Batch startup without paper

The problem

The two documents that define a structural batch live in a folder.

The work order is printed from ERP data and any tooling adjustment is noted by hand. The first-piece dimensional check sheet is handwritten per quality procedure. Both can sit undigitized for 4-8 hours, and when an IATF 16949 auditor asks to reconstruct a batch, it's sheet by sheet.

• The work order prints from the ERP with part number, quantity, operation and die, and then it gets annotated by hand: a tooling shim, a blank holder adjustment, a coil change mid-batch.
• The first-piece dimensional check is handwritten because the control plan says so, with nominal, tolerance, measured value and the inspector's pass or fail.
• Between the signature and digitization, four to eight hours pass — and considerably more across a weekend shift.
• During that gap the batch physically exists and is being welded downstream, but exists in no queryable system, which is exactly the state you do not want when an IATF 16949 auditor picks a batch at random and asks you to walk them through it. Reconstructing it means going to the folder, finding the right sheet and hoping the annotations are legible.

How it fits the IRIS system

Connect in photo mode — the operator keeps filling in their paperwork.

Connect photo mode: the operator photographs the work order and the dimension sheet with their phone or industrial tablet. A grounded LLM extracts part number, batch, measured dimensions and pass/fail result, saving it to central memory with no change of habit.

It is the lowest-friction case of the twelve, because it asks the line for nothing new: the operator adds a photo to what they already did with pen and paper. No screen to learn, no field to fill in, no step added to a startup that already sits on the shift's critical path. That is usually why it is the first one approved by a plant that has never deployed AI on the floor.

See the full IRIS architecture →

Before and after

Startup on paper versus startup captured

Aspect	Today, on paper	With iLEAN Connect
Capture latency	4-8 h, more across a weekend	Second zero
Reconstructing a batch for the auditor	Sheet by sheet in a folder	One query by work order number
First-piece dimensions	Handwritten, unqueryable	History by part number and die
Tooling adjustment noted by hand	Lost unless somebody types it	Extracted and tied to the batch
Coil change mid-batch	A note in the margin	A recorded event on the batch
Operator's habit	—	Unchanged: two photos

From 4-8 hours of latency to zero. From manual batch reconstruction for audits to instant web lookup by work order number.

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 how many batch startups you run per day.
• Transcription time recovered by production and by quality.
• From 4-8 hours of latency — more across a weekend — down to second zero.
• And reconstructing a batch for an audit stops being a folder search and becomes a lookup by work order number.

Estimated payback 4-9 months depending on daily startup volume. Estimate to validate.

And the fair question from the production manager

“What if it misreads a measured dimension on the first-piece check?” — reading fields from a document with a known structure 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 inspector confirms or corrects it before anything crosses into the system.

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