Line startup stops living on a sheet of paper

The problem

The startup paperwork is the only source that knows why the line did what it did.

With several packaging lines running three shifts, a plant generates dozens of startup documents and shift reports a day. A clerk retypes them twenty-four to seventy-two hours later, keeping only what fits the boxes on the OEE report. The result is that the system records forty minutes of downtime but not why: the cause, handwritten in the margin by the operator who was standing at the machine, never reaches continuous improvement. The whole improvement cycle feeds on generic categories filled in after the fact instead of observed causes.

• Packaging order — which brand, which format and which batch are due on this line and this shift. It comes down from planning on paper, sometimes with a change scribbled in pen.
• Startup checklist — signed by hand by the line lead. It is the proof that what had to be checked was checked before releasing the first container.
• Shift report — the real downtime, with the cause written in the margin by whoever was standing at the machine. It is the most valuable of the three and the first one lost.

How it fits the IRIS system

Connect in photo-the-analog mode — the line lead's phone digitizes the paper without touching the machine.

Connect in photo-the-analog mode. Step by step: (1) the line lead photographs the packaging order and the checklist with a phone or industrial tablet at startup; (2) an anchored LLM recognizes the document structure and extracts brand, format, batch, line, shift, start time, checklist signature and the handwritten causes and durations; (3) the result goes to the early human verification tablet at the line, where the supervisor confirms with two taps; (4) only then does it cross into central memory, anchored to the packaging batch. Zero habit change and zero training: the operator already knows how to take a photo.

Two photos at startup. Connect extracts brand, format, batch, shift, signature and the handwritten causes. The supervisor confirms with two taps and the batch exists in the central system. The line no longer depends on somebody retyping it later.

• Photo of the packaging order — the line lead takes it with the phone or industrial tablet at startup. Connect extracts brand, format, batch, line, shift and start time.
• Photo of the signed checklist — the evidence of who checked what and when stays anchored to the packaging batch, not to a binder.
• Handwritten text is read — an anchored LLM recognizes the document structure and recovers the causes and durations written in the margin, which is exactly what never reaches analysis today.
• Human verification before entry — the result appears on the tablet at the line; the supervisor confirms with two taps or corrects. Only then does it cross into central memory.
• Zero habit change — the operator already knows how to take a photo. No app to learn, no new screen to install, no PLC to touch.

See the full IRIS architecture →

Before and after

Startup retyped after the fact vs. startup captured at second zero

Aspect	Startup with clerical retyping	With iLEAN Connect
Data latency	24-72 h until somebody types it	Second zero
What gets captured	Only what fits the report fields	The whole document, handwriting included
Micro-stop cause	Lost in the margin of the report	Anchored to the batch and available for analysis
OEE analysis	Generic categories filled in later	Causes observed by whoever was there
Checklist evidence	Paper binder, hard to retrieve	Signed photo anchored to the packaging batch
Clerical workload	Several person-days a month per plant	Two confirmation taps per startup

Latency of 24-72 h to second zero. Selective retyping by form field to the full handwritten text anchored to the batch. Micro-stop causes lost to real causes available for OEE analysis.

Impact estimate

Impact estimate for your plant — 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.

• Brewery with several packaging lines on three shifts: dozens of startup documents and shift reports a day.
• Connect pilot on a single line, with the phone or tablet the line lead already carries. No construction work and no downtime.
• Indicative payback between 4 and 9 months, depending on lines and shifts.
• The hard lever is not the clerical saving: it is that the improvement cycle starts feeding on real causes instead of categories filled in after the fact.

Estimated payback 4-9 months depending on the number of lines and shifts, through recovered clerical retyping time and, above all, through the quality of OEE analysis. *Estimate to validate*.

And the fair question from the production manager

"What if the OCR misreads a handwritten cause and pollutes the analysis?" — hallucination is a problem of <b>free</b> generation, not of <b>anchored</b> tasks. Here the AI only moves a specific data point from one medium to another, and in that kind of task the best models dropped below <b>1.5%</b> error ^{<a href="#source-openai">[1]</a>}. And even so, nothing enters the system without the supervisor seeing it: early human verification sits before the master record, not after it.

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