Replacing a rigid ordering system with intelligent search, multi-cart management, and the foundation for a self-learning product recommendation and stock management engine.
#b2b2b #b2b2c #ecommerce #multi-cart #dropshipping #ML #AI #checkout #UX #informationarchitecture #IA #wayfinding #keywords #orderfulfillment #smartsearch #distribution #imagingproducts #machinelearning
Supplies Network is a managed print and supply services company — think the sourcing and fulfillment backbone behind your office’s printer supplies and imaging equipment. Multiple distribution centers across North America enable fast drop-shipment directly to end customers
The B2B2C model is what makes it interesting: Supplies Network’s customers aren’t retail stores — they’re IT service companies, managed print providers, and similar businesses who support their own clients’ technology needs. When a local IT shop needs to get printer supplies or imaging equipment to the cupcake shop down the street, Supplies Network fulfills that order. Everything ships with the IT company’s branding. To the cupcake shop, it looks like their IT provider has deep product resources. Supplies Network is invisible by design.
This model created complex ordering needs that the existing portal could not support.
The existing customer portal was built on rigid, outdated architecture. It looked and functioned like a system from 2002 — because it was.
Search required exact SKUs. Customers had to know the precise part number to find a product. A unix-based lookup forced users to navigate rigid category and subcategory hierarchies. If a customer searched for “hp printer ink” and didn’t use the right terminology, they got nothing useful back.
Ordering was painfully manual. Customers often needed to place orders for 20 or more of their own unique end-customers in a single session. There was no way to manage multiple shipping destinations efficiently. Customers got so frustrated with the volume of orders that they would call Supplies Network directly and have someone on staff place the orders for them.
No product intelligence. There were no related product suggestions, no alternative recommendations, no way to surface compatible items. If a product wasn’t in the catalog, the system returned a JavaScript alert dialog telling the customer to call their sales rep.
Checkout was a gauntlet. Shipping options, payment, and address entry were spread across multiple disconnected screens with inconsistent patterns. Address entry had no validation or standardization.
I was the sole IA on the project, responsible for:
The old checkout was a series of disconnected screens. I consolidated the flow and introduced key improvements:
Address standardization — when a customer entered a new shipping address, the system validated it against standardized formats and offered similar matches when the exact address wasn’t found. This reduced shipping errors and returns.
Ship-To vs. Drop-Ship clarity — clear explanations of what each option means for the packing list and labeling, since the distinction matters enormously in a B2B2C model (the end-customer’s name goes on drop-ship labels, not the reseller’s).
Microcopy with purpose — throughout the wireframes, I focused on copy that built confidence. “Save Address” replaced a generic “OK” button. “Cancel” was de-emphasized as a hyperlink to reduce accidental clicks. Tooltip explanations appeared for shipment types. Every label was written to reduce the cognitive load on customers processing high-volume orders.
Beyond the portal redesign, I developed the initial logic for what would become a learning system. The concept: by analyzing customer order patterns — what products were purchased together, how frequently, by what types of resellers — the system could begin to predict warehouse stocking needs and suggest related products.
This manifested in the wireframes as “Related Products” and “Related Services” modules in the search results sidebar, which surfaced contextually based on what the customer was viewing. The keyword-weighting system was designed to get smarter over time as more purchase data flowed through it.
The architecture was built on first principles: customer purchases create tiered connections between products, which surface as faster search results, smarter suggestions, and for the distribution centers, the ability to predict stocking needs based on common purchase pairings.
The single biggest architectural change was moving from rigid hierarchies to a keyword-weighted search. I developed the entire metadata structure and search architecture independently, presenting it through my internal account manager with minor adjustments before pitching it to Supplies Network’s full boardroom.
After the pitch was approved, their software development manager worked alongside me to restructure how the data was organized for every item in their database — but the information architecture, the keyword-weighting logic, and the search experience design were entirely my work.
The new search leveraged metadata at every level — customers could search by product name, SKU, brand, compatible printer model, or even common misspellings and still find what they were looking for. Results surfaced with faceted filtering by brand, price range, yield, color, and whether items were remanufactured. Categories appeared as clickable links above results for quick refinement.
This wasn’t just a better search box. It was a fundamental re-architecture of how product data was organized, tagged, and surfaced.
The most significant UX challenge was the multi-cart system. A single Supplies Network customer might need to ship to dozens of their own end-customers in one session. The old system had no concept of this.
I designed a persistent “Active Cart” selector that lived in the global header — always visible, always accessible. Each cart was tied to a specific shipping destination (customer name + address + zip code). Customers could switch between carts at any point, and a “Create a New Cart” option sat right in the dropdown alongside existing carts.
Product availability was tied to shipping zip code, which meant each cart needed its own warehouse fulfillment calculation. The system displayed real-time warehouse stock levels by region (MO, TX, CA, PA) on product detail pages, with a quick availability checker by zip code.
The shopping cart itself was designed to handle the complexity: line items with quantity, pricing, back-order status, and a clear subtotal — with the multi-cart context always visible so customers never lost track of which end-customer they were ordering for.