Fabric / Case study 02
AI Order Cloud
Designing AI around the work—not the novelty.
The challenge
Fabric wanted AI to become a meaningful part of its commerce platform. The easy answer was a chatbot. The harder, more valuable question was where AI could remove real operational friction without asking users to trust a black box.
Context
Why this problem mattered.
Merchandising and fulfillment teams work across dense systems, high-volume catalogs, and time-sensitive decisions. AI could accelerate that work, but only if it was contextual, explainable, and integrated into workflows people already understood.
My role
Where I led.
I led the design direction and helped move the initiative from an executive ambition to a shared product strategy. I facilitated alignment with executives and functional leaders, organized the design team around common principles, and partnered with product and engineering through concept development, testing, and systemization.
The approach
Turn evidence into direction.
- 01
Align before designing
I brought executives, department leaders, and key stakeholders together to identify where AI could create credible customer and business value. The workshop gave design a strategic role early enough to influence the premise—not just the interface.
- 02
Start with consequential work
Customer conversations narrowed our focus to merchandising and supply-chain leaders. Both groups spent significant time gathering data, maintaining accuracy, and moving between systems to make decisions.
- 03
Prototype the relationship
The team explored how users would move between conversation and action, inspect the rationale behind a recommendation, and recognize AI-generated content without losing their place or sense of control.
Key decisions
The choices that shaped the work.
Go deeper than chat
I pushed the direction beyond a bolt-on assistant and toward AI embedded in the workflows where people already made decisions.
Make trust visible
Recommendations needed rationale, clear provenance, and an easy path to accept, revise, or reject an action.
Create a reusable language
We translated the interaction principles into AI-specific patterns and components that could scale across the product.
Outcome
What changed.
49%faster data analysis
90%accuracy rating in beta
40%improvement in self-resolution
The beta results supported the central bet: AI was most useful when it helped people complete specific work with the context and controls needed to trust the result.