The quality spreadsheet stops being retyped by hand

The problem

Two parallel worlds and one person copying cells between them.

Two parallel worlds coexist on the floor: the real spreadsheet, with live data on dough moisture, filling weight, and cooking temperature, and the official ERP, with the same data retyped weeks later by someone spending half a day a month copy-pasting and introducing errors.

• The spreadsheet holds the live data: dough moisture after resting, dosed filling weight, dough-to-filling ratio per format, cooking temperature in the tunnel oven and blast chilling time.
• The official ERP holds a retyped version, incomplete and out of date. The practical lag is counted in weeks, and sometimes the data lands once the batch is already on the shelf.
• Somebody spends at least half a day a month copying and pasting between the two, and every copy is a chance to get a decimal wrong on a moisture reading or a dosed weight.
• Transcription errors surface at exactly the worst moment: preparing a client audit or answering a complaint about piece weight out of tolerance. And when the ERP figure and the spreadsheet figure disagree, the argument is no longer about the process, it is about which of the two numbers is the right one.

How it fits the IRIS system

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

Connect listen-inward mode: 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, and pushes it to the ERP via API after validation.

Dough moisture and filling weight data stop living in a folder the ERP never sees. They are in the system the same day they are measured, which is when they are useful for correcting a formula rather than just documenting it. A drifting dough-to-filling ratio stops being discovered three weeks later, with the product already frozen and shipped.

See the full IRIS architecture →

Before and after

Today's lab versus the connected lab

Aspect	Today	With iLEAN Connect
Lag between real data and ERP	Weeks	Second zero
Manual retyping	Half a day a month, at least	Zero
Dough moisture and filling weight	On a sheet the ERP never sees	In the ERP the day they are measured
Copying errors	Surface at the worst moment	Eliminated
The sheet's format	Has to be respected	Taken as it is, even if it changes
Quality module licenses	Several, just for typing	One, for whoever validates

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

Impact estimate

Impact estimate — to be validated against your actual 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.

• 💰 CFO angle: the ERP bills per named user. If several lab people log in today only to enter results, the typical reduction is 50-80%, recurring every year.
• Estimated payback 4-9 months, counting recovered time and license savings.
• Half a day a month of retyping turned into zero, and with it the decimal errors that surfaced at the worst moment.
• And weeks of lag between the figure measured in the lab and the figure in the ERP brought down to second zero.

Typical 50-80% reduction in quality/lab module licenses. Estimated payback 4-9 months. *Estimate to validate*.

And the fair question from the production manager

“Several people have been changing that sheet for years — does it hold up?” — that is the expected case: parsing does not depend on fixed positions, it interprets columns, units, merged cells and margin notes without asking the lab to standardize anything. It is an anchored task, where the best models drop below 1.5% error [1], and the quality lead validates on screen before the push.

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