Chat-based AI Shopping Agent

NoviBox is an online gift shopping platform aiming to enrich everyday lifestyles with uniquely designed products. Their research revealed that most users spend a considerable amount of time browsing and selecting gifts, resulting in prolonged decision-making, user drop-off, and lower purchase conversion rates.

To address this, NoviBox set out to introduce an AI shopping assistant to enhance the gift selection, purchasing, and post-sale experience through data-driven personalized recommendations.

We began with desk research that surfaced a striking insight: 47% of consumers would buy items via a chatbot. A competitive analysis of Urban Outfitters, Amazon, and Bing Copilot revealed a fragmented landscape — clean but rigid preset flows, strong after-sales UI with no purchase recommendations, and powerful AI suggestions that required redirection to other platforms.

A survey of 100 respondents (15 questions across user needs, interaction styles, and feature design) confirmed that 93% already understood AI-driven features and chatbots, validating a high baseline of user readiness for an AI-first shopping experience.

The first core flow lets users explore products through an AI-powered swipe experience. By simply collecting users’ preferred styles, the assistant curates a personalized feed and guides users to add items to their cart or purchase directly within the chat.

Usability testing revealed that users needed a more prominent “Review My Favorite List” prompt, along with discount information and detailed product descriptions surfaced earlier in the flow.

The second flow comprehensively gathers users’ gift-buying needs — occasion, recipient, budget — and returns a refreshable recommendation list with three potential matches at a time.

This flow achieved a 92.9% task success rate with only a 7.1% drop-off. However, the initial misclick rate of 37.5% prompted us to add a direct “View Product Details” button so users could build comprehensive understanding before committing.

For post-purchase support, the assistant collects user feedback, provides human customer support hand-off, and surfaces policy and refund/return services — all within the chat interface.

The Order Help module supports after-sales services like quick order inquiries, address modifications, order cancellations, and return/refund processing, keeping users inside a single conversational flow rather than navigating disconnected help pages.

We tested two interaction paradigms across the full prototype suite. 69% of testers preferred Version A, which reduced the user misclick rate by 24% compared to Version B.

The checkout flow surfaced a critical insight: an 82.5% drop-off rate at the add-to-cart stage, caused by the absence of a direct purchase button. Users had to view product details before purchasing, and the “View Details” button lacked clear visibility. These findings drove a redesign that surfaced purchase actions earlier and more prominently.