From Configurator to Casting: The Custom Order Handoff
Short answer: When a customer finishes configuring a piece, what should reach production is not a screenshot and an email. It should be a structured order containing a bill of materials, a stone specification, sizing, engraving instructions, a manufacturable 3D file and a routing through the correct production stages. Everything between the click and the casting flask should be data moving on its own. In most jewelry businesses, that stretch is still a person retyping.
This is the least glamorous part of a customization program and the part that decides whether it makes money.
The gap nobody photographs
Walk the process in most businesses that sell custom pieces online and it looks like this.
The order lands. Someone opens it, reads the option list, and writes a job sheet. A CAD technician opens the base model, applies the changes described in words, and produces a file. Someone checks it against the original order. It goes to print, to casting, to setting, to finishing. Somewhere along the way the customer emails to ask where it is, and the answer requires two internal messages and a look at a whiteboard.
Every step in that chain is competent work by capable people. It is also almost entirely transcription. The information that a person is carefully moving from one place to another was already complete and unambiguous at the moment the customer clicked confirm.
The cost of that gap shows up in four places: lead time, error rate, the ceiling on how many custom orders you can take in a week, and the quiet fact that nobody can tell a customer where their piece is without asking someone.
What a complete handoff contains
If you build only one thing this year, build this payload. A configured order should arrive in production as a structured record with the following:
Identity. Order reference, customer, channel, promised date, priority.
Bill of materials. Metal type and karat, estimated weight, findings, and the specific components by SKU. Not "14k yellow" as text, but a real material record your costing system already recognizes.
Stone specification. For each stone: shape, dimensions, carat, quality band, setting type, position. Center and accents listed separately, with sourcing status against inventory.
Geometry. The actual 3D model with the customer's choices applied. Not the base model plus a note. This is the field most commonly missing, and it is the one that forces the CAD rebuild.
Sizing. Finger size, band width, profile, and any resizing tolerance. For a stone set band, the stone count derived from that size.
Personalization. Engraving text, font, placement, plus any hallmarking requirements.
Routing. Which stages this piece needs and in what order, with the ones it can skip marked as skipped. A plain band and a pave halo do not travel the same road, and the system should know that without being told each time.
Provenance. A record of exactly what the customer saw and approved, including the render. When a piece is questioned on delivery, this ends the conversation in seconds.
If your current handoff is a PDF, compare it against that list. The distance between the two is your project scope.
Getting the geometry right
The hardest field on the list is geometry, and there are three honest ways to produce it.
Parametric base models. Build each product family once as a parametric model with the configurator's variables driving its dimensions. The customer's selections resolve into a finished model automatically. Highest quality, most manufacturing control, most up front work per family. This is the correct answer for a core catalog you will sell for years.
Component assembly. Model heads, shanks, halos and accents as compatible components and assemble the selected combination programmatically. Faster to scale across many combinations, needs careful tolerance work at every junction.
AI assisted generation. For one off and semi bespoke pieces that fall outside the catalog, AI systems now turn sketches, reference images and written briefs into production ready CAD in a fraction of the time a manual rebuild takes. This is what The Goldsmith AI does, and it changes the economics of the long tail of custom work specifically, where parametric modeling was never going to pay for itself.
Most mature operations end up running all three: parametric for the core, assembly for the variations, AI generation for the bespoke.
Where the data has to land
A handoff is only as useful as the systems it reaches. Four connections, in the order they usually pay off:
Production management. The job exists as a record with stages, owners and dates, from the moment of order rather than the moment of transcription.
Inventory. Stone and finding availability checked at the point of order, not at the point of setting. A customer told at checkout that a stone adds two weeks is a satisfied customer. A customer told the same thing in week three is a refund.
Costing and ERP. Actual metal weight and stone cost captured against the order, so margin on custom work is a number you can see per piece rather than a feeling you have per quarter.
Customer communication. Stage changes push status automatically. This one is cheap to build and disproportionately reduces inbound queries, because most "where is my order" emails are asking a question your factory floor already knows the answer to.
The three checkpoints worth keeping human
Automation in a jewelry workshop should not aim at removing judgment. It should aim at removing typing. Three points deserve a person:
Design approval on bespoke work. When a piece is genuinely one of a kind, a designer should see it before it is printed.
Stone selection above a threshold. Set a value or a carat above which a human chooses the actual stone. The system can handle everything under it.
Final quality inspection. The last look before a piece goes to a customer stays with the person whose eye you trust. Automate everything that gets the piece to that moment, and nothing about the moment itself.
Sequencing a build without stopping production
Nobody can pause a working factory to install a system. A sequence that works alongside live production:
Step 1. Define the payload. Agree the exact fields above with your production team, because they are the people who know what is missing from the current job sheet.
Step 2. Instrument the current process. Measure how long each stage takes today and where orders actually wait. Almost every business is surprised by the answer, and the surprise usually redirects the project.
Step 3. Automate one product family end to end. Bands are the usual first choice, for the reasons in our band configurator guide: clean geometry, unambiguous data.
Step 4. Connect inventory, then costing, then customer status. In that order, because each one makes the next more useful.
Step 5. Extend to the remaining families and let the exceptions be exceptions. The goal is not one hundred percent automated. It is that the routine cases stop consuming the attention that the difficult ones deserve.
We go deeper on choosing what to automate first, across the whole operation rather than just custom orders, in Workflow Automation for Jewelry Manufacturers.
What this is worth
Three things change when the handoff is real, and they compound.
Lead time falls, because the waiting between stages was always longer than the work within them. Error rate falls, because the two places errors enter a custom order are transcription and interpretation, and both are removed. Capacity rises without hiring, because the constraint on custom volume in most workshops is not bench hours, it is coordination hours.
And a fourth thing, less obvious. Once orders are structured data, you can finally see your custom business: which combinations sell, which stages actually cause delay, what a bespoke piece really costs to make. Most jewelry businesses have never had that view, because the information was distributed across job sheets and memory.
How we approach it
AMG Dynamics builds the connection between the storefront and the shop floor: configurators that output production data, ERP and inventory integrations, automated routing, and the CAD pipelines that sit between them. We work with jewelry brands, manufacturers and retailers from London, Delaware and Mumbai, alongside 3D visualization, AI systems and luxury websites.
We take this part seriously because it is where customization stops being a marketing feature and becomes a business.
Want to see where your custom orders actually lose time? Send your current process to admin@amgdynamics.com and we will map it against the payload above.
The production side of this, from structured intake to routing, is on our jewelry workflow automation page.
Frequently asked questions
What should a configurator send to production? A structured order containing order identity, bill of materials, stone specification, the actual 3D geometry with the customer's choices applied, sizing, engraving details, production routing and a record of the render the customer approved. A screenshot and an option list is not enough, because it forces a CAD rebuild and a manual interpretation.
Do we need to rebuild the CAD file for every custom order? Not if the configurator produces geometry. Parametric base models resolve customer selections into finished models automatically, component assembly handles wide combination ranges, and AI assisted CAD generation covers bespoke pieces outside the catalog. Manual rebuilds should be the exception rather than the default.
Can this connect to our existing ERP? In most cases yes. The practical work is defining the data contract between the configurator and the systems downstream, then building the integration. Older or heavily customized ERP systems take longer, but the payload itself does not change.
How much should stay manual? Design approval on bespoke pieces, stone selection above a value threshold, and final quality inspection. Automation should remove transcription, not judgment.
What is the first thing to automate? One product family, end to end, usually the one with the cleanest geometry and highest volume. Prove the payload works on a low risk family before extending it across the catalog.