Paperless Build Sheet at Unit Start

The problem

The build sheet knows what the unit is, but only on paper.

The paper build sheet is the physical source of truth for every unit, but any last-minute note (option change, prior issue, delivery priority) only lives on paper until someone transcribes it, sometimes several stations later. That lag causes rework on mis-built options and forces a manual reconstruction of what happened to a specific unit whenever an issue surfaces.

• In a mixed-model plant every body shell starts with a printed build sheet: model, body color, trim level, destination market and option content. It is the physical reference every station builds from.
• Last-minute changes are written on it by hand — an option swapped, a prior issue on that unit, a delivery priority. That note exists on one sheet of paper riding on one skid.
• Someone transcribes it several stations later, if at all. Meanwhile the downstream stations build from the MES content, which does not know about the change yet.
• When a unit comes off the line with the wrong option, rebuilding what happened to that VIN means walking the line, asking people and hunting for the sheet.

How it fits the IRIS system

Connect in photo mode — the build sheet keeps riding with the body.

Connect photo mode: the operator photographs the build sheet with a phone or tablet; a grounded LLM extracts model, color, trim, market and notes, tying them to the unit number in central memory. Early human verification confirms the read before it crosses into the MES.

Nothing changes at line start: the sheet is still printed and still travels with the unit. One photo is added, and from that second the handwritten note stops being a secret between one piece of paper and whoever happens to read it. It is the case with the least friction on the floor in this matrix, which is why it usually opens the door.

See the full IRIS architecture →

Before and after

The build sheet on paper versus the build sheet captured at line start

Aspect	Today, on paper	With iLEAN Connect
Last-minute option change	Visible only to whoever reads that sheet	Pushed to every downstream station
When the VIN's content is complete	Stations later, after transcription	Second zero, at line start
Mis-built option rework	Found at end-of-line inspection	Avoided before the station builds it
Reconstructing a unit's history	Walk the line and hunt for the sheet	One query by VIN
Delivery priority noted by hand	Lost if nobody types it	Structured and tied to the unit
Line-start operator's routine	—	Unchanged: one photo per body

A last-minute note that depends on someone physically seeing it → available to every downstream station from zero second.

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 units started per day.
• The return comes from mis-built options that stop needing rework, and from transcription time recovered at line start and in the office.
• From hours of lag between the handwritten note and the MES to a record that exists at second zero.
• And when an issue surfaces on a specific VIN, its history is queried in seconds instead of rebuilt from memory and paper.

Estimated payback 4-9 months depending on daily unit volume. *Figure to be validated*. (Payback 4-9 months · from hours of lag to zero-second record)

And the fair question from the production manager

“What if it misreads a trim code or the handwritten market on the sheet?” — reading fields from a build sheet whose layout never changes is an anchored task, where the best models drop below 1.5% error [1]. And there is a second net before the MES: the line-start operator sees the read on the station tablet and confirms it, and any option code that does not exist for that model is flagged instead of accepted.

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