The melt flow index reaches the ERP before the truck reaches the customer

The problem

The batch gets bagged, labelled and sometimes shipped before its result is in the system.

And all of it gets written into a spreadsheet that lives in a shared folder the ERP cannot see. Two parallel worlds: the spreadsheet, where the live data is, and the official ERP, with data transcribed days later. Somebody spends half a day a month copying and pasting, and introduces errors doing it. But the transcription is not the serious part. The serious part is that the batch gets bagged, labelled and far too often shipped before its melt flow result is in the system. As long as that data does not exist in the ERP, automatic batch hold simply cannot exist. There is nothing to hold against.

• The live data lives in a spreadsheet in a shared folder the ERP cannot see.
• Two parallel worlds: the spreadsheet, with today's data, and the official ERP, with data transcribed days later.
• Somebody spends half a day a month copying and pasting, and introduces errors. But the transcription is not the serious part.
• The serious part is that the batch ships before its melt flow result is in the system. While the data travels, so does the truck.

How it fits the IRIS system

Listen-inward mode: it watches the folder, not the person. And it holds the batch.

Connect in listen-inward mode watches the folder, not the person. It reads the file when it is saved, parses the sheet with an anchored model, shows the summary for validation and pushes it to the ERP by API against the right batch. And if the melt flow index falls outside the range of the committed grade, it flags the batch on hold before it ever reaches dispatch. The lab carries on working exactly as before. And here is the part a finance director signs without needing to believe in AI at all. The ERP charges per active user. If the lab and quality team log into the ERP purely to key in results, those are several licences being paid for typing. With Connect only the person who validates logs in: the typical reduction is between fifty and eighty per cent of the quality-module licences, and it is a recurring saving every year.

If the melt flow index falls outside the range of the committed grade, it flags the batch as held before it reaches dispatch. The lab carries on working exactly as before.

See the full IRIS architecture →

Before and after

The lab result, before and after

Aspect	Today	With iLEAN Connect
Lag between measurement and ERP	Days	The moment the file is saved
Who types it in	Somebody, half a day a month	Nobody
An out-of-range batch	Found after shipping	Held before dispatch
Quality module licences	One per person entering data	One, the lead who validates
Evidence for the customer	Spreadsheet and ERP, not matching	One series against the batch
The lab spreadsheet	Everyone tries to kill it and fails	It stays: it is the tool that works

Impact estimate

Impact estimate — to validate against 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 4-9 months, plus the recurring annual licence saving.
• Typical reduction of 50-80% in quality module licences.
• The licence saving is immediate and depends on no improvement hypothesis: it is a cost line that goes down.
• And an out-of-range batch can no longer be shipped, which is the return you actually feel.

Payback 4-9 months · typical 50-80% cut in quality-module ERP licences. Estimated payback is four to nine months, with the recurring annual licence saving on top. It is an estimate to validate, and it validates quickly: the real ERP invoice and the number of active lab users are all it takes. Conservative ranges, an estimate to validate against the plant's real data.

And the fair question from the production manager

"What if the sheet changes format?" — listening mode does not depend on a frozen template: it parses the sheet with an anchored model, respecting its structure as it is [1]. If a new column shows up, it is flagged for review rather than silently accepted.

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