The lab's Excel is not the problem — the problem is that nobody in the ERP sees it.

The problem

Two parallel worlds in the same plant — the real one (Excel) and the official one (ERP).

In plants that make food products there is a pattern that repeats with surprising consistency: quality control runs on a lab Excel, and the ERP runs on a transcription someone does at the end of the month.

• The lab's Excel is the REAL operating tool. The operator knows it, opens it fast, needs no training for it, customizes it with their own columns (moisture, weight, color, hardness, the acceptance range for each parameter), and saves the data in the shift's shared folder. It works.
• The ERP is the official system of record, but nobody opens it to enter quality data. It takes ten minutes per batch between starting the session, navigating to the module, filling in the fields, validating and closing. And it demands an extra license for every person who logs in. The lab does not use it.
• Result: two parallel worlds. The real one (Excel, where the shift's live data lives) and the official one (ERP, where the data is transcribed at month's end by a person who spends half a day copying and pasting). They coexist, but they are never the same data at the same time.

The hidden cost of this duality has four layers that pile up: (1) the person who transcribes — half a day of administrative work every month; (2) a lag of up to 30 days between the lab's real data and what the ERP sees, exactly when management asks for the dashboard; (3) the transcription errors that slip through — a misplaced decimal in moisture can change a batch's verdict; (4) when the IFS or BRC auditor arrives, there is no way to cross-check the Excel against the ERP because the cells do not match — the data is in both places but with different naming and misaligned dates, so it has to be rebuilt by hand.

How it fits the IRIS system

iLEAN Connect does not replace the Excel — it connects it to the ERP without anyone noticing.

The problem is not the Excel. The problem is that the Excel and the ERP never talk, and forcing the lab to operate inside the ERP is expensive (in time and in licenses) without solving anything. iLEAN Connect, in "inward listening" mode, seals that crack: the Excel stays where it is, and the data flows to the ERP on its own.

Connect watches the lab's shared folder. When the quality manager saves the Excel, Connect reads the new cells, identifies them, deposits them in the central memory, and after one tap of human validation inserts them into the ERP automatically. The operator changes nothing. The ERP stops needing 5 lab licenses.

The concrete mechanics of Connect applied to the quality-control Excel at a food products plant:

• "Inward listening" mode — Connect watches the lab's shared folder where the shift's Excel lives. It does not watch the operator; it watches the file. When it detects a new save, it activates.
• Delta-only reading — Connect reads only the cells that have changed since the last save. It does not reprocess the whole Excel or the old data: only the new data for the batch in progress. This is what lets the system work in seconds, without saturating the network or the ERP.
• Parsing with an anchored LLM (no free generation) — the model identifies the column-and-row structure of the lab's Excel (sourdough moisture, average piece weight, bake color, hardness, batch, shift), without asking the operator to change a single column of their template. If the plant changes its Excel format, Connect re-maps. The operator never notices.
• iLEAN central memory instantly — the data crosses into the central memory the moment it is saved, not at month's end. This is what makes live dashboards possible, plus trend alerts (moisture climbing three shifts in a row) and real-time batch dossiers for the auditor.
• Human validation before the ERP — the data sits in a pending state until the quality manager validates it with one tap (on the earpiece, on the tablet, in the mobile app). Once validated, Connect inserts it into the ERP via API or RPC, depending on what the vendor supports.
• CRITICAL — savings on ERP licenses. Today the ERP demands a license per active user. If 5 lab people access the ERP just to enter quality data, that is 5 full licenses. With Connect, those 5 people stay on their Excel and the ERP receives the data automatically without anyone from the lab entering the system. Typical result: from 5 licenses down to 1 — the validating manager's, and even then that manager often validates from their phone without opening the ERP. The vendor's invoice changes.

See the full IRIS architecture →

Before and after

Excel + manual transcription to the ERP vs. Excel + automatic Connect

Aspect	Today — Excel + monthly transcription	With iLEAN Connect "inward listening"
Data worlds	Two parallel ones (real Excel / official ERP)	One — the operator's Excel, the ERP receives what is validated
Time spent transcribing	Half a day of work every month	Zero — Connect reads and pushes on its own
Lag between real data and ERP	Up to 30 days	Seconds — after human validation
Lab ERP licenses	5 (one per person entering data)	1 (only the validating manager)
Transcription errors	Inevitable — the miscopied decimal	Zero — the data is the same as in the original Excel
IFS/BRC auditor asks for Excel ↔ ERP cross-check	Manual reconstruction, days of work	Automatic per-batch dossier, same data on both sides
Changes to the lab's Excel template	Retraining, new procedure	Connect re-maps — the operator never notices

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Mid-sized food products plant, 2 shifts, quality control on a per-batch Excel in a shared folder, 4-6 lab people with ERP access exclusively to enter quality data.
• Connect pilot on the lab folder (no machine touched, no Excel touched, no ERP touched in the initial phase — just delta reading and deposit into central memory). First visible value within a few weeks.
• Estimated payback between 4 and 9 months, depending on how many Excels iLEAN absorbs (one per shift, one per line, one per SKU) and the volume of manual transcription avoided.
• Direct savings on ERP licenses: a typical 50-80% reduction in quality/lab module licenses, depending on how many users log into the ERP today just to transcribe. This lever alone can pay for the pilot in the first year.
• Recovery of half a day per month for every person who today spends time copying from Excel to the ERP — time that goes back to the lab, not cut from the org chart.
• IFS/BRC compliance reinforced: automatic Excel ↔ ERP cross-check per batch, dossier rebuildable with no manual work when the auditor arrives.
• The combined hard lever (licenses + time + zero transcription errors + audit) usually convinces the CFO before the quality manager — and that speeds up the decision.

And the quality manager's reasonable doubt

“What if the AI misreads a cell in the Excel and puts a wrong value into the ERP?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely recontextualizes a piece of data from one system to another (reading an Excel cell and mapping it to the equivalent ERP field), the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN deposits the data as pending and the person signs before it crosses into the ERP. The three safety rings exist precisely for this.

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