The metallurgical lab spreadsheet, pushed into the ERP without retyping

The problem

The lab works in Excel because Excel is fast and the ERP is not.

The lab works in Excel because Excel is fast and the ERP is not. But then somebody has to retype, and the gap between the real figure and the ERP figure is measured in days or weeks. Meanwhile the planner does not know which heats are confirmed fit, and when a metallurgical question appears, files have to be opened by hand to reconstruct the history. Sand control suffers the same problem with a different consequence: a drift in moisture, compactability or active clay shows up in the casting defect before it shows up in any report. And there is a cost nobody looks at: every lab person who logs into the ERP purely to enter results consumes a full license.

• Spectrometer chemistry, thermal analysis, nodule count, hardness and tensile live in the lab's spreadsheet; moisture, compactability and active clay in the sand control sheet. Somebody retypes both.
• The gap between the real figure and the ERP figure is measured in days or weeks. Meanwhile the planner does not know which heats are confirmed fit and schedules them anyway.
• A sand drift shows up in the casting defect before it shows up in any report, because the report is typed after the fact and read after that.
• And every lab person who logs into the ERP purely to enter results consumes a full quality module license, year after year, for a task that is pure copying.

How it fits the IRIS system

Connect in listen-inward mode — the folder, not the person.

Connect in "listen inward" mode:

It reads the sheet the lab already has, in the layout it already has, the moment it is saved. Results are structured by heat and by control point, crossed with the heat sheet and the furnace curve, and the quality lead validates on the tablet before the push. Chemistry, energy and outcome end up in the same place for the first time, and the lab keeps working the way it works. The planner sees which heats are confirmed fit in the schedule itself, without phoning the lab.

• iLEAN watches the shared folder and reads the file when it is saved. What it observes is the folder, never people.
• A grounded parser copes with every lab having its own sheet layout and structures results by heat and by control point.
• It cross-references them with the heat sheet and the furnace curve, so chemistry, energy and outcome end up in the same place.
• The quality lead validates on the tablet and, after that signature, Connect pushes to the ERP through its API.

See the full IRIS architecture →

Before and after

Today's lab versus the connected lab

Aspect	Today	With iLEAN Connect
Real figure to ERP figure	Days or weeks	Minutes
Heats confirmed fit	The planner asks the lab	Visible in the schedule
Sand drift	Discovered in the defect	Visible in the report first
History of a heat	Files opened by hand	One query by heat number
Sheet layout	Has to be respected	Read as it is, even as it changes
Quality module licenses	Four	One, the quality lead who validates

from a gap of days or weeks between the real figure and the ERP figure to minutes · from manual reconstruction of history to a query by heat · from a sand drift discovered in the defect to a drift visible in the report · from 4 quality-module licenses to 1.

Impact estimate

Impact estimate — to be validated against the 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.

• Estimated payback 3-7 months — the fastest in this matrix.
• CFO angle: the ERP bills per named user. If four lab people log in purely to enter results, that is four licenses: 50-80% reduction, recurring every year, in the order of thousands depending on vendor pricing.
• Transcription time released in the lab and in quality, on top of the license saving.
• And a sand drift caught in the report instead of in the shot-blasted casting.

estimated payback 3 to 7 months — the fastest in the pool. The saving on quality/lab module licenses is typically 50-80% and it is recurring, in the order of thousands per year depending on vendor pricing, on top of the transcription time released. *Estimate to validate against the actual ERP invoice*.

And the fair question from the production manager

“That spreadsheet has been changed by three lab techs over ten years — will it cope?” — that is the anticipated case: parsing does not depend on fixed cell positions, it interprets columns, units, merged cells and margin notes without asking anyone to standardize anything. It is an anchored task, where the best models drop below 1.5% error [1], every value is checked against its metallurgical range — a carbon equivalent of 9 is rejected, not pushed — and the quality lead validates before the push.

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