Trusting AI with high-ticket orders.
How a made-to-order jewellery brand put guardrails around fdx and let it run bespoke orders, without risking the brand.
A made-to-order jewellery house
Heroyne designs and sells personalised, made-to-order jewellery. Every custom piece is a small project: metal, stone, sizing, engraving, timeline and price, agreed with the customer before a maker ever picks it up.
AI on high-ticket orders is dangerous
On a piece worth hundreds or thousands, one wrong quote, a promised stone that isn't in stock, or a timeline the makers can't hit doesn't just lose a sale, it erodes the brand. That risk is exactly why most jewellers won't let AI near their bespoke orders, and why Heroyne was wary too.
Guardrails and deterministic workflows
fdx doesn't improvise. Heroyne's bespoke process runs as a defined, repeatable workflow: the agent only quotes against the live catalogue and real lead times, stays inside pricing and policy limits, and pauses for human approval on every high-value decision. Every action is logged and reversible.
Confidence to let it run
With guardrails in place, Heroyne trusts fdx with its highest-value orders. The agent handles the conversation and the legwork; the brand keeps control of every call that matters, and nothing goes to a customer unchecked.
“On a piece worth thousands, one wrong quote can cost you the customer's trust. The guardrails are what let me say yes to AI.”