timely product recommendations. Before, it normally took several days for the list of recommended products on its site to be updated.
Vertex AI Search for commerce has significantly improved the user engagement on our site by processing more diverse data with more sophisticated AI models. What's even more impressive is how little development work it requires.
Roger Yang
CTO, PopChill
Now, the recommendations are refreshed every two hours based on the latest user event data. Overall, the number of clicks on recommended items has increased by 2.5X.
"Vertex AI Search for commerce has significantly improved the user engagement on our site by processing more diverse data with more sophisticated AI models," notes Yang. "What's even more impressive is how little development work it requires. We only had one engineer working on the feature development for two weeks."
Enabling accurate product search with Vertex AI
With Vector Search in Vertex AI, we're able to greatly enhance the quality of the search results on our site and drive up the conversion rate.
Roger Yang
CTO, PopChill
Having successfully optimized its product recommendation feature, PopChill continued improving the user experience that it offers with Google Cloud. To help its users find the pre-owned items they want more easily, PopChill leverages Vector Search in Vertex AI to provide more relevant search results and enable image search.
Prior to the adoption of Google Cloud, the team used an open-source search engine to support its ecommerce platform's search feature. Since abbreviations and nicknames are commonly used by consumers to look for luxury brands, the open-source search engine was not able to generate all relevant results with its limited word-breaking and keyword contextualization capabilities.
By vectorizing product description and search keywords through Text embeddings API, as well as image data through Multimodal embeddings API, PopChill can now allow its users to find all relevant products with one keyword or image.
"Offering accurate search results can be challenging for luxury items, as users tend to search with keywords that need to be contextualized, and it's often easier to search with pictures. With Vector Search in Vertex AI, we're able to greatly enhance the quality of the search results on our site and drive up the conversion rate," says Yang.
On top of that, Vertex AI has helped PopChill accelerate the development process by providing developer-friendly tools like APIs. It took the team only two weeks to create a functional image search prototype with minimal resources.
Ensuring safe luxury item trading with Gemini
To guarantee safe trading on its ecommerce platform, PopChill used to rely heavily on manual work. For example, it needed one employee to spend half of a day reviewing chat messages between buyers and sellers to prevent fraud or platform policy violation. By training Gemini 1.5 Flash in Vertex AI to identify the messages that show intention to break the platform's rules and pass them to its staff for second review, PopChill is able to avoid human errors and enhance work efficiency.
Furthermore, PopChill utilizes Gemini to automatically categorize products and verify if product descriptions are correct. It also employs Vertex AI Vision to verify seller's identity through ID card images, which used to need one dedicated personnel.
The automation enabled by Gemini and Vertex AI Vision has freed our employees from repetitive work while reinforcing safety on our ecommerce site. We can now build trust among users and in our platform more efficiently.
Roger Yang
CTO, PopChill
"The automation enabled by Gemini and Vertex AI Vision has freed our employees from repetitive work while reinforcing safety on our ecommerce site," notes Yang. "We can now build trust among users and in our platform more efficiently."
Optimizing luxury item trading experience with AI
The excellent AI and data capabilities of Google Cloud have enabled us to effectively improve the user experience of our pre-owned luxury item trading platform in a short time. We believe that it will continue facilitating our business expansion and service optimization in the future.
Roger Yang
CTO, PopChill
Next, as scam messages spread through pictures become increasingly rampant, PopChill plans to use Multimodal AI to pick out scam text in images more effectively. To provide better user experiences, it will leverage Gemini to build a customer service chatbot and generate more appealing product descriptions based on images. Satisfied by the performance of the data and AI solutions of Google Cloud, the team will also migrate its infrastructure from another cloud platform to reduce cost.
Yang says, "The excellent AI and data capabilities of Google Cloud have enabled us to effectively improve the user experience of our pre-owned luxury item trading platform in a short time. We believe that it will continue facilitating our business expansion and service optimization in the future."
PopChill is a pre-owned luxury item ecommerce platform established in 2021. Providing authentication service and diverse features, PopChill ensures safe trading for both buyers and sellers.
Industry: Retail
Location: Taiwan
Products: Google Cloud, BigQuery, Vector Search, Vertex AI, Vertex AI Search for commerce, Vertex AI Vision
