Transforming Marketplace Support with Savvycom’s AI-Driven Recommender System

Our client specializes in business transformation through cloud technologies, delivering tailored solutions to enhance competitiveness for businesses across APAC and Southeast Asia. The goal is to optimize their all-in-one customer service tool. This tool connects businesses with customers on Shopee, Lazada, and TikTok, integrating seamlessly with Zendesk and Sunshine Conversation to manage customer interactions efficiently.

Resource Quality and Response Time

The client struggled with inadequate resources, impacting their ability to address customer queries and issues promptly. This lack of quality resources led to inefficiencies in managing customer support and hindered their capacity to maintain high customer satisfaction levels.

Complex Customer Interactions Across Multiple Marketplaces

With a diverse presence across multiple marketplaces like Shopee, Shopify, and Lazada, managing customer interactions became increasingly complex. Each platform had its own set of communication tools and customer service protocols, making it difficult for the client to provide a consistent support experience across all channels.

Build a Recommender System

Create a model using shop data to generate personalized product recommendations for buyers. This system needed to be integrated with Zendesk and provided as an app for various marketplaces, including Shopee, Shopify, and Lazada.

Integrate Marketplace Communication with Zendesk

Develop a centralized solution to manage customer tickets from different shops in one place. This integration would enable them to offer comprehensive customer support through Zendesk for all marketplaces. Additionally, agents should be able to offer product recommendations based on the recommender system’s outputs.

Marketplace Integration

• Python Apps: Savvycom developed custom Python applications for each marketplace (Shopee, Shopify, Lazada) using their REST APIs. These apps synchronized shop and seller data, including orders and products, with a MongoDB database.
• Recommender System: A recommender application was built using Python and TensorFlow. This app utilized the synchronized shop data to train models that generated personalized product recommendations for each customer.
• REST API: A REST API was created using Python and FastAPI to interact with the recommender app, allowing for model training and retrieval of product recommendations.

Data Management

• CRUD Operations: A separate REST API was developed using JavaScript and NestJS to handle CRUD (Create, Read, Update, Delete) operations with the shop data. This ensured efficient data management and accessibility.

Amazon Chat Integration

• Email-Based Solution: Since Amazon does not provide a Chat API, Savvycom devised a custom solution using email messages as a replacement for chat. This involved setting up and configuring a mail server using Dovecot and Postfix.
• Mail Server Interaction: A REST API was developed using Python and FastAPI to interact with the mail server, allowing agents to read and reply to messages.

…

By implementing these solutions, we addressed the client’s challenges, achieved their business objectives and enhanced their overall operational efficiency.