Shift start-up stops living on a piece of paper

The problem

The birth certificate of the shift sits in a drawer.

The problem is not that it is filled in by hand. The problem is what happens next: the sheet gets annotated with what actually happened, that station four is running a temporary fixture, that an operator is missing, that the light leather variant runs after the break. And then it sits in a drawer for a full shift, or forever. When the customer raises a claim about a seat three weeks later, reconstructing what happened at that start-up depends on finding that piece of paper. And on whoever wrote it remembering. That is not traceability, that is archaeology.

• The start-up sheet says which variants run this shift and the state the line was left in; the first-off record is what quality signs before releasing production. Both are handwritten, and both get annotated with what actually happened during the first hour.
• Station four on a temporary fixture, one operator short, the light leather variant moved to after the break: that is the information that matters later, and it lives only on that sheet.
• Between the signature and any system, eight to twelve hours pass — or it never gets digitized at all.
• When the customer raises a claim on a seat three weeks later, reconstructing that start-up means finding the paper and trusting the memory of whoever wrote it, if they are still on that shift. That is not traceability, that is archaeology.

How it fits the IRIS system

Connect in photo mode — the line leader adds two photos, nothing else changes.

With iLEAN Connect the line leader takes two photos at start-up, with the same phone already in their pocket. A grounded model recognizes the layout and extracts the fields: line, shift, planned variants, manning, inherited issues, first-off result and signature. All of it lands in central memory, tied to that line and that shift. And I want to stress this, because it is what makes it work: the operator learns nothing new, types nothing, changes not one gesture of their routine. They keep filling in the same sheet. All we add are two photos.

It is the lowest-friction case of the twelve and usually the first one signed off: nobody learns a tool, nobody types. The sheet stays the same sheet. What changes is that from now on a history of shift start-ups exists, and every root-cause analysis on foaming, trim or torque later leans on it. The first question in any customer claim — what was this line running and who was on it — stops being a question.

See the full IRIS architecture →

Before and after

Start-up on paper versus start-up captured

Aspect	Today, on paper	With iLEAN Connect
Latency to a system	8-12 h, or never	Second zero
Planned variants and manning	Handwritten, in a drawer	Structured per line and shift
Inherited issues	Lost unless someone remembers	Extracted and tied to the shift
First-off result and signature	A page in a folder	Attached to the shift record
Answering a customer claim	Find the paper, ask who wrote it	One query by line and shift
Line leader's routine	—	Unchanged: two photos

Impact estimate

Impact estimate — to be validated with 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 the number of lines and shifts.
• Supervision time spent today transcribing and filing start-up sheets, recovered — and the drawer of paper stops growing.
• From 8-12 hours of latency — or none at all — down to second zero, queryable by line and shift.
• And a history of shift start-ups that did not exist before, which is what feeds every root-cause analysis on foaming, trim and torque later.

Estimated payback 4-9 months · from 8-12 h of latency to second zero The direct return is the supervision time spent today transcribing and filing: estimated payback of four to nine months depending on the number of lines and shifts, an estimate to validate with your own figures. But the real value is different: from here on there is a history of shift start-ups that simply did not exist before, and it feeds every root-cause analysis we will talk about later.

And the fair question from the production manager

“What if it misreads the manning or the variant list on a sheet full of crossings-out?” — reading fields from a document with a known layout is an anchored task, where the best models drop below 1.5% error [1]. And there is a second net: the summary goes to the line tablet and the line leader confirms or corrects it before the record enters central memory.

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