The pellet batch is born traced, not on a sheet that gets lost

The problem

Paper is not the problem. The hours before it exists for the system are.

And today that sheet sits on the office desk until somebody collects it at the end of the day. Between start-up and the moment the data exists for the system, four to eight hours go by. When a customer complains three weeks later and someone has to answer which material that batch came from, the answer is on a piece of paper that has to be found, and that was sometimes filled in from memory.

• The blend sheet sits on the office desk until somebody collects it at the end of the day.
• Between start-up and the moment the data exists for the system, four to eight hours go by.
• When a customer complains three weeks later and you have to say what material that batch came from, the answer is on a sheet that has to be hunted down.
• And that was sometimes filled in from memory — which is worse than not having it: it gives a certainty that is not there.

How it fits the IRIS system

Two photos at start-up — the operator fills in the same sheet as always.

With Connect, the operator takes two photos at start-up: the order and the blend sheet. Nothing in the routine changes, the sheet still gets filled in exactly as before. iLEAN recognises the structure of the document and extracts the fields: batch number, polymer, blend suppliers and their proportion, start-up temperature profile, screw speed and remarks. The result is that the batch exists in central memory at second zero, and it exists with the real blend rather than the planned one. That is what later makes it possible to correlate what a batch was made of with how it behaved: the melt flow index it returned, the colour, the fines load. Without this first step, no downstream root-cause analysis has anything to work with.

The batch exists in the central memory at second zero, and it exists with the real blend, not the planned one. That difference is what everything downstream rests on.

See the full IRIS architecture →

Before and after

The blend sheet on paper vs. the sheet captured

Aspect	On paper today	With iLEAN Connect
When the batch exists in the system	4-8 h later, or at shift close	Second zero
Which blend is recorded	The planned one	The real one, supplier by supplier
Start-up profile and screw speed	On the sheet, cross-referenced with nothing	Tied to the batch
Answering a claim from three weeks back	Hunt for the sheet	A query by batch
A sheet filled in from memory	False certainty	The photo remains as evidence
Operator routine	—	Unchanged: still writing by hand

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, depending on batches per day.
• From 4-8 h of latency to traceability at second zero.
• Recovers the support team's transcription time, which today is half a day gone every week.
• And the real blend on record is the basis of root-cause analysis: without it, everything else measures on sand.

Payback 4-9 months · from 4-8 h of latency to traceability at second zero. Estimated payback is four to nine months depending on the number of batches per day. It gives the production manager back the transcription time, but the real value is elsewhere: inbound traceability stops depending on a sheet of paper not getting lost. This is an estimate to validate against the plant's actual volumes. Conservative ranges, an estimate to validate against the plant's real data.

And the fair question from the production manager

"What if it misreads one supplier's proportion?" — reading fields off a sheet with a known structure is an anchored task, where the best models drop below 1.5% error [1]. And nothing enters on its own: the summary goes to the tablet and somebody confirms or corrects it before it crosses into the system.

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