Work / AI-Empowered Seller Dashboard
Building an AI-enabled B2B commerce platform that empowers 2,000+ merchants to list faster and ship smarter.
Ezze Live is a fast-growing live commerce platform for unique items like vintage fashion, sneakers, and trading cards. As our global audience grew, the business expanded from a single sales channel into a multi-channel B2B platform for merchant scaling.
My ResponsibitliyAs Lead Product Designer, I owned the 0-to-1 strategy and execution of the Ezze Seller Dashboard, architected an AI-empowered web system that aligned complex global supply chain logistics with the demands of high-volume sellers.
Lead Product Designer
Team
Product Managers, Engineering, Sellers
Timeline
2022–2025
Expanding to a multi-channel ecosystem broke our legacy workflows. As product complexity and order volumes surged, listing and fulfillment led to extreme merchant fatigue and severe profit erosion.
Challenge #2 Fragmented Shipment Chaos
Isolated shipments made separate processing mandatory. Sellers were forced to generate individual labels and pack separate boxes for every transaction, causing redundant labor, high shipping costs, and a staggering error rate.
Built for the platform's early days, the initial architecture was rigidly single-channel and highly manual.
*Stage 1, Select product type & Default product form*
Part #1 Listing
Sellers typed every product detail from scratch for every unique item. There was no in-page continuity. Switching sales formats or adding the next item required exiting the page and restarting the entire process.
*Stage 1, Add tracking*
One item equaled one tracking number. If a buyer purchased 5 items, the system generated 5 separate labels, requiring sellers to physically pack 5 individual boxes.
Lacking in-page routing, simply changing a product's destination from "Live" to "Store" meant aborting the task and starting over from a blank form.
Issue #2 Fulfillment Overload
The new channel triggered a surge in order volume, making the 1-to-1 manual shipping model physically and operationally unscalable.
To understand how these system limitations impacted daily operations, I interviewed three distinct merchant types.
Pain PointsListing volume is manageable, but fulfillment is highly inefficient. When a buyer purchases multiple pairs, the system generates isolated orders. Shipping separate bulky shoeboxes to the exact same address wastes shipping costs and materials.
User NeedsA workflow to process and group multiple large items for the same buyer together.
Pain PointsEvery vintage piece is a 1-of-1, requiring sellers to manually type unique product details from scratch. Loyal fans often buy multiple items, but the system forces sellers to ship each item in a separate package.
User NeedsA faster method to input visual product details without repetitive typing. The ability to pack a single buyer's purchases into one physical box.
Pain Points Shipping is straightforward for micro-items, but listing is highly time-consuming. Manually typing hyper-specific card details for hundreds of individual cards severely slows down operations.
User NeedsAn efficient way to capture dense, specific product information without slow manual keystrokes.
To address the Stage 1 blockers, we launched a transitional update to support multi-channel scaling.
&What We Built...
*Stage 2, Product From w/ AI Trigger*
Both Live Sales (Buy it Now, Auction, Giveaway) and Store were integrated into the same page as independent sections, allowing sellers to list for both channels simultaneously.
#2 Manual AI Assistance
Introduced a click-to-trigger AI tool for image and text extraction.
*Stage 2, Automatically merged orders & Ship now (Add tracking)*
The system automatically grouped items purchased by the same buyer into one shipping label.
Testing this update with our initial merchants revealed a new reality.
The Synthesis: New Operational Frictions
#1 Listing Friction
The monolithic form structure offered zero error isolation, making the process high stakes. Combined with a manual trigger AI, the workflow remained repetitive and fragile.
#2 Fulfillment Friction
Auto bundling was too rigid. It ignored physical constraints like volume and weight, and lacked the flexibility to handle simple order modifications like partial cancellations.
To eliminate the frictions from Stage 2, I led a ground-up redesign focusing on automation, error isolation, and flexible fulfillment.
Part #1 Intelligent Listing
Auto-Fill AI Replaced manual triggers with a seamless "Image-to-Info" flow. The AI automatically extracts product details and predicts package dimensions immediately upon image upload.
Smart Memory The system auto-applies frequent seller choices for categories and shipping. By learning from past behavior, it provides high-probability defaults that allow sellers to breeze through recurring decisions.
Isolated Channel Editing To manage expanding channels, I replaced the monolithic long form with Scoped Editing Modules. Channel-specific edits are confined to independent overlays, ensuring any error is isolated and never compromises the entire product entry.
*Stage 3, Product Form w/ auto AI*
*Stage 3, Product Form Prototype*
Smart-Suggest Bundling Shifted from "Forced" to "Suggested" bundling. The system flags potential bundles but requires seller confirmation, allowing for manual overrides when items are too bulky or fragile.
Independent Order Management Each sub-order within a bundle maintains its own clear interface. Sellers can process cancellations or partial refunds for a single item without disrupting the workflow for the rest of the shipment.
*Stage 3, Bulk shipping*
*Stage 3, Order detail w/ tracking*
*Stage 3, Add Tracking Prototype*
Key User Testing Insights
The feedback confirmed that shifting from manual input to intelligent validation successfully resolved the fatigue of high-volume sellers while providing the flexibility needed for complex logistics.
#1 AI Auto-Fill & Smart Memory
By instantly extracting details from photos and remembering repetitive attributes, the system turned exhausting data entry into a quick verification step. Sellers reported that raw listing times were cut in half, allowing them to process dozens of unique items without heavy keystrokes.
#2 Smart-Suggest Bundling & Independent
Orders Merchants gained the speed of automated label grouping alongside the critical authority to manually split shipments for bulky or fragile combinations. Furthermore, the independent order interface allowed them to process partial refunds without breaking the entire grouped shipment.
Addtionally, #3 Isolated Channel Editing
Scoped modules completely eliminated the anxiety of losing massive forms. Sellers can now safely adjust storefront pricing or live auction bids independently, ensuring that an error in one channel never overwrites the core product data.
Launched in early 2025, the intelligent ecosystem successfully onboarded and scaled over 2,000 registered sellers, driving a significant increase in multi-channel adoption and daily active listings.
3x
faster listing speed
40%
reduction in fulfillment errors
150%
increase in daily active listingsNext project →