The test spreadsheet stops living in a world of its own

The problem

The real test results and the official ones live in two different places.

The quality lead fills in a spreadsheet every batch with rating tests — drop, stacking, leak-tightness — in a shared folder the ERP never sees. It's two parallel worlds: the real spreadsheet with live data, and the official ERP with data transcribed late by someone who spends time every month copy-pasting, and sometimes makes mistakes.

• Every batch, the quality lead fills in a spreadsheet with the rating tests: drop, stacking and leak-tightness, plus notes on board moisture or adhesive. Each row is the proof that a batch earned its rating.
• That spreadsheet sits in a shared folder the ERP never sees. It is the most up-to-date quality record in the plant, and the least connected.
• The official ERP holds data transcribed late by someone who spends time every month copy-pasting — and sometimes makes mistakes. A drop height typed wrong or a stacking load in the wrong column looks valid until someone compares it with the original.
• The gap between the real test and the ERP record runs to days, which is exactly when a customer asks whether a batch passed. Then someone opens the spreadsheet anyway, because the ERP cannot be trusted to be current.

How it fits the IRIS system

Connect listening inward — it reads the folder when the file is saved.

iLEAN Connect's "listen inward" mode reads the shared folder when the file is saved — it doesn't watch the person. It parses every cell with an anchored LLM, and after the lead's validation, it automatically syncs to the ERP via API.

The quality lead keeps the spreadsheet she already trusts. What changes is that the result exists in the ERP as soon as she validates it, so the answer to "did this batch pass the stacking test?" stops waiting for the end of the month. And nobody spends an afternoon copying cells into a screen that already had the answer elsewhere.

See the full IRIS architecture →

Before and after

The rating-test spreadsheet today versus the spreadsheet synced

Aspect	Today	With iLEAN Connect
Drop and stacking results in the ERP	Days late	On validation
Monthly copy-paste work	Recurring	Zero
Typing mistakes in test values	Found when a customer asks	Not introduced
Layout of the test spreadsheet	Must be respected	Read as it is, even if it changes
Lab licenses in the ERP	Several named users	A single validation license
What the quality lead does	Fills in, then transcribes	Fills in, then validates

Before: a days-long gap between the real test and the ERP record, plus monthly transcription time. After: zero latency, zero transcription.

Impact estimate

Impact estimate — to be validated against your license 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: typical 50-80% reduction in quality and lab module licenses, from several user licenses down to a single validation one. The saving recurs every year, which is why finance usually looks at this case first.
• Estimated payback 4-9 months.
• From a days-long gap between the real test and the ERP record to zero latency, and zero transcription.
• And the test history per batch is in the ERP when a chemical or pharma customer asks for it. That history is also the raw material of the certification dossier. Nothing has to be copied twice.

CFO angle: typical 50-80% reduction in quality/lab module licenses (from several user licenses down to a single validation one). Estimated payback 4-9 months. Estimate to be validated.

And the fair question from the production manager

“That spreadsheet has merged cells and notes everywhere — can it really parse it?” — yes. Reading a spreadsheet whose columns and units are known, even when the layout moves, is an anchored task, where the best models drop below 1.5% error [1]. And nothing is synced until the quality lead validates what was read. A value outside the expected range is flagged, not pushed.

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