The shop's most valuable quality data stops living in a folder

The problem

The shop's most valuable quality data lives in a folder.

Dimensional reports carry the part submission file and first article inspection, and they are also what decides whether a detail goes back to the machine. While they live in a folder, every lookup is a manual search and every ERP entry is a transcription with risk. There is also a quiet cost that almost nobody looks at: several metrology and quality people hold an ERP seat purely to enter this data. Every one of those licenses is paid for every year. And when the customer asks about the dimensional report of a mold delivered two years ago, the answer depends on someone remembering which folder it was in.

• Every detail and every tryout part leaves a CMM report with dimensions, deviations and a verdict. It carries the part submission file and first article inspection, and it decides whether a detail goes back to the machine.
• While it lives in a folder, every lookup is a manual search and every ERP entry is a transcription with risk.
• Several metrology and quality people hold an ERP seat purely to enter this data, and every one of those licenses is paid for every year.
• When the customer asks for the dimensional report of a mold delivered two years ago, the answer depends on someone remembering which folder it was in.

How it fits the IRIS system

Connect in listening-inward mode — the directory, not the people.

Connect in listening-inward mode: iLEAN observes the directory — not the people — and reads the file as it is saved.

iLEAN observes the folder and reads the report the moment it is saved: dimensions, deviations and verdict are normalized and associated with the mold and detail number. The quality manager validates, the API posts to the ERP. The report stops being a file and becomes project data, retrievable years later by mold. The decision to send a detail back to the machine is taken from the verdict in the system, on the same day it was measured.

• A grounded LLM parses the report: it normalizes dimensions, deviations and verdict.
• It associates them with the corresponding mold and detail number.
• It passes the quality manager's validation gate.
• It posts to the ERP through the API. The report stops being a file and becomes project data, retrievable years later.

See the full IRIS architecture →

Before and after

Today's metrology versus the connected metrology

Aspect	Today	With iLEAN Connect
Measurement to data in the system	Days or weeks	Minutes
Transcription	Half a day per person per month	Zero
Report of a mold delivered two years ago	Hunted through folders	Direct lookup by mold
Detail sent back to the machine	Decided from a printout	Decided from the verdict in the system
Part submission file	Rebuilt by hand	Assembled from posted reports
Quality module licenses	Four or five	One, the manager who validates

from days or weeks between measurement and data in the system → minutes. From half a day per person per month spent transcribing → zero. From hunting for an old dimensional report in folders → a direct lookup by mold.

Impact estimate

Impact estimate — to be validated against your actual ERP invoice.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• 💰 CFO angle: if four or five people log into the ERP purely to type measurement results, only the person who validates needs a seat — a typical 50% to 80% cut in that module's licenses, recurring every year.
• Estimated payback 3-7 months, among the fastest because it stacks two savings.
• Half a day per person per month of transcription recovered.
• And the dimensional report of any delivered mold is a query, not an archaeology exercise.

estimated payback 3-7 months, among the fastest because it stacks two savings. Financial angle: if four or five people log into the ERP purely to type measurement results, only the person who validates needs a seat — a typical 50% to 80% cut in that module's licenses, recurring every year. *Estimate to validate against the actual ERP invoice*.

And the fair question from the production manager

"Our CMM exports change format whenever the program is edited — does it cope?" — that is the anticipated case: parsing does not depend on fixed positions, it interprets characteristics, nominals, deviations and verdicts wherever they sit in the file. It is an anchored task, where the best models drop below 1.5% error [1], and the quality manager validates before the API post.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.