Quoting wire frames: from half a day per frame to about 15 minutes

Client context

A European automotive supplier with plants in Europe and North America. The company produces bent-wire parts and welded wire frames for car seats - high-volume work, hundreds of thousands of pieces a year, for multinational automotive customers.

The real problem

In automotive, the offer decides everything. A request for quotation arrives one to three years before production starts, the project then runs five to eight years, and once the customer nominates a supplier, the price is essentially locked for the life of the program. Quote too high and you lose the project. Quote too low and you live with it for years.

Each request arrives as CAD data in a different format, often without a bill of materials. A welded frame holds up to forty bent-wire parts. For every one of them, an estimator had to open the model in a viewer, follow the wire line by line, count the bends by hand, note the lengths on paper and type it all into the calculation spreadsheet. Then the customer revises the design - some requests go through twenty versions - and the analysis starts again. The customer expects the updated offer in days.

Where we started

We did not start with software. We spent the first sessions walking through how an offer is actually built: what arrives from the customer, what the estimator extracts from it, how the calculation spreadsheet works, where the ERP takes over, and which steps eat the days. One rule was set by the client and we built everything around it: the customer's drawings are the customer's intellectual property. Nothing goes to a public cloud, nothing is stored outside the company's control.

What we built

A wire-analysis tool fitted into the process the team already runs:

• Upload the customer's 3D file, get the breakdown. The tool sorts the assembly into wires and other parts, and for each wire reads the diameter, length and number of bends, with a 3D view of the frame beside the table.

• Plain geometry, no guessing. The analysis is algorithmic computation on the CAD geometry. There is no AI producing the numbers, so a result can always be traced and checked. AI helped us build the tool quickly; it does not run inside it.

• The expert stays in charge. Bends are counted by two independent methods. Where they agree, the row is green. Where they disagree, the row is flagged and the estimator checks that one part and corrects it. The Excel export records which values were detected automatically and which were set by hand.

• It feeds the existing tools instead of replacing them. The result lands in the same calculation spreadsheet the sales team has always used. The ERP remains the system of record. The tool stands on an open-source CAD core, so there are no new licence fees and the client owns the code.

• A packaging module is next. Freight is 25 to 30 percent of a frame's price, and a truck full of stacked wire frames is mostly air. The same geometry engine now simulates how frames nest into boxes and boxes into a trailer. The first runs packed over a hundred frames more into the same truck than the current standard packing - a simulation result the logistics team is now validating against reality.

Delivery ran in weekly cycles: feedback on Thursday, updated version on the server within days. Version 1 was accepted into daily production use about six weeks after kickoff, delivered in a fixed scope with follow-up fixes included.

The results

• Half a day to a full day → about 15 minutes to break down a new frame, in the estimator's own assessment after weeks of real use - including the manual check.

• The tool found errors the manual process had missed. On one of the first test frames it counted 16 bends where the historical spreadsheet said 15. Later, the packaging analysis surfaced that two standard boxes on record do not actually fit side by side in the truck on record - a data conflict nobody had noticed.

• +100 frames per truck in the first packing simulation against the current standard - pending physical validation.

• Twenty design revisions per request stop being a threat. Re-running the analysis on an updated model costs minutes, so the team answers revision waves at the pace the customer demands.

What we learned

The first version classified about 70 percent of one complex frame correctly, and a simple four-bend wire once came back as twelve. We could have chased full automation for months. Instead we accepted, together with the client, that the target is different: the tool does the boring majority, flags its own doubts, and the expert settles them. The shop's craft also lives outside the CAD file - a large arc is drawn as one radius but bent on the machine as five separate bends. Rules like that come from the estimator's head, and the tool absorbs them release by release. That is why it works: it was built with the people who quote frames every day, not around them.

Thinking about the manual steps in your own quoting?

Book a free 30-min process review call. We map the process first and automate the highest-volume, lowest-judgment steps - fixed scope, step by step, your data on your side and your code at the end.