Mobboxed: How an AI Sales Agent Reduced Cart Abandonment by 25%

E-Commerce Has a 2 AM Problem

Refurbished electronics are a high-trust purchase. When a customer is buying a used iPhone for ₹45,000, they have questions. "Are there scratches on the screen?" "What is the battery health?" "Does the warranty cover water damage?"

For Mobboxed, a rapidly growing refurbished mobile seller, these questions were bottlenecking sales. If a customer browsed the site at 2 AM and had a question about battery health, they would leave. By the time the human support team replied at 10 AM, the customer had already bought from a competitor.

Building the AI Sales Agent

We didn't just want a chatbot that replies with FAQ links. We needed an agent that could actually sell. We built a custom AI integration using Next.js and the OpenAI API, but the real magic was in the RAG (Retrieval-Augmented Generation) pipeline.

When a user opens the chat on a specific product page (e.g., iPhone 13 Pro), the system immediately injects the live PostgreSQL database context into the LLM's system prompt. The AI instantly knows the exact stock count, the specific battery health of the device the user is looking at, the IMEI status, and the warranty terms.

User: "Does this phone have any scratches?"
AI Agent: "This specific iPhone 13 Pro (Sierra Blue) is graded 'Excellent'. The screen is flawless. There is a microscopic scuff on the bottom left bezel, which is why it's discounted to ₹42,000. It comes with a new charging cable and a 6-month warranty. Would you like me to add it to your cart?"

Direct Cart Manipulation

We gave the AI the ability to execute function calls. If a customer says "Yes, add it to my cart," the AI triggers a React state update that literally adds the item to the global cart state and slides open the checkout drawer.

The AI isn't just answering questions; it's actively guiding the user through the conversion funnel.

The Business Impact

The results were immediate. Within the first month of deployment, Mobboxed saw a 25% reduction in cart abandonment, specifically during off-hours (8 PM to 9 AM). The AI was successfully resolving objections and closing sales while the founders slept.

Furthermore, the human support team's ticket volume dropped by 60%, allowing them to focus entirely on complex returns and B2B bulk orders.

This is the future of e-commerce: not just a static catalog, but an interactive, intelligent shopping experience.