The good data lives in a spreadsheet the ERP never sees

The problem

The good data lives in a spreadsheet the ERP only sees weeks later.

Two parallel worlds exist. The laboratory spreadsheet, in a shared folder, where the real process data sits. And the ERP, where the official data sits, arriving weeks late because somebody has to copy it across. That transcription costs in the order of half a working day a month, and along the way decimals get lost and batches get swapped. Not because the laboratory works badly, but because between the real figure and the official figure there is a person copying by hand. The discrepancy always surfaces at the worst possible moment: when the auditor cross-checks the quality record against the batch produced and the two sources do not match.

• The laboratory spreadsheet holds the live data: average piece weight, core temperature at the tunnel outfeed, filling percentage, béchamel viscosity, microbiology, the pre-fryer oil control. It is filled in every shift, in a shared folder, by whoever is on duty.
• The ERP holds the official data, which arrives weeks late because somebody copies it across. Between the real figure and the official figure there is a person copying by hand.
• That transcription costs in the order of half a working day a month, and along the way decimals get lost and batches get swapped.
• The discrepancy surfaces at the worst moment: when the auditor cross-checks the quality record against the batch produced and the two sources do not match.

How it fits the IRIS system

Connect in inward listening mode — the folder, not the people.

Connect in "inward listening" mode.

The laboratory keeps working exactly as it does today, in the same sheet, each analyst in their own way. Only the transcription disappears — and with it the licenses that existed purely so that four or five people could type control data into the quality module. The batch-by-batch cross-reference that the auditor does by hand today happens automatically, because every row is written to the ERP already tied to its batch.

• The system observes the shared folder — the folder, not the people — and reads the delta when the file is saved.
• A grounded model interprets the sheet's structure even though every analyst fills it in their own way.
• The result is shown to the manager for sign-off on the validation screen.
• After sign-off it is written into the ERP by API, tied to the corresponding batch. The laboratory keeps working exactly as it does today; only the transcription disappears.

See the full IRIS architecture →

Before and after

Today's laboratory versus the connected laboratory

Aspect	Today	With iLEAN Connect
Gap between the sheet and the ERP	Weeks	Second zero
Manual transcription	Half a working day a month	Zero
Quality record versus batch	They never quite meet	Cross-referenced batch by batch
Lost decimals and swapped batches	Found by the auditor	Eliminated
Sheet format	Has to be respected	Read as it is, even as it changes
Quality module licenses	Four or five, to enter data	One, the person who validates

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.

• Estimated payback 4-9 months.
• 💰 CFO angle: the ERP is billed per active user. If four or five people from the laboratory and quality log in purely to enter control data, that is four or five licenses of the quality module. With Connect only the person who validates goes in: typical reduction of 50-80% in that module's licenses, recurring every year.
• Half a working day a month of laboratory time recovered.
• And quality records that meet the batch on the day the auditor asks, not weeks later.

estimated payback of 4 to 9 months. CFO angle: the ERP is billed per active user; if four or five people from the laboratory and quality log in purely to enter control data, that is four or five licenses of the quality module. With Connect only the person who validates goes in: typical reduction of 50-80% in that module's licenses, with a recurring annual saving. *Estimate to be validated against the real ERP invoice*.

And the fair question from the production manager

“Every analyst fills in that sheet differently — does it hold up?” — that is the anticipated case: parsing does not depend on fixed cell positions. The model interprets columns, units, merged cells and margin notes without asking anyone to standardize the sheet. It is an anchored task, where the best models drop below 1.5% error [1], and the quality manager signs off on the validation screen before anything is written to the ERP.

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