From a manual, specialist-led process to a self-serve fixture of the Shopify Fulfillment Network. One triad, three disciplines, nine people. Sole designer on the workstream.
To bring products into SFN, a merchant emailed an onboarding specialist. The specialist replied with a spreadsheet template. The merchant filled it in. The specialist transcribed it into SFN's backend tools. Every product, every variant, every dimension by hand.
Average time to induct a catalog before products could even ship to SFN. The single largest source of merchant attrition between install and active use.
Stated job: support new merchants. Actual job: transcribe spreadsheets. 100% of bandwidth absorbed by manual induction. No capacity for the work the org needed them to do.
Slow merchant ramp on the front end. Specialist headcount sunk on the back end. SFN's value prop was speed, and the first step in the journey was the slowest.
The brief was "design a self-serve product induction flow." The real problem was that no UX, no matter how polished, would unblock merchants while the operating model still required a human in the loop for every catalog. The product to design was the workflow itself. The product to remove was the specialist from the loop.
Walking in: no precedent, no scope boundary, no definition of done. With no shared definition of done, every design conversation was a litigation. I wrote four principles and ran every decision through my team
A merchant should induct an entire catalog in a single sitting. No waiting on a specialist. No multi-day round trips.
Progressive disclosure over front-loaded complexity. The merchant sees what they need, when they need it.
Submitted = synced. No silent failures. No catalog limbo. The merchant always knows what's in SFN.
The end-state has no human in the loop. Anything less and the operating model doesn't shift.
A self-serve workflow the merchant never needs to ask a question about, and the specialist never needs to answer one about. Anything short of that left the operating model intact. If we shipped a polished UX that still required a specialist on the back end, we would have lost the bet.
Each one a no I had to say and a conversation I had to lead.
Partnered with the rest of the design pod to ship targeted UX enhancements to surrounding onboarding while reserving my innovation budget for induction itself.
Worked with that team's leaders to get the adjacent flow prioritized on their roadmap. Influence without scope creep.
Ran user testing and interviews when engineering pushed back on scope and timing. The data shifted the conversation from "what's cheapest" to "what does the merchant actually need."
Product induction is a workflow within a workflow. Each decision below was made against the four principles and backed by usability research.
A single long form was the obvious shape and what engineering wanted to ship. Research showed it was the worst possible shape for the merchant inducting hundreds of SKUs in one sitting.
Break induction into discrete steps, each a single decision the merchant could make without context-switching. Validation at each step, not at the end.
User testing showed 3× higher completion on the stepped flow vs. a single-form prototype. Cognitive load was the deciding factor.
Most merchant catalog complexity lives in variants: size, color, packaging. Missing variant information needed to be rectified before completing the workflow.
Surface a Variant Attention page as a dynamic step after selecting products. Add the right level of friction so merchants aren't surprised at the end of the workflow.
Variant-related induction errors dropped to near-zero after launch. The pattern was later adopted by other SFN teams handling complex SKU data.
Merchant catalogs are long. Attention spans aren't. Merchants inducting 200 SKUs in one sitting were failing halfway through and losing their work.
Show selection data in real-time, surfacing the monetary impact of inducting certain products (like best sellers). Help merchants optimize for sales while they work.
Pre-launch, the most common drop-off was 60-80% through induction. Post-launch, partial-submission users completed at 94%.
Saying no protects the work that has to be staff-quality. Choosing what not to touch is as much a craft decision as what to ship.
Adjacent roadmaps are part of your design surface. The next-step flow your work hands off to is your problem too, even when it isn't your scope.
User research is what unblocks engineering pushback at the cross-triad table. It changes the question from "what's cheapest" to "what's right."
In the induction step within a month of shipping. Merchants who installed SFN actually converted to onboarded users.
4+ hours of manual induction collapsed to ~8 minutes self-serve. The lowest time-to-completion of any onboarding step.
All 12 onboarding specialists redeployed to general merchant support within 4 months of MVP. The operating-model shift the org needed.
Induction moved out of onboarding-only and into the SFN app's primary navigation. Merchants can induct any time, not just during setup.
Other SFN workstreams used the induction pattern for their own product-data workflows. The pattern became infrastructure, not just a feature.
By the end of year one, the operating model had fully shifted. Specialists were on the support side, not the data-entry side.