Atelier: AI-Assisted Design Interactions in Bespoke Jewelry Sales

Abstract

In bespoke jewelry sales, translating a customer's vision into a visual design and price quote takes days. This thesis asks what happens when that feedback loop collapses to seconds. We present Atelier, a design tool that gives jewelry professionals seemingly instantaneous AI-generated visuals, CAD renderings, and price estimates during a live sales interaction. We conduct a within-subjects study with ten bespoke jewelry professionals across wholesale and customer-facing contexts. First, we find a trust asymmetry: professionals trust AI-generated visuals significantly more than AI-generated prices, not because of pricing accuracy but because of pricing opacity. Second, we find that willingness-to-adopt across modalities and features is shaped by the interaction context - wholesale versus customer-facing - rather than by the individual's role. We then iterate on Atelier in response to these findings and derive three design considerations for AI tools in consultative retail - trust scaffolding, curatorial control, and visible inputs and outputs - that generalize to any domain where professionals mediate AI outputs for someone else.