Case study — a forward-deployed-engineering scope, live

Three systems that never talked to each other. One answer to "why was it late."

Auric & Stone ships fine jewelry from Milan to Asia Pacific and the Middle East. Their order system, warehouse (WMS) and carrier (TMS) each define "on time" differently — and none of them know what the others recorded. Waybill unifies all three and answers delay questions with the actual evidence trail, not a guess.

—
orders unified
—
late, % of shipments
4
disconnected source systems
2
conflicting timestamp definitions resolved
The unglamorous 90%

Building the agent took a day. This took the rest of the week.

This is what actually happens in a forward-deployed engagement: none of these systems share a key, a timezone convention, or a definition of "ready to ship." Below is the real seam between them — the same class of problem behind most late-delivery disputes.

Resolved crosswalk — the field two teams disagreed on

Loading…

Ask it directly

Ops questions, answered with the evidence trail

Try an order id, a customer name, or an aggregate question — Waybill decides which systems to query, sometimes in more than one step, and shows that work before it answers.

W
Waybill · Auric & Stone Global Distribution
● unified across 4 systems
demo · synthetic data
Why was order AS-2026-0023 late? Which orders to Riyadh are late right now? Give me this week's delay report
How it was built

The forward-deployed-engineering pattern

01

Unify disconnected systems

Orders, WMS and TMS each keyed differently — order id, then AWB. Support tickets weren't keyed at all. Built the crosswalk that joins them.

02

Resolve conflicting definitions

"Loading time" meant something different to the warehouse than to transportation. Documented and encoded the negotiated resolution, not a guess.

03

Answer with evidence, not vibes

The agent cites the exact system and field behind every claim, and says plainly when a cause was never documented — instead of inventing one.

Waybill is a technology demo by CognitionSync. "Auric & Stone" and all orders, customers, tickets and timestamps are synthetic and deterministic — generated once, not fetched from a real business. This page models the class of problem behind most forward-deployed-engineer and enterprise-AI-deployment engagements: making a model work inside someone else's messy, undocumented, real systems.