THE QUALITY EXCEL SHEET STOPS BEING RETYPED

The problem

Two parallel worlds, and half a day a month spent copying between them.

Every shift, the quality supervisor fills an Excel sheet with final moisture, salinity, particle size and pH in a shared lab folder the ERP never sees. One person spends half a day a month copy-pasting that data into the official ERP, with weeks of lag and error risk on a critical figure.

• Every shift the quality supervisor fills a spreadsheet with final moisture, salinity, particle size and pH, in a shared lab folder the ERP has never seen. The file works perfectly well; it just lives nowhere useful.
• One person spends half a day a month copy-pasting that into the official system, which is why the ERP is permanently weeks behind reality on the figures quality cares about most.
• Residual moisture is not a soft figure: it is what supports the declared shelf life of the cube and the sachet, and it is being moved between systems by hand.
• Every copy-paste is a chance to shift a decimal on the one number a customer might one day ask you to prove, and nobody checks the copy against the original.

How it fits the IRIS system

Connect listens inward — it watches the file, never the person.

Connect 'listens inward': it watches the shared folder (not the person), reads the delta when the file is saved, parses it with a grounded LLM, validates it on screen with the supervisor, and pushes it to the ERP via API after validation.

The spreadsheet is not standardized, replaced or taken away from the lab. It is read where it already lives, in the format the lab already trusts, and it keeps working the day somebody adds a column in March without telling anybody. That tolerance is what makes the case deployable in weeks rather than quarters.

See the full IRIS architecture →

Before and after

The lab as an island versus the lab connected

Aspect	Today	With iLEAN Connect
Gap between measuring and the ERP	Weeks	The same shift
Retyping quality results	Half a day a month	None
Residual moisture per batch	In a folder the ERP never sees	In the ERP the day it is measured
A decimal shifted while copying	Surfaces when a customer asks	Cannot happen
The spreadsheet's format	Would have to be standardized	Taken exactly as it is
Quality and lab ERP licenses	Several, purely to type	One, for whoever validates

Weeks of lag between real data and ERP data → zero seconds. Half a day a month of transcription → zero.

Impact estimate

Impact estimate — to be validated against your ERP license bill.

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: the ERP bills per named user, every year, whether that user does one thing or twenty. If several lab people log in only to enter results, the typical reduction is 50 to 80% in the quality and lab module, and it recurs.
• Estimated payback 4-9 months, before the license line is counted at all.
• Half a day a month of transcription returns to the quality team, and the ERP stops lagging reality by weeks on the figures that matter most.
• And the shelf-life figure you would defend in front of a customer stops travelling between systems by copy-paste, which is the one journey nobody ever audits.

CFO angle: if several lab staff log into the ERP today just to enter data, with Connect only the validator needs a license — typically 50 to 80% fewer licenses in the quality/lab module. Estimated payback of 4 to 9 months. Estimate to be validated.

And the fair question from the production manager

"What if somebody changes the sheet and it stops reading it?" — parsing a spreadsheet whose meaning is stable is an anchored task, under 1.5% error [1] for the best models, and a new column does not break it the way a rigid template would. The second net is the validation screen: the supervisor sees the parsed values next to the figures they measured before anything reaches the ERP, so a bad read is a correction on the spot and never a wrong record in the system of truth.

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