Shulex
Boosting Shulex’s Performance and Scalability in VOC Services with Zilliz Cloud
50%
less search latency and data analysis costs
30%
faster report generation
Minute-level
response to unexpected traffic surges
Since transitioning from the open-source Milvus vector database to the fully managed Zilliz Cloud, we’ve experienced significant improvements in business performance. We’ve achieved lower operational costs, increased search speed, a more flexible system architecture, and a more stable user experience. Zilliz Cloud also provides expert support to resolve issues quickly and effectively. Overall, Zilliz Cloud has given us greater convenience and a competitive edge, and we are very pleased and optimistic about this change.
Shengyi Pan
About Shulex
Shulex (Voc.AI), a leading AI company based in Silicon Valley, is dedicated to empowering sellers and brands to identify market opportunities and develop winning products through advanced VOC (Voice of Customer) solutions. By harnessing the power of AI, particularly large language models (LLMs) like ChatGPT and sentiment analysis, Shulex helps businesses gain deep insights into customer feedback. With over 200,000 registered customers, including global brands like P&G, Anker, and Hisense, Shulex's innovative VOC solutions have proven invaluable across a wide range of industries.
The Challenge: High Operational Costs with the Self-Managed Milvus Solution
VOC, or Voice of Customer, is a service that captures, analyzes, and leverages customer feedback to understand customer needs, preferences, and experiences. Shulex offers advanced VOC solutions to help brands gather and analyze data from various sources—such as surveys, reviews, social media, and customer support. This process ultimately enhances products, services, and overall customer satisfaction by aligning offerings more closely with customer desires.
Shulex needed a solution capable of understanding the intent and semantic meaning of customer feedback in natural language while efficiently managing large-scale data to achieve this goal. Initially, Shulex implemented a self-managed vector search solution using Milvus, an open-source vector database great for storing, indexing, and retrieving unstructured text data.
The self-managed Milvus solution initially performed well in storing massive volumes of vectors and conducting semantic searches. However, Shulex's rapid growth led to an explosion in data volume. The VOC review analysis service alone generated over 10,000 e-commerce-specific review labels, producing billions of vectors. While the self-managed Milvus solution could handle this large-scale data, it required significant internal resources for operation and maintenance. The high operational costs and the lengthy recovery times during system issues became problematic, leading to customer dissatisfaction.
Chenhui Li, Tech Lead at Shulex, explained, "As our business has expanded, the demands on our vector database have increased. We need a solution that minimizes our operational costs, offers elastic scaling capabilities to manage vast amounts of vector data and unexpected traffic surges, provides faster vector search speeds, and ensures a high service level agreement (SLA).”
After discussions with the expert team at Zilliz, the company behind Milvus, the Shulex team decided to migrate to Zilliz Cloud, the fully managed version of Milvus, to address these challenges. Zilliz Cloud now powers Shulex’s two core services: VOC review analysis and intelligent customer service.
Powering VOC Review Analysis with Zilliz Cloud
Shulex has been recognized as a top-ranked Amazon product selection tool thanks to its powerful VOC review analysis capabilities. Leveraging Zilliz Cloud, Shulex processes and analyzes vast amounts of Amazon reviews and social media data in real time, delivering clients instant insights into product feedback, including user demographics, usage scenarios, purchasing motivations, product strengths, and areas for improvement.
How the VOC Review Analysis System Works
Shulex combines the power of a large language model (LLM) with Zilliz Cloud to create a sample review database. This approach allows any new or unlabeled reviews to be compared against existing review data for similarity, making the process efficient and scalable. The system follows the following key steps:
Creating a Sample Review Database within Zilliz Cloud: Create a database within Zilliz Cloud to store original reviews, their vector representations, and sentiment labels (positive or negative) generated by an LLM.
Selecting Product Categories for Analysis: Choose specific products from a wide range of categories for detailed review analysis.
Encoding Reviews into Vector Embeddings: Convert tens of thousands of reviews from selected categories into vector embeddings and store them in Zilliz Cloud.
Generating Sentiment Labels with an LLM: Feed the selected reviews into an LLM like GPT-4, which assigns a sentiment label to each review. These labels are then stored in Zilliz Cloud and mapped with their corresponding review text.
Semantic Similarity Search with Zilliz Cloud: Zilliz Cloud performs a semantic similarity search on new or existing unlabeled reviews, matching them with the most semantically similar reviews in the sample database and retrieving the corresponding sentiment labels.
Generating Statistical Reports: The system analyzes customer sentiment related to specific product attributes based on Zilliz Cloud's classification and generates detailed reports.
Benefits of Zilliz Cloud for Shulex’s VOC Review Analysis
50% less search latency and 30% faster report generation: Zilliz Cloud offers superior vector similarity search performance with a search engine over five times faster than the open-source Milvus. It handles high-frequency searches at 1,000 queries per second (QPS), reducing search latency by 50% and improving report generation speed by 30%.
50% reduction in data analysis costs: Zilliz Cloud eliminates the need to run all product reviews through an LLM to obtain sentiment labels. Instead, it compares the semantic similarity between new or unlabeled reviews and those stored in the sample review database, significantly reducing the reliance on LLMs and cutting data analysis costs by at least 50%.
Minute-level response to unexpected traffic surges: During unexpected spikes in customer traffic, especially during promotional periods, the previous system required manual scaling, leading to wait times of up to an hour. Zilliz Cloud’s elastic scaling capabilities allow clusters to scale in or out within minutes, eliminating delays and enhancing customer satisfaction.
Building a RAG-based VOC Chatbot with Zilliz Cloud
Shulex also offers a VOC Chatbot that quickly transforms company and external data into FAQs and creates a professional customer service bot in just two minutes. This approach boosts response efficiency while cutting operational costs.
Built with the Retrieval Augmented Generation (RAG) technique, the VOC chatbot combines the power of large language models with vector databases. It extracts public web links, embeds knowledge from company files, emails, and support tickets into vector embeddings, and stores them in Zilliz Cloud. This method creates a company-specific knowledge base, enhancing the LLM with an external memory.
Zilliz Cloud enables the RAG-base chatbot to swiftly process and retrieve large volumes of vector data, facilitating real-time knowledge retrieval. The chatbot reliably supports Shulex’s intelligent customer service, handling 90 queries per second with a stable recall rate of over 98%. Now, the VOC chatbot manages over 80% of all customer service tasks.
Shengyi Pan, CTO of Shulex, commented, “Since transitioning from the open-source Milvus vector database to the fully managed Zilliz Cloud, we’ve experienced significant improvements in business performance. We’ve achieved lower operational costs, increased search speed, a more flexible system architecture, and a more stable user experience. Zilliz Cloud also provides expert support that resolves issues quickly and effectively. Overall, Zilliz Cloud has given us greater convenience and a competitive edge, and we are very pleased and optimistic about this change.”
Conclusion
Integrating Zilliz Cloud into Shulex's operations has proven to be a game-changer. By enabling faster, more efficient processing and analysis of vast data, Zilliz Cloud has significantly enhanced Shulex’s core services—VOC review analysis and intelligent customer service.
Zilliz Cloud's performance in high-speed semantic searches, reduction of data analysis costs by 50%, and rapid response to traffic surges have empowered Shulex to maintain its competitive edge in the market. As a result, Shulex can deliver deeper insights to its clients with greater speed and efficiency while keeping operational costs in check. The transition to Zilliz Cloud has streamlined Shulex's processes and provided a more robust and scalable foundation for future growth.
As our business has expanded, the demands on our vector database have increased. We need a solution that minimizes operational costs, offers elastic scaling capabilities to manage vast amounts of vector data and unexpected traffic surges, provides faster vector search speeds, and ensures a high service level agreement (SLA).
Chenhui Li