Lab data, from spreadsheet to ERP with no transcription

The problem

The lab measures every batch; the ERP finds out days later.

The ERP requires someone to manually transcribe lab data -half a day a month lost, plus a gap between real lab data and what purchasing or sales can see.

• The mill lab records per batch the values that decide whether the flour meets the specification: moisture, wet and dry gluten, protein, falling number and the alveograph's W and P/L. Together they define which customer and which application each batch can go to.
• They live in a spreadsheet, because that is where the technicians work fast. The ERP requires someone to retype them by hand.
• That costs about half a day a month of a qualified technician's time, and leaves a gap of days between what the lab knows and what purchasing or sales can see.
• Meanwhile the mill pays for several quality module licenses just so people can type data that already exists in a file.

How it fits the IRIS system

Connect "listens inward" — the file is read when it is saved.

Connect "listens inward" reads the file when it's saved (without watching anyone), parses it with a grounded LLM and stores it in central memory. After human validation, it flows to the ERP via API.

The lab keeps its spreadsheet and its way of working. Connect reads the file when it is saved, without monitoring anybody's screen, and the data reaches the ERP validated without a second pair of hands typing it. Sales can answer a customer's question about a batch's gluten without calling the lab.

See the full IRIS architecture →

Before and after

Lab data today versus lab data connected

Aspect	Today	With iLEAN Connect
Alveograph W and P/L in the ERP	Retyped by hand	Pushed via API after validation
Falling number per batch	Days of lag	Zero lag
Transcription time at the flour mill	About half a day a month	None
Quality module licenses	4 users typing	1 user reviewing
Sales asking for a batch's gluten	Ask the lab	Visible in the ERP
The lab's spreadsheet at the flour mill	—	Kept as it is

From half a day a month of transcription and days of lag, to data available at zero lag for the whole organization.

Impact estimate

Impact estimate — to be validated with your numbers.

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 3-6 months.
• Estimated savings of 50 to 80% on quality/lab module licenses, from 4 licensed users down to 1 — to be checked against your ERP vendor's pricing.
• About half a day a month of transcription recovered for the lab team. Time that goes back to analysis, which is what the lab is for.
• And lab results available to purchasing and sales at zero lag instead of days later. A customer asking about the falling number of last week's batch gets an answer in the same call.

Estimated savings of 50 to 80% on quality/lab module licenses (from 4 licensed users down to 1), payback 3-6 months. Estimate to validate against your ERP vendor's pricing.

And the fair question from the production manager

"What if it reads a falling number into the wrong batch?" — parsing a spreadsheet whose columns and batch codes are known is an anchored task, where the best models drop below 1.5% error [1]. On top of that, each value is checked against its physical range and its batch code before it is shown, and a person validates before anything flows to the ERP. A wrong batch code simply does not match any open batch and is sent back for review.

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